Many parents have heard the warning: Don’t use baby talk with babies and toddlers. Instead, caregivers are often encouraged to speak properly and use adultlike language, out of concern that simplified speech could confuse children or delay language development.

But my research, which I highlighted in in my new book, “Beyond Words,” suggests the opposite is true. The sing-song voice many adults instinctively use with infants, sometimes called “baby talk” but more accurately known as “parentese” or infant-directed speech, actually helps children learn language.

Far from confusing babies, exaggerating phrases like “Loooook at the doggie!” capture their attention, help them detect patterns in speech and strengthen social bonding.

And the funny mistakes children make along the way, such as saying “goed,” instead of “went,” or “mouses” instead of “mice,” are not signs that children are learning language incorrectly. They are evidence that children are actively working out the rules of language for themselves.

What parentese really is

When many people think of baby talk, they imagine nonsense phrases like “goo goo ga ga” or made-up words like “num nums.” But that’s not what linguists and developmental psychologists mean by parentese.

Parentese uses real words and grammatically correct sentences, but with exaggerated intonation, a higher pitch, stretched-out vowels and a slower rhythm. Think of the way a caregiver might naturally say: “Hi, baaaaby! Are you huuungry?”

There is little evidence that occasional playful nonsense words harm children’s language development. But studies suggest that parentese in particular helps babies pay attention to speech, recognize patterns and engage socially.

Adults across cultures tend to speak this way to infants instinctively. Even people who swear they never use baby talk often slip into it around babies.

Researchers have found that infants actually prefer listening to parentese over regular adult speech. The exaggerated sounds and slower pacing make language easier to process. Babies are better able to pick out individual sounds, notice word boundaries and recognize patterns. In other words, parentese helps tune babies into language.

It also strengthens emotional connection. Language learning does not happen in isolation. Babies learn through warm, responsive interaction with caregivers during feeding, play, bath time and everyday routines.

Interestingly, humans are not the only ones who respond to this style of communication. Studies have even shown that cats react more positively when people use a baby-talk voice with them.

Babies are not passive learners

Children do not learn language simply by copying adults word for word. They actively test hypotheses about how language works. That is why toddlers make predictable and surprisingly logical mistakes.

One common example is overgeneralization. A child learns that people form the past tense of many verbs by adding “-ed,” so they produce forms like “goed,” “eated” or “comed.”

These are not random errors. In fact, they show that the child has understood a grammatical rule and is trying to apply it consistently. The problem is simply that English is full of irregular exceptions. The same thing happens with plurals. Children may say “foots” instead of “feet” or “mouses” instead of “mice.” Again, the logic behind these errors is sound.

Linguists sometimes say that children are little scientists, constantly testing patterns and revising their understanding as they receive more input from the world around them.

Why toddlers call everything a ‘dog’

Young children also make predictable mistakes with meaning.

A toddler might learn the word “dog” and then use it for every four-legged animal they encounter. Linguists call this overextension. On the flip side, some children use words too narrowly. A child may use “dog” only for the family pet and not recognize that other dogs belong in the same category. Linguists call this tendency underextension.

These mistakes reveal how children organize and categorize the world around them. They are gradually mapping words onto objects, people and experiences.

Pronouns are another tricky area. Small children often confuse “me” and “you” because these words constantly shift depending on who is speaking. If a parent says, “I’ll pick you up,” the child hears themselves called “you.” But when they try to repeat the sentence, they may not yet understand that the labels switch from speaker to speaker.

This is why toddlers sometimes say things that sound unintentionally cute or confusing. But beneath the confusion is a sophisticated learning process.

Even the Cookie Monster gets it wrong

Children’s speech errors are so recognizable that they often appear in popular culture. Sesame Street’s character Cookie Monster famously says things like “Me want cookie,” while Elmo often refers to himself in the third person: “Elmo wants this.” These speech patterns mirror real stages of child language development. Young children commonly confuse pronouns or refer to themselves by name before mastering forms like “I,” “me” and “mine.”

Despite occasional complaints from adults, there is no evidence that hearing this kind of speech harms children’s language development. If anything, it reflects the natural experimentation children go through.

‘Pasketti’ and ‘wabbit’

Pronunciation develops gradually too. Young children often simplify difficult sounds and groups of consonants. “Spaghetti” becomes “pasketti,” “rabbit” becomes “wabbit” and “yellow” may come out as “lellow.”

Speech-language specialists call these simplifications phonological processes. They are a normal part of development because some sounds are physically harder to produce than others. Sounds such as r, th, sh and ch tend to develop later because they require more precise control of the tongue and mouth.

Most children naturally outgrow these pronunciation patterns as their speech matures. However, persistent difficulties can sometimes signal a speech or language disorder, which may require professional support.

Mistakes are part of learning

Parents are often under enormous pressure to do everything right, including helping their children learn to speak a language. But children do not learn language by avoiding mistakes. They learn through interaction, experimentation and repetition.

Parentese helps babies focus on speech and engage socially. The funny mistakes toddlers make reveal that they are actively piecing together the complex system of language and are often signs of normal development. Language acquisition is messy, creative and remarkably sophisticated.

Speaking in an exaggerated sing-song voice to a baby is not something parents and caregivers need to feel embarrassed about.

Far from harming language acquisition, it may help lay the foundation for it.

Karen Stollznow does not work for, consult, own shares in or receive funding from any company or organization that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.

Four percent of Americans – roughly 12 million people – believe that “lizard people” secretly control the Earth. At least, that was the finding of an infamous 2013 public opinion survey.

Do so many people really believe such outlandish claims? Or do results like these partly reflect people giving silly answers or deliberately skewing surveys for fun?

US psychiatrist Alexander Scott believes the latter plays a significant role.

Using the survey as an example, he coined the term “the Lizardman constant” to describe the idea that a certain amount of noise and trolling will always exist in surveys about unusual beliefs.

As Scott warned: “Any possible source of noise – jokesters, cognitive biases, or deliberate misbehaviour – can easily overwhelm the signal.”

As researchers who study uncommon beliefs such as conspiracy theories, we wanted to investigate how this kind of cheeky trolling can muddy the waters.

Trolls and true believers

Building on earlier Australian research, we surveyed New Zealanders to test how common dishonest or joking responses were in conspiracy theory surveys.

We did this in two ways. First, we directly asked people a yes/no question at the end of the survey:

“Did you respond insincerely at any earlier point in this survey? In other words, did you give any responses that were actually just joking, trolling, or otherwise not indicating what you really think?”

Second, we included in the survey a “conspiracy theory” so ridiculous we could assume most, if not all, people who said they believed it were taking the mickey.

We asked them if they believed:

The Canadian Armed Forces have been secretly developing an elite army of genetically engineered, super intelligent, giant raccoons to invade nearby countries.

In our representative online sample of 810 New Zealanders, 8.3% of respondents confessed to being insincere in the survey.

Another 7.2% said they thought the Canadian raccoon army theory was probably or definitely true. That proportion – similar to findings from Australia – would equate to more than 300,000 adult New Zealanders.

To complicate things slightly, there was some overlap between those admitting to insincere answers and those claiming to believe the raccoon conspiracy. Combined, 13.3% of respondents fell into one or both groups – roughly one in eight people not appearing to take the survey seriously.

Importantly, these respondents were also much more likely to endorse other conspiracy theories, inflating estimates of how widespread those beliefs really are.

For instance, 6.5% of the full sample endorsed the claim that governments around the world are covering up the fact that 5G mobile networks spread coronavirus.

But once we removed the insincere responders, that figure dropped by more than half to 2.7%.

Across 13 different conspiracy theories, the estimated proportion of believers fell substantially once those respondents were excluded.

---

---

Another interesting insight from our study was that people endorsing contradictory conspiracy theories were much more likely to show signs of responding insincerely.

Previous studies have found some people appear to believe conspiracy theories that directly contradict each other. In our survey, for example, some participants agreed both that COVID-19 is a myth and that governments are covering up the fact that 5G networks spread the virus.

But nearly three-quarters of those respondents also showed signs of joking or dishonest answers.

This suggests genuinely believing contradictory conspiracy theories may be less common than previously thought.

Not every conspiracy believer is joking

Our findings add further weight to the idea that surveys may overestimate how many people truly believe some conspiracy theories – thanks, in part, to trolls.

But does that mean all conspiracy theory research is bunk?

Fortunately not. Most research in this area is not focused on counting conspiracy believers, but on understanding why people hold these beliefs and what effects they can have.

We tested several well-established findings from earlier conspiracy theory research to see whether they still held up once insincere respondents were removed from the data.

For example, previous studies have found that people who endorse conspiracy theories are more likely to see the world as a dangerous and threatening place.

We found the same pattern. In fact, removing insincere respondents made little difference to the broader relationships identified in earlier research.

Nevertheless, we recommend that future surveys include ways to gauge whether respondents are answering sincerely and account for this in the analysis. At the very least, researchers should acknowledge that trolls and joking responses can distort their results.

While our research suggests some people are taking the mickey in surveys, it also shows a significant minority genuinely appear to believe some of these claims.

In some cases – such as believing authorities are covering up the fact that the Earth is flat – this may be relatively harmless. But other conspiracy beliefs can lead to real-world harm.

Good-quality research is essential for understanding how sincere believers end up down these rabbit holes, and how those beliefs influence real-world behaviour.

Research into why people embrace conspiracy theories – and the real-world consequences of those beliefs – remains important.

But when surveys suggest millions may believe in lizard overlords or genetically engineered raccoon armies, it is also worth remembering the “Lizardman constant”: some respondents may simply be having us on.

---

The authors acknowledge the contributions of Rob Ross, Mathew Ling and Stephen Hill to this article.

---

John Kerr is supported by a Royal Society Te Apārangi Mana Tūānuku Research Leader Fellowship.

This research was supported by the Marsden Fund Council from Government funding, managed by Royal Society Te Apārangi.

Mathew Marques does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.

AI-powered digital twin technology could transform how doctors understand and treat heart disease. But if the medical data used to build these virtual models overlook biological differences between women and men, the promise of truly personalised medicine may remain incomplete.

Artificial intelligence is beginning to reshape how doctors study and treat heart disease. One of the most ambitious ideas is the “digital twin”: a computer model built from a patient’s medical data that allows researchers to simulate how a disease might develop and how treatments might work.

In cardiology, these models combine medical imaging, clinical records and biological data to create a virtual version of the heart. In the future, doctors could potentially test treatment strategies on this digital model before applying them to the patient.

But an important scientific question is emerging: What if the medical data used to build these models are missing important biological differences between women and men?

As digital health technologies move closer to clinical practice, ensuring these tools reflect the full diversity of human biology is becoming increasingly important.

In our research at Aix Marseille University on patient-specific computational models of inflammatory heart disease (MYOCAR3 funded by Civis Alliance), we are beginning to see how differences in immune responses between women and men could influence how these diseases develop and how they might appear in future digital models.

The promise of digital twins in heart medicine

Digital twins are attracting growing attention across Europe as a way to advance precision medicine.

Instead of treating patients based on average responses observed in large populations, researchers hope to build personalised models that capture the unique biological characteristics of each individual. Several European initiatives are exploring this approach.

The European Virtual Human Twin Initiative (VHT), supported by the European Commission, aims to accelerate the development of digital twin technologies for healthcare. Other projects, such as SimCardioTest, focus on building patient-specific cardiovascular models to improve diagnosis and treatment planning.

These efforts bring together engineers, clinicians and data scientists to better understand complex heart diseases. But the success of these models depends heavily on one crucial factor: the quality and representativeness of the data used to build them.

When medical data fails to represent everyone

Over the past decade, researchers have increasingly recognised that biomedical research has sometimes treated male biology as the default.

A widely cited analysis published in Nature reported that male animalshistorically outnumbered females by roughly five to one in many preclinical studies.

In cardiovascular medicine, this imbalance matters.

Heart disease remains the leading cause of death worldwide, responsible for nearly 18 million deaths each year, according to the World Health Organization.

Yet heart disease does not affect women and men in exactly the same way. Symptoms, disease mechanisms and responses to treatment can differ.

Inflammatory heart disease provides a striking example. Myocarditis, an inflammation of the heart muscle, can occur after viral infections and, in rare cases, after vaccination.

Global estimates suggest that myocarditis affects around 1.8 million people each year and occurs two to four times more frequently in men than in women, particularly among young adults.

Research published in journals such as Circulation suggests that these differences may be linked to variations in immune responses, hormonal influences and cardiac tissue biology.

For scientists developing digital heart models, this raises an important question: if datasets do not fully capture these biological differences, can digital twins accurately reproduce how the disease behaves in different patients?

From sex differences to gender-sensitive medicine

These concerns are part of a broader shift in biomedical research towards what is known as sex and gender-sensitive medicine.

This emerging field recognises that both biological sex and sociocultural gender factors influence health, disease progression and responses to treatment.

Researchers are increasingly working to integrate these dimensions into medical research, clinical practice and healthcare education.

For example, the University Hospital Zurich Heart Centerhas developed consultations dedicated to gender-sensitive cardiology. Researchers analyse international datasets, identify patterns across large patient cohorts and generate new clinical data to better understand how sex and gender influence cardiovascular disease.

At the same time, European scientific collaborations are working to strengthen how sex differences are considered in research.

The European Initiative COST Action EU-SABV is the first Europe-wide effort that focuses on improving how “sex as a biological variable” is integrated into biomedical research, helping ensure studies produce findings that are both rigorous and relevant for diverse patient populations.

Together, these efforts aim to generate better data sets, the essential foundation for reliable digital health technologies.

Building better digital medicine

Digital twins represent one of the most exciting frontiers in cardiovascular medicine. In the future, these models could allow doctors to simulate disease progression, test therapies virtually and tailor treatments to individual patients.

But the promise of digital medicine ultimately depends on the data that shape these models.

If those data fail to reflect biological differences between women and men, even the most advanced algorithms may miss part of the picture.

Ensuring that digital twins reflect the full diversity of human biology will, therefore, be essential. Only then can these technologies fulfil their promise of truly personalised medicine, not for an “average” patient, but for every patient.

---

This article was co-written with Morgane Evin (PhD – Aix-Marseille University) Hao Gao, (PhD – University of Glasgow) and Dr Franck Thuny (Hôpital Nord, APHM in Marseille).

---

A weekly e-mail in English featuring expertise from scholars and researchers. It provides an introduction to the diversity of research coming out of the continent and considers some of the key issues facing European countries. Get the newsletter!

---

L’auteur remercie l’Alliance européenne des universités civiques pour le financement de la bourse CIVIS 3i pour l’étude de recherche MYOCAR3.

The day before the Farrer byelection on May 9 in which Pauline Hanson’s One Nation party delivered a seismic shock to the Australian political landscape, her party apparatchiks banned the ABC from attending its election-eve press conference.

Thirteen days later, another party apparatchik told a journalist from Guardian Australia to “shut up” during a press conference in Adelaide about the party’s policy on oil and gas. Hanson was later heard describing the journalist as a “nasty bitch”.

And a week before Farrer, at the byelection in the Victorian state seat of Nepean, the One Nation candidate, Darren Hercus, refused to speak to the ABC because, he said, the ABC was biased.

The response of the media industry and the profession of journalism to these antidemocratic outbursts has been supine: a shameful abrogation of their obligation to defend the freedom of the press.

In Farrer, the other journalists stood by and watched as the ABC reporters were ejected. In the ensuing two weeks, not a single word of condemnation has been uttered publicly by any industry or professional leader as one abusive episode followed another.

Yet across the Pacific we see exactly how this plays out in Trump’s America.

A far-right populist leader attains power and then turns on those elements of the media he does not like, branding them the enemy of the people, undermining public trust in their reporting, and shutting them out from the access they need to do their job.

Hanson is not there yet, but her party’s instincts are clear. The ABC and Guardian Australia have put her and her party under close scrutiny, and this is her party’s response. (Although in fairness it should be added that in terms of the Farrer incident, Hanson herself said the ABC should not have been ejected.)

However, the ABC has been in Hanson’s gunsights for years. As far back as 2017 she made a deal with Malcolm Turnbull’s government: you give me an inquiry into the ABC and I’ll support the changes you want to make to media ownership laws.

It was simply a stunt to divert resources within the ABC and generate negative headlines for the national broadcaster. It led to no change because there was no basis for change.

The proximate cause of her wrath this time was an ABC story revealing a One Nation candidate in the recent South Australian state election was wanted for questioning in the United Kingdom on allegations of sexual touching.

So less than 24 hours before the polls opened in Farrer, Hanson’s chief of staff, James Ashby, ejected two ABC journalists from the party’s press conference, saying contemptuously, “Bye, bye to the ABC”.

As The Age’s media writer later noted, it was straight out of Donald Trump’s playbook.

Yet we have not heard a word of condemnation from the ABC’s editor-in-chief, Hugh Marks, or the broadcaster’s chair, Kim Williams. Nor has there been editorial commentary or an opinion column in any of our major daily newspapers. What about the Media, Entertainment and Arts Alliance? Silence.

To its credit, the ABC TV program Media Watch did not pull its punches. Its presenter, Linton Besser, described One Nation’s attitude to the press, and in particular the ABC, as ugly. Alarm bells should be ringing, he said, because the slurs about “fake news” and “the enemy of the people” might very well be hurled at others too.

Tellingly, he reported the ABC had declined to comment on the Farrer incident.

Otherwise, the nearest any of Australia’s main media outlets came was an article in The Age and Sydney Morning Herald describing the events in Farrer and setting out the background. Useful as a reference, but it did nothing to defend the principle at stake: that in a democracy, the media must be free to cover matters of public interest, and their scope to do so must not be subject to the whims and vagaries of political leaders or parties.

Instead, the media has been consumed by One Nation’s historic victory and the prospect that it will make further gains.

Hanson is presented in an heroic light: the Nine papers quote the London Telegraph referring to Hanson as “Australia’s flame-haired answer to Farage”, a reference to the UK’s far right leader Nigel Farage, who also made historic gains in recent local government elections there.

Less heroically, she is also characterised by these newspapers as “mother duck”.

The Australian tells us “the shake-up is just starting”.

And The Age and SMH capture the mood of the electorate: “Voters tell Canberra: ‘Get stuffed’.”

None of this is to say her party’s result in Farrer, its winning of four seats in South Australia and its continuing high ride in the opinion polls is anything other than a story of immense significance. It deserves all the attention it is getting.

But to ignore her party’s anti-democratic behaviour shows wilful blindness to what is happening in the United States, and suggests a complacency that it can’t happen here.

Ironically, an American journalist, Sinclair Lewis, has a lesson in this for Australia’s media. In 1935 he wrote a novel called “It Can’t Happen Here”, predicting with terrifying accuracy what Trump is doing to the American republic.

On the face of it, the exclusion of the ABC from a party press conference may appear to be a small thing. Moreover, there is a healthy belief in newsrooms that the public are not interested in journalists writing about journalists.

But this is not a story about journalists. It is a story about the functioning of the Australian democracy. It is a story requiring the insight to see a large principle in a small thing, a quality we are entitled to expect in the leaders of the fourth estate.

Denis Muller does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.

You’ve probably heard of Ozempic or Wegovy. These are the injectable drugs that have become household names for weight loss and diabetes.

Now, researchers are investigating whether these medications known as GLP-1 agonists or GLP-1 drugs could treat everything from cancer and brain disease to depression, addiction and endometriosis.

Some findings are genuinely exciting. Others are being oversold. Here’s what the science actually says.

First, how do these drugs work?

GLP-1 (glucagon-like peptide-1) is a hormone your gut naturally releases after eating. It tells your pancreas to produce insulin and signals to your brain that you’re full. These drugs mimic that hormone.

But GLP-1 receptors aren’t just in the gut. They’re found in the heart, kidneys, liver and brain. That’s what makes scientists think these drugs might do far more than manage weight.

---

Read more: The rise of Ozempic: how surprise discoveries and lizard venom led to a new class of weight-loss drugs

---

Where the evidence is already solid

Beyond diabetes and obesity, GLP-1 drugs have now earned regulatory approval in several new areas.

A trial of more than 17,000 people found semaglutide (the active drug in Ozempic/Wegovy) cut the risk of serious heart attacks and strokes by 20%, even in people without diabetes.

In a trial of almost 1,200 patients, semaglutide outperformed a placebo in treating a type of advanced liver disease.

Tirzepatide (Mounjaro) has also been shown to significantly reduce the severity of sleep apnoea, mostly because weight loss puts less pressure on the airways.

GLP-1s and cancer: promising but no clinical trial evidence

Obesity is a risk factor for at least 13 cancers, so reducing weight using GLP-1 drugs can also be expected to limit cancer risk. This was shown in a study of 86,000 adults with obesity. It found GLP-1 users had a 17% lower cancer risk.

New data suggests GLP-1 users were also less likely to see cancer spread to other organs, but this work is yet to be verified by other researchers. The anti-inflammatory effects of these drugs, which appear to work independently of weight loss, may be playing a role.

However, there have not yet been any well-controlled clinical trials that establish the link between GLP-1 drugs and preventing cancer.

Endometriosis: early but promising signs

Endometriosis affects roughly one in ten women of reproductive age. This is where tissue similar to the womb lining grows outside the uterus.

Because GLP-1 receptors are also present in reproductive tissue, these medications have shown promise in improving symptoms, with a survey of 161 women supporting this.

But, similar to cancer, there are no randomised human trials.

Addiction and smoking

GLP-1 receptors are concentrated in the brain’s reward pathways. These same circuits drive cravings for alcohol, nicotine and drugs.

An analysis of more than 1.3 million people found GLP-1 users had significantly lower rates of opioid overdose and alcohol intoxication.

A randomised trial found semaglutide reduced drinking in people with alcohol use disorder.

Early quit-smoking trials are encouraging, too.

The brain: the least clear picture for GLP-1 therapy

This is where the story gets genuinely complicated.

There are real biological reasons GLP-1 drugs could help with neurodegeneration and mental ill-health. They reduce brain inflammation, interact with dopamine (the brain’s motivation chemical) and support the gut-brain axis (the communication network that carries signals to and from the gut and brain).

However, current clinical evidence is conflicting.

For Alzheimer’s disease, researchers gave 204 participants with mild to moderate disease liraglutide (a GLP-1 that pre-dated Ozempic) and measured how much brain volume they lost. Those taking the drug showed significantly less shrinkage in key brain regions, including their temporal lobe and overall grey matter.

However, a large phase 3 trial of oral semaglutide found it wasn’t effective at slowing clinical disease progression.

Similarly, exenatide (another earlier GLP-1) showed no evidence for disease modification in a phase 3 Parkinson’s disease trial.

For mental health, current evidence is also mixed. Meta-analyses and large cohort studies show significant reductions in depression and anxiety scores among GLP-1 users.

But a separate observational study found people on these drugs had almost double the risk of major depression.

Another paper found that people with a genetic tendency toward low dopamine levels may face higher risk of depression and suicidal thoughts on these medications.

There are also case reports of serious psychiatric episodes appearing within weeks of starting treatment.

We don’t yet know who these drugs will help, and who they could seriously harm.

---

Read more: Taking a drug like Ozempic? What you need to know about risks of suicidal thoughts and contraception failure

---

What we need to be cautious about

Crucially, most of the new uses for these medications haven’t yet been tested in proper clinical trials. Large real-world studies are useful, but they can’t rule out crucial confounding factors. This means the effects may be due to external influences.

For example, most major GLP-1 trials have enrolled people with obesity or diabetes. People with mental health conditions, neurodegenerative diseases, or addiction were largely excluded. Yet, these are the very populations now being considered for treatment.

Long-term effects are also unknown. A study of more than 200,000 patients found a 2–2.5 times higher risk of drug-induced pancreatitis (dangerous inflammation of the pancreas).

Rapid weight loss also strips lean muscle, not just fat, affecting strength and metabolism, especially in older adults.

Studies have also indicated these medications carry a risk for thyroid cancer, prompting a warning on drug labels, but evidence is highly conflicting.

Time and further research will tell, but there are genuine safety concerns associated with the widespread use of these medications.

So, while the science here is genuinely exciting, we should continue to approach with informed caution.

---

Read more: Mounjaro is more effective for weight loss than Ozempic. So how does it work? And why does it cost so much?

---

Paul Joyce receives funding from The Hospital Research Foundation, Cancer Council SA, and the Australian Research Council. He is director of the Australian Controlled Release Society.

Humans are the only species known to use fully symbolic language: a system capable of expressing abstract ideas, imaginary worlds and endless combinations of meaning. But how did we get there?

The origins of language have fascinated philosophers, scientists and storytellers for thousands of years. Despite all our advances in linguistics, archaeology and cognitive science, we still don’t know exactly how language began.

That uncertainty hasn’t stopped people from trying to solve the mystery. In fact, some of the earliest theories of language’s origins are among the strangest and most entertaining ideas in the history of science.

Bow wow, ding-dong

In the 19th century, scholars proposed a flurry of curious theories to explain how speech first emerged. Many of these theories were given playful nicknames by the German philologist Max Müller, who intended them partly as satire. Yet the theories were genuine attempts to tackle one of humanity’s biggest questions.

The most famous is probably the Bow-Wow Theory. This suggested language began through imitation of natural sounds. Early humans, according to this theory, copied the noises around them: animal cries, splashing water, thunderclaps and birdsong. Words such as “buzz”, “hiss”, “bang” and “splash” seem to support the idea because they sound like what they describe.

But there is a problem. Different languages hear the same sounds differently. English dogs go “woof” or “bow-wow”, but in Turkish they go “hev-hev”, while Indonesian dogs go “guk-guk”. Even animal noises, it turns out, are filtered through culture and language.

And onomatopoeic words (words that imitate sounds) make up only a tiny fraction of our vocabularies. Most words sound nothing like their meanings. For instance, there is nothing inherently tree-like about the word “tree”.

That brings us to the Ding-Dong Theory, which argued that sounds and meanings are naturally connected in some deeper, almost mystical way.

Some words do seem to fit their meanings uncannily well. “Mini”, “teeny” and “itsy-bitsy” feel small and delicate. “Lump”, “rump” and “plump” sound heavier and rounder.

Modern linguists call this sound symbolism. One famous experiment asked participants to match two nonsense words, “bouba” and “kiki”, to two shapes: one rounded and one jagged. Most people matched “bouba” with the soft shape and “kiki” with the sharp one.

The effect is real, but it is limited. Most language still appears to be arbitrary, which means there is no natural reason why a particular sound should mean a particular thing.

Pooh-pooh, la-la, ye-he-ho

Other theories focused less on imitation and more on emotion and social interaction.

The Pooh-Pooh Theory proposed that speech began with instinctive emotional cries such as “ouch”, “oh” or perhaps less publishable exclamations uttered after stubbing a toe. According to this idea, language evolved from spontaneous vocal reactions to pain, surprise, fear or joy.

Again, though, there are complications. Interjections vary widely across languages. English speakers say “ouch”. Greeks say “aou”. Czechs might exclaim “ach”. Emotional sounds are not nearly as universal as they seem.

Then there is the wonderfully named Yo-He-Ho Theory, which suggested language emerged from rhythmic chants used during collective labour, like sailors chanting “yo-heave-ho” while hauling ropes, or workers singing together to coordinate physical effort.

The theory may sound quaint, but modern researchers do think rhythm, cooperation and social bonding played important roles in human evolution. Language is, after all, deeply social.

Another proposal, the La-La Theory, linked language to music. Charles Darwin entertained the possibility that speech evolved from musical calls used in courtship and emotional expression. Before humans spoke, perhaps we sang?

Some modern theories echo this idea. One hypothesis suggests that, as early humans began walking upright, parents increasingly needed to soothe babies from a distance. Sing-song vocalisations, cooing and proto “baby talk” may have helped strengthen emotional bonds and eventually paved the way for speech.

Gestures, symbols and brains

Today, most scientists think no single theory fully explains language origins. Instead, language probably emerged gradually through a combination of gestures, vocalisations, facial expressions, social cooperation and increasing cognitive complexity.

Some researchers argue that language began with gestures before shifting to speech. Others believe language evolved as a tool for social bonding, allowing larger groups of humans to cooperate and share information. Still others see language as tied to the evolution of symbolic thought itself: our ability to imagine, plan, remember and communicate abstract ideas.

Biology is also a factor. Humans have developed unusually precise control over the tongue, lips and vocal tract. We have evolved specialised brain regions linked to language processing.

But anatomy alone cannot explain language. Parrots can mimic speech sounds. Many animals communicate. None, however, appear to possess grammar and symbolism on the human scale. And, frustratingly, early language leaves no evidence behind. Spoken words don’t fossilise.

That lack of evidence is one reason the topic became so controversial that, in 1866, the Société de Linguistique de Paris banned discussions about language origins altogether, dismissing the field as hopelessly speculative.

Of course, theories about language origins also appear in religion and mythology. In Greek mythology, the messenger god Hermes was associated with language and communication. In the Hindu tradition, the goddess of knowledge and speech Saraswati bestowed Sanskrit upon humanity. In the Judeo-Christian tradition, language was a gift from God, who enabled Adam to name the animals in the Garden of Eden.

These stories reflect something deeply human: our urge to explain where language came from, because language itself feels almost magical. Every theory of language origins captures a small piece of the puzzle. Imitation, emotion, rhythm, music, gesture, cooperation and symbolic thought probably all played some role.

But none can provide a complete answer. The truth is that language evolved so long ago, and likely so gradually, that we will never pinpoint a single moment when it began, unless someone invents a time machine.

The birth of language will probably remain one of humanity’s greatest unsolved mysteries. Still, the theories themselves tell us something important. Humans are always trying to explain what makes us human. And language may be the most human thing of all.

Karen Stollznow does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.

Some time earlier this year, an employee at tech giant Meta built a system to track how much each staff member was using artificial intelligence (AI).

Named “Claudeonomics” after the Claude chatbot, the system created a leaderboard ranked by the number of tokens each user was exchanging with AI models, with leaders given titles such as “Token Legend”. (Tokens are tiny chunks of text, each around four characters long, that language models use for processing.)

Meta is not alone in its fascination with “tokenmaxxing”: AI labs OpenAI and Anthropic, e-commerce company Shopify, and tech investment firm Sequoia capital are all reportedly monitoring AI usage and rewarding heavy users, some of whom burn billions of tokens in a week.

Reducing a person’s performance to a single metric can be appealing for management in large corporations. But the choice of what to measure isn’t a neutral one – and if we’re not careful, it can start to rewrite our vision of what we actually value.

The score keeps the score

One of the more full-throated advocates of tokenmaxxing is Jensen Huang, chief executive of chipmaker Nvidia, who envisions a future in which tech employees negotiate high token budgets and consume tokens at rates commensurate with their salaries. Around 80% of those tokens are currently processed via Nvidia’s chips, so Huang’s enthusiasm makes sense.

But is token consumption a helpful metric for those of us who do not profit directly from AI processing volume?

In a recent book, The Score, philosopher C. Thi Nguyen analyses the rise of metrics throughout modern society and offers some helpful insights.

As Nguyen emphasises, what we measure shapes our goals. We develop metrics as tools of convenience; they standardise our measurement of values so we can compare large numbers of otherwise disparate things.

This standardisation comes at the expense of variation and distinctiveness, Nguyen argues. In business, it can make workers seem interchangeable.

Determining which employees in a large organisation are consuming the most tokens in a week is fairly straightforward. But it tells us nothing about the quality or impact of their work.

Bad metrics, bad results

In the past, questionable metrics have contributed to dramatically bad outcomes.

Prior to the 2008 global financial crisis, for example, many financial institutions had sophisticated systems of measures designed to incentivise selling as many loans as possible, as quickly as possible. Perhaps unsurprisingly, many of those loans turned out to be far riskier than anyone realised.

Nguyen emphasises that these types of metrics can tempt us into thinking they are unavoidable. But one of the central lessons of moral philosophy is that we ought to pause at moments like these and ask a couple of basic questions: what is a good life, and what values are actually worth chasing?

Huang and others usually don’t present tokenmaxxing as an answer to these question. But that’s how it functions. What is worth devoting your professional and creative energy to? Simple: grinding through tokens.

A new vision of the good life?

Silicon Valley has, of late, produced a striking number of manifestos and quasi-constitutions.

Consider Anthropic’s Claude’s Constitution, published in January 2026, which sets out the company’s aspirations for its model’s values and speech. Or look at venture capitalist Marc Andreessen’s Techno-Optimist Manifesto, which makes the case for ambitiously accelerating technological advancements in the service of promoting human flourishing.

Some of the most influential texts in the history of moral and political philosophy take this form. Thomas Jefferson wrote one – the US Declaration of Independence. Karl Marx and Friedrich Engels wrote another – The Communist Manifesto.

One way to view these Silicon Valley proclamations, and trends like tokenmaxxing, is as repackaging familiar commonplaces of corporate life – recasting mission statements and key performance indicators in a loftier register. But another is to see them as attempts to do something far more ambitious: sketch the outlines of a new and far-reaching vision of the good life.

On that view, the metrics used to measure progress against the vision matter. Tokenmaxxing, for example, is already creeping beyond the bounds of the tech industry – one report from the Wharton School at the University of Pennsylvania suggests many organisations are prioritising staff AI usage and spending as metrics.

Metrics can be useful – if we’re careful

Metrics do have their place in an ordered and complex society. There are many instances in which we might happily defer to the scores produced by simple metrics, trading nuance for convenience. Aggregate ratings on product or restaurant review sites, for example, can simplify our decision-making, even if they aren’t tailored to our specific preferences.

The problem is what Nguyen calls “value capture” – when we uncritically allow external metrics to determine our own goals and behaviour. Resisting this process involves questioning what is being measured and reframing it.

Instead of counting tokens, for example, we might use an equivalent metric such as energy consumption. Energymaxxing might sound more like conspicuous wastage, rather than improved performance.

Counting tokens is one measure of AI activity, which is itself intended as a measure of productivity, which in turn leaves aside the question of what is being produced. Not only is tokenmaxxing a dubious metric in itself, but it may also distort our vision of what matters.

Victoria Lorrimar receives funding from the John Templeton Foundation.

Tim Smartt does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.

New South Wales parliament is debating a bill this week that seeks to ban abortions performed on the basis of fetal sex.

If passed, health practitioners who perform such abortions would face professional misconduct charges and lose indemnity insurance coverage for the procedure.

At first glance, this might appear to be a defensible measure to address a practice that sits uneasily with gender equality.

But there’s little evidence sex-selective abortions are occurring in Australia.

The South Australian Law Reform Institute has warned such prohibitions would restrict and delay access to time-sensitive care. They would also prove unworkable and unenforceable.

When and how can you determine fetal sex?

Fetal sex can be determined through non-invasive prenatal testing. This blood test is taken from the pregnant person between ten and 14 weeks’ gestation, and costs around A$500 to $800 out of pocket.

Around 25–30% of pregnant people use this type of testing.

Fetal sex can also be determined through routine ultrasounds performed at around 20 weeks’ gestation.

When can you get an abortion?

Australia progressively decriminalised abortion between 2002 and 2023. It’s now legal in all states and territories.

Decriminalisation shifted decision-making power from doctors to pregnant people, recognising them as the authorities on their own pregnancies. Pregnant people are no longer required to account for their decisions to medical professionals.

However, gestational limits apply in most states and territories (aside from the Australian Capital Territory), ranging from 14 to 24 weeks.

Beyond those limits, medical signoff is still required.

What does the law say about sex-selective abortion?

The law varies by state.

South Australia explicitly prohibits it.

During its decriminalisation debate, NSW passed a statement of parliamentary opposition to the practice.

No other state or territory has specific provisions.

How common is sex-selective abortion?

The evidence base for the bill is thin. The SA Law Reform Institute found “little, if any, evidence that abortions purely on the basis of gender are a real issue in Australia”.

A 2020 NSW review similarly found sex-selective abortions “are rarely performed for the sole purpose of sex selection”.

Proponents cite a recent study as evidence of the practice among people with specific migrant backgrounds.

The study’s own authors, however, describe their findings as “indirect evidence” of a skewed sex ratio at birth. They note only that this “may be indicative of prenatal sex selection” and explicitly state it “does not establish causality”.

It’s actually an anti-abortion strategy

Sex-selection amendments are a well-documented anti-abortion strategy designed to foment stigma and discourage health practitioners from providing abortions.

This bill is one of several currently before state parliaments to limit access to abortions. It’s part of a coordinated effort to erode abortion access and contest the principle that abortion is health care.

Legislation is only part of the strategy. The parliamentary process is also used as a mechanism to advance narratives that frame abortion as morally indefensible. The goal is to undermine not just access, but the legitimacy of abortion itself.

The language of the second reading speech reflects this. The speech contains no reference to fetuses or embryos, only “unborn babies” and “girls”.

There are no “pregnant people” or even “women”, only “mothers”. Doctors who perform abortions are “abortionists.” Advocates are the “pro-abortion lobby”.

This is the lexicon of the anti-abortion movement. It constructs a worldview in which the fetus has independent moral status, the pregnant person exists only as a mother, and abortion is something only the unscrupulous would defend.

The bill’s stated justification rests on one study’s indirect and observational findings about “two migrant communities”, who are portrayed as culturally at odds with Australian values of gender equality.

The bill positions abortion restrictions as a protective, progressive measure, obscuring Australia’s uneven and incomplete record on gender equality.

What could the bill mean for pregnant people and providers?

In practice, the bill reintroduces medical gatekeeping for pregnant people. It will return pregnant people to a regime in which their reasons for terminating a pregnancy are subject to medical and legal scrutiny.

This burden is likely to fall unevenly on racialised communities: in practice, people from some ethnic communities may face greater scrutiny of their decisions from health providers.

It also imposes greater professional and personal risks on abortion providers. Australia already faces a shortage of abortion providers. Exposing health practitioners to professional sanction and voiding their indemnity insurance deters providers.

The international literature is clear: bans on sex-selective abortion do nothing to address the underlying causes of son preference. Those causes are social, economic and cultural.

What sex-selective abortion bans do is restrict reproductive autonomy – itself foundational to gender equality. This bill will not advance gender equality. It will restrict abortion access and expose providers to sanction.

Erica Millar receives funding from the Australian Research Council. She is a member of the South Australian Abortion Action Coalition.

Anna Noonan does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.

Egisto Ott is no James Bond. But the stories the 63-year-old Austrian told a Viennese jury recently would make good plotlines. Ott worked as an intelligence officer in Austria’s now-defunct Federal Office for the Protection of the Constitution and Counterterrorism. He was also moonlighting for the Russians.

Prosecutors say Ott, who was sentenced to four years in prison on May 20, handed over information to fellow Austrian Jan Marsalek, the fugitive former executive of the collapsed payments firm Wirecard. Marsalek ran a cell of Bulgarians who were convicted in London in 2025 of spying for Russia. They called themselves the “minions”.

In 2023, the London Metropolitan police in cooperation with MI5 secured chat messages between Marsalek and the minions, which led to Ott. It turned out Ott had provided sensitive data on dissidents, investigative journalists and a Russian intelligence defector. The trial also revealed that Ott had obtained the infamous “canoe-trip-mobiles”.

In 2017, high-ranking Austrian civil servants went on a canoe trip in a tributary of the Danube River. They managed to fall into the water and had their phones sent in for repairs. Their mobile data was copied by Ott and subsequently ended up in Moscow, along with Marsalek’s favourite Viennese chocolate cake, a Sachertorte. According to the chat messages, the minions had a stressful time finding the correct one (there are rival Sachertorte recipes).

What sounds like a comic opera has a sinister backstory. Since the 1950s, Austria has hosted several international organisations that are regularly targeted by intelligence services. These include Opec (the Organization of Petroleum Exporting Countries), the International Atomic Energy Agency and the United Nations Office on Drugs and Crime.

However, Austria’s reputation as a spying hub dates back even longer. The Austrian capital, Vienna, was known for espionage before and after the second world war. Arnold Deutsch, the recruiter of the Cambridge Five spy ring that passed information to the Soviet Union, hailed from Vienna. Its leading light, Kim Philby, was also talent-spotted by Soviet intelligence in the city in 1933.

But Vienna was never just a playground for Soviet intelligence. After the war, when the city was divided into four sectors for allied occupation, the UK’s foreign intelligence service, MI6, started its most creative cold war operations. Peter Lunn, head of MI6’s Vienna station, built listening stations in the city to tap Soviet phone lines.

He hid his listening tunnels underneath ordinary shops in the British zone. The first tunnel was built beneath a police station. Later, MI6 built another tunnel under a jewellery shop and then installed intelligence officers posing as a young, rich couple in a Viennese villa. While they were partying upstairs, their colleagues listened in to Russian military traffic downstairs.

The only surviving witness of a listening station today is Sir Rodric Braithwaite, whom I first interviewed in 2024. As a 19-year-old conscript, Braithwaite worked with British Army Field Security in the Aspang listening station, next to the Aspang Bahnhof (a train station on the outskirts of Vienna).

It wasn’t an uplifting experience. He sat there for long shifts with earphones on, handling old equipment and pressing recording buttons. But his memories of the tunnel are valuable because to this day MI6 has not released any photos, let alone recordings, that were made during these operations.

The Third Man

They have also not revealed the details of another highly creative intelligence operation. In 1948, a British team arrived in Vienna to film The Third Man, a thriller set in the city. They were eager to shoot scenes in the Soviet sector.

Four key people involved in the making of the film were working for British intelligence: novelist Graham Greene, director Carol Reed, “Austria advisor” Elizabeth Montagu and, most importantly, producer Sir Alexander Korda. Korda’s film production company had been providing covers for British intelligence officers in Europe since the 1930s.

Whether the filming of The Third Man was connected to Lunn’s tapping operations, or whether MI6 had to smuggle something out of the Soviet sector, is a matter of conjecture. But “odd people” appeared on the set.

The film’s sound engineer, Jack Davies, remembered a British technician who turned up out of the blue. After filming, the technician vanished completely and Davies never came across him again – something rather unusual in the small world of British film technicians.

The script girl, Angela Allen, who I first interviewed for my book Das Haus am Gordon Place (Vienna ‘48) in 2024, also realised that something odd was going on. She noticed that Carol Reed was under enormous stress in Vienna and kept himself awake with Benzedrine. He stopped taking the drug once they were back in England, filming in London studios.

Allen, who is 97 now, wasn’t surprised to find out years later that Korda was working for the British intelligence services. She told me: “He had enormous charm. He could make his people do everything for him.”

Perhaps that is one reason why Ott and Marsalek failed. To succeed as a spy in Vienna, you need to be a great illusionist like Alexander Korda.

Karina Urbach’s book about spies in Vienna, Das Haus am Gordon Place (Vienna ’48), won the German crime award. Her interviews with Angela Allen and Sir Rodric Braithwaite can be watched here: https://vimeo.com/1086100608

Mining billionaire Gina Rinehart is bankrolling the acquisition of a 9.5% stake in Southern Cross Media by Bruce McWilliam, who worked for Murdoch’s News Corp for nine years and is also a former Seven Network executive.

This venture is costing Rinehart $26 million. It does not buy her a direct stake in Southern Cross, but if McWilliam cannot uphold his side of a security deed he has signed with her, she could take control of it.

Southern Cross is one of Australia’s biggest media organisations. It owns the Seven Network, 7news.com.au, the Triple M and Hit radio brands, a raft of regional radio stations, and West Australian Newspapers.

The Rinehart-McWilliam-Murdoch axis is a formidable force, part of a new combination of media, political and mining interests, reminiscent of that which formed the Liberal Party in the 1940s. The other key figures are News Corp chair Lachlan Murdoch, One Nation leader Pauline Hanson and Liberal Party director Tony Abbott.

This is the lens through which it is instructive to assess the media’s coverage of One Nation’s rise since the Farrer byelection on May 9.

To see the parallels with the 1940s, we need to join a few dots.

Rinehart is a benefactor to Hanson. She recently bought her a light aircraft worth $1 million.

She is also a benefactor to Lachlan Murdoch. Her company Hancock Prospecting is sponsoring Sky News, owned by Murdoch’s News Corporation, to the extent of a little over $1 million for a Sky event in Dubbo called the Bush Summit.

Lachlan Murdoch is chairman of News Corporation. In 2023, he appointed Tony Abbott to the board of the News subsidiary, Fox Corporation, a day after Rupert Murdoch announced his retirement. In May this year, Abbott was elected unopposed as federal president of the Liberal Party.

Lessons from the 1940s

The parallels with the 1940s can be seen in volume two of Sally Young’s magisterial two-volume history of the Australia media, Media Monsters, where she describes the machinations that led to the formation of the Liberal Party.

The right was in disarray. Robert Menzies’ comically ill-named United Australia Party had been trounced by Labor at the 1943 election. In the aftermath, Menzies was re-elected leader but made it a condition that he had the right to form a new party.

He was backed by an entity called Collins House. This was a collection of companies connected by networks of powerful business figures who dominated mining and manufacturing. An influential figure was Lachlan Murdoch’s grandfather, Keith Murdoch. As managing director of the all-powerful Herald and Weekly Times (HWT) newspaper group, he provided a vital connection between the Collins House group and the most senior level of politics.

The HWT and other major media proprietors of the day anointed Menzies and his proposal for the new Liberal Party, at a dinner of Collins House magnates in Melbourne in 1944.

The difference between the political circumstances of the 1940s and those of today is that today there are two right-wing political parties contending for supremacy: the Liberal Party and One Nation.

Rinehart seems to be having a million quid each way on which will prevail. By contrast, if the recent coverage of One Nation by The Australian is any guide, Lachlan Murdoch has already cast his vote decisively for the Liberal Party.

The media sober up

For a fortnight after One Nation’s historic win in Farrer, the media, including News Corp media, were intoxicated by the attendant excitement and controversy: the shredding of Liberal Party support; Hanson’s ambition to be prime minister; the possibility of a Liberal-One Nation coalition.

Then, led by The Australian, the media began to sober up. On May 23, its editor-at-large, Paul Kelly, wrote that the Nationals, Liberals and One Nation were locked in a bitter competition with “life or death” consequences.

From that point on, The Australian applied the blowtorch of journalistic scrutiny to One Nation, and The Age and Sydney Morning Herald swiftly followed.

With its customary disregard for journalistic ethics, The Australian made a point of reporting that One Nation’s South Australian parliamentary team was looking like a “rainbow coalition”, one of its MPs having come out as gay with a partner who was an Indonesian Muslim.

But then it got into some serious public-interest journalism. For two days it pursued the party over its handling of rape allegations against an adviser, Sean Black.

It accused Hanson of shirking her parliamentary duties by being absent from 88% of Senate estimates hearings over the past decade. It also drew attention to the fact One Nation had failed to lodge audited financial records for three years in Queensland, and disparaged its policy proposal for citizen-initiated referendums.

On June 3 it drew on all this to publish a thundering editorial. One Nation was drifting further out to the fringes. It would be divisive and disruptive. It had appeared to lurch into blind confusion. Hanson was “not fit in any sense” for the role of prime minister.

On June 6, it led page one with a full-frontal attack, carrying the self-revealing headline: “Hanson hit”. It said Hanson had been caught out misleading voters, raising further questions about her capacity to be prime minister.

The Age and SMH were by then taking up the theme.

Suddenly Hanson was reportedly not sure if she would pitch for the prime ministership. She had admitted having had to close down party branches that had been “infiltrated by extremists”. She had insisted she would not be influenced by Rinehart despite having adopted one of Rinehart’s key policies. In other words, she was all over the place.

On June 6, the papers’ political and international editor, Peter Hartcher, described her as a firebrand provocateur who specialises in grievances without solutions and turns to scapegoats instead – Asians, First Nations people, Muslims. He pointed out that Hanson had answered “no” when asked by another journalist whether she could think of any error that Donald Trump might have made since taking power.

The same day another Age/SMH commentator, Paul Sakkal, wrote about what he called the collection of right-wing forces barracking for Hanson: openly white supremacists, people who rallied alongside neo-Nazis, supporters of the so-called sovereign citizen Dezi Freeman, who had killed two policemen. “A serious governing party cannot retain these relationships.”

A right mess

The big question after all this is how the forces brought together through the new media-politics-mining combination will resolve the obvious tensions involved in creating an effective force on the right of Australian politics.

Murdoch, through The Australian, has clearly signalled his contempt for One Nation, and already has Abbott on his team through Fox Corporation.

Rinehart, with her substantial holdings via McWilliam in Southern Cross Media, could go either way: backing Hanson or the Liberals. And her record indicates she would use her power to influence editorial decision-making to support her choice.

In 2012 she became the largest shareholder in the Fairfax company, with 14.99%. However, she refused to sign the company’s charter of editorial independence, and as a result was refused a seat on the board. She sold out in 2015.

Her history in refusing to sign the Fairfax charter is a strong indicator she would want the option of using her position on any media board to influence editorial decisions.

The old Fairfax newspapers, The Age, the SMH and the Australian Financial Review, are now owned by the Nine Entertainment Company, and stand outside the new cabal. A crucial question is whether they might prove to be a countervailing force.

One Nation set off this earthquake in Australian politics, but how the media play into the aftershocks will be a significant factor in the shaping of the new landscape.

Correction: this article originally referred to Gina Rinehart as “billionaire heiress to the Lang Hancock mining empire”. This has been amended to “mining billionaire”.

Denis Muller does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.

Artificial intelligence (AI) giant Anthropic has expanded access to a highly advanced model deemed too dangerous for public release, including Australia in the select handful of users.

The large language model, known as Claude Mythos, is now being rolled out to an additional 150 organisations across 15 countries, including the Australian government and several local businesses, as part of Project Glasswing.

In an era where large-scale AI launches are happening on a day-by-day basis, this limited, gradual release may seem particularly surprising. But Mythos is not like most other AI systems. Instead it’s an automated tool for assessing software to find critical bugs and vulnerabilities.

This managed release is deliberate, as the discovery of vulnerabilities in computer systems is useful for those who want to defend them and those who want to hack them.

However, the real nature of the impact of AI systems on cybersecurity is significantly more complex.

Finding hundreds of severe vulnerabilities

Under initial testing, Mythos has been able to identify multiple new high-risk vulnerabilities. Left unfixed, such flaws allow attackers to easily steal data or induce system crashes.

While these reports are promising, the raw data needs context. Of the 23,000 vulnerabilities flagged by Mythos, only 6,200 were estimated as high-risk by Mythos. However AI isn’t perfect, as human experts could only validate two in every three of these vulnerabilities as high-risk. Even still, the nature and severity of identified vulnerabilities has led developers to say that with Mythos “defenders finally have a chance to win, decisively”.

And winning this battle is extremely valuable.

Over the last few years, Australians have repeatedly been the victims of costly cybersecurity incidents, including Optus, Medibank Private, the Melbourne International Film Festival, and Canvas.

This barrage of attacks likely explain why the Australian Signals Directorate welcomed Australia’s inclusion in Anthropic’s Project Glasswing. While this AI-driven security offers huge potential benefits, the government so far has been tight-lipped on the specifics of how Mythos will actually be used.

Dangerous in the wrong hands

While discovering vulnerabilities is useful, defenders need to be able to respond to them. This is problematic when tools like Mythos produce large numbers of false reports, which have the potential to overwhelm unprepared cybersecurity teams.

More concerningly, while access to Mythos is currently tightly controlled, it will not be long until similar tools are available to help support hackers.

And it’s not just the vulnerabilities that AI can discover that pose risks.

AI systems more broadly are incredibly vulnerable to being tricked or exploited, with highly damaging consequences.

Just this week, hackers used Meta’s AI powered chatbot to gain access to high-profile Instagram accounts, including Barack Obama’s. They did so by tricking AI chatbots into changing account details. And, even after Instagram announced it fixed the issue, within hours there were reports of further accounts being compromised.

A similar attack known as Echoleak last year revealed how tying Microsoft Copilot to email accounts could introduce significant risks. This was made possible by sending emails to accounts monitored by Copilot’s AI. These emails tricked the AI into leaking large amounts of private and confidential information, without the email ever needing to be opened by a human. No longer do we live in a world where hackers need to convince users to click a malicious link, if they can instead convince the AI that reads emails to act dangerously.

Both Echoleak and the Instagram hacks underscore the risks we face as more and more organisations tie their critical functions to AI systems that are difficult to audit, and easy to exploit – even by just being persuasive.

A new balance point

All of this suggests the current cybersecurity landscape might be shifting to a new balance point, where defenders and hackers race to develop and exploit powerful AI tools.

Tools like Mythos aren’t a silver bullet. While they provide defenders with an additional set of eyes on where to look, it still will require expertise to work out what is real, and what isn’t.

But the advent of the AI era has already fundamentally changed the risks associated with poor cybersecurity practices. Every day a user or service provider delays a software update on one of their devices is a day where a vulnerability can be exploited.

For cybersecurity teams, ensuring compliance is already a difficult enough process that will only get worse when the speed of vulnerability discovery increases.

While they are high value targets for hackers, large organisations will likely remain safe, as they will have the resources to access and deploy tools like Mythos. But smaller, less resourced companies will likely not have the capacity to access these tools – or to react to the upcoming tsunami of cybersecurity updates.

And if they fall behind on these updates, these smaller companies will likely find themselves at far more risk than they ever have been before.

The cybersecurity divide between those with and without resources will only grow. Bridging this gap is not just an IT challenge – it’s a public safety concern that will affect us all.

Andrew Cullen does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.

Public hearings are underway this week to highlight the impacts of the government’s new National Disability Insurance Scheme (NDIS) bill to tighten eligibility and save costs.

Over the past two days, the Senate inquiry heard that if the bill passes in its current form, it risks entrenching gender inequities in the NDIS and further excluding women and girls.

We have long known the NDIS has a gender problem.

Women and girls only make up 38% of the scheme. Men outnumber women in every age category (except for 55 and over) and dominate nearly every disability type within the scheme.

From the age of 15, access requests from men are also approved at a higher rate than access requests for women.

Tighter eligibility

From January 1 2028, the bill will require scheme applicants to access all “appropriate” treatments (meaning known, evidence-based and available in Australia) likely to “materially” (meaning noticeably) improve or alleviate the impact of the impairment, before NDIS access is granted.

Applicants have always been asked to demonstrate they’ve tried other treatments before applying for the NDIS. But this new rule is likely to place an even higher burden on people with impairments that are difficult to diagnose and medically complex to treat.

Under the new permanency rules, people may have to try lots of potentially marginal treatments that might slightly improve functioning, even if their conditions are not understood, or the treatment is expensive or difficult to access.

Women are more likely than men to have medically unexplained or chronic conditions, such as myalgic encephalomyelitis/chronic fatigue syndrome and fibromyalgia, which causes chronic pain.

These “pay to prove” dynamics also disproportionately affect those with fewer financial resources. Women with disability are more likely than men with disability to live on lower incomes. They also have higher expenses and lower earning capacity due to greater caring responsibilities.

Tightening access to the scheme in this way – without first addressing costly and difficult-to-access treatment pathways – risks excluding these women and girls from key supports.

Support can only be provided for approved conditions

If the bill passes, the NDIS will only fund supports for needs that directly relate to NDIS-recognised impairments.

This reverses changes to NDIS legislation introduced two years ago, which better recognised the complex way people actually experience impairments.

We have spoken to people in our current study about chronic pain who have explained that pain from one condition (for example, a connective tissue disorder) can affect the functional impacts of another impairment (for example, autism or a psychosocial disability, for which they receive NDIS support). For them, it’s impossible to distinguish between support needs “arising directly” from their NDIS-recognised impairment and support needs that are indirectly related to that impairment.

Women are also more likely than men to experience multiple chronic health conditions and disabilities, especially in age groups under the NDIS cut-off of 65.

Narrowing the lens of assessment and restricting access in this way also has gendered consequences.

Cuts to social participation funding

The bill gives the minister power to make cuts to entire categories of supports in the future, without introducing legislation or consultation.

We got a taste of what this could look when the government announced it would cut participants’ social and community participation budgets.

The 50% across-the-board cuts will shift these responsibilities back onto informal carers – largely women.

There are more than twice as many female primary carers as male primary carers. Of those providing primary care to children with disability, the overwhelming majority (84.7%) are women.

Some 43.8% of primary carers also have disability themselves. This means that when it comes to carers, we are often talking about women with disability who are the primary carers for children with disability.

These cuts will increase unpaid caring responsibilities. The bill’s explanatory memorandum acknowledges this:

Due to the gendered nature of caring, women are more likely to be impacted by changes to the supports available […].

These changes may lead carers to cut back on paid employment, deepening women’s socioeconomic exclusion.

Cuts to social and community participation funding are also likely to increase social isolation and reduce natural safeguards of community connection for people with disability.

Women with disability are disproportionately likely to experience violence, so cutting them off from vital community participation supports poses an unacceptable risk.

What needs to be done?

The bill’s explanatory memorandum says:

opportunities to increase gender equality will be considered as part of the design and evaluation of future market reforms to delivering social and community participation and capacity building activities.

However, no timeframes, benchmarks or accountability mechanisms are provided for when or how this work will occur.

The Australian government’s approach to gender-responsive budgeting requires new policy proposals to include gender analysis proportionate to the scale, scope and likely impact of the reform.

Given the scale and implications of the proposed NDIS reforms, we need a comprehensive and publicly available gender impact analysis before this bill is passed.

We also need more certainty on what can be done for those outside the scheme who need foundational supports. The Australian government has announced the Thriving Kids initiative. However, there is limited detail on planned foundational supports for other participant groups.

Researchers and advocates have been calling for an NDIS gender strategy for years. The National Disability Insurance Agency (NDIA) began work on this in early 2025, but by 2026 this work had been paused to prioritise broader scheme reforms. Advocates, such as Women with Disabilities Australia, continue to draw attention to the gendered issues of these reforms.

The likely consequences of these reforms show a gender strategy is needed more than ever.

Sophie Yates receives funding from the National Health and Medical Research Council.

Molly Saunders does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.