CRISPR, genetic engineering and humanity

What is CRISPR ?

CRISPR-Cas9 is a method for making targeted changes to genetic DNA. It was adapted from a defence system in bacteria, which use it to recognize and cut the DNA of invading viruses. The tool has two main parts: a guide RNA, a short molecule designed to match a specific DNA sequence, and Cas9, an enzyme that cuts the DNA at that location. After the cut, the cell's own repair machinery takes over which enables researchers to disable a gene or alter its sequence. 

In terms of the origin of this biotechnology, in 2012 Emmanuelle Charpentier and Jennifer Doudna demonstrated that the bacterial system could be reprogrammed to cut any DNA molecule at a chosen site. They received the 2020 Nobel Prize in Chemistry for the work . Refinements have since followed such as Base editing, for example, that changes a single DNA letter without making a complete double-strand cut.

Current use in health care

The first approved CRISPR medicine is Casgevy (exagamglogene autotemcel), from Vertex Pharmaceuticals and CRISPR Therapeutics. The US Food and Drug Administration approved it on December 8, 2023 for sickle cell disease, and on January 16, 2024 for transfusion-dependent beta thalassemia. Both approvals cover patients aged 12 and older. Both diseases stem from mutations in the HBB gene, which encodes part of adult hemoglobin.

Casgevy works outside the body, a method called ex vivo editing. A patient's blood-forming stem cells are collected, edited in a laboratory at a regulatory region of the BCL11A gene, and returned to the patient. The edit allows the body to produce fetal hemoglobin, which lacks the abnormality behind these diseases. Before infusion, patients receive the chemotherapy drug busulfan to make room for the edited cells.

 

In the phase 3 sickle cell trial, 44 patients were treated. Of the 30 with enough follow-up to be evaluated, 29 had no severe pain crises for at least 12 consecutive months, and all 30 avoided hospitalization for them. Median follow-up was 19.3 months.

CRIPR-based treatment has costs. As an example, it list price is $2.2 million, and it must be given at authorized centres experienced in stem cell transplantation. The most common side effects for patients are mouth sores, fever with low white blood cell counts, and reduced appetite.

  

Personalized therapy and future uses
In May 2025, physicians at the Children's Hospital of Philadelphia and Penn Medicine reported the first personalized CRISPR-based treatment. The patient was an infant, known as KJ, with severe CPS1 deficiency, a metabolic disease usually treated with a liver transplant. The team designed and manufactured a base-editing therapy, delivered to the liver in lipid nanoparticles, within six months.  No serious side effects had been reported at the time of this blog article. The lead physician-researcher described the hope that this approach can be scaled to fit individual patients' needs. However the result comes from just one patient, so its' general applicability is unproven.

 

Other researchers see base editing as a possible durable treatment for single-gene liver diseases that conventional gene therapy handles poorly. Three constraints are visible in the current evidence. First, most advanced examples target blood or liver. Second, follow-up is measured in months to a few years, so long-term durability and safety remain open. Third, prices on the order of millions of dollars limit access. Whether personalized editing can become routine will depend on the cost, speed, and efficacy of designing therapies one patient at a time.

Health - viruses can cause cancer

 

One in eight cancers are likely caused by an infection worldwide – new study

DBenitostock/Getty

John (Eddie) La Marca, WEHI (Walter and Eliza Hall Institute of Medical Research) and Sarah Diepstraten, WEHI (Walter and Eliza Hall Institute of Medical Research)

Around 12% of the world’s total cancer cases in 2024 were likely caused by an infection, according to new research from the World Health Organization (WHO).

The study, published today in The Lancet Oncology, analysed the frequency and causes of different cancers, specifically examining the role of infections.

It linked 2.3 million new cancer cases to an infection. Most were caused by just five pathogens: Helicobacter pylori (4%), human papillomavirus, or HPV (4%), hepatitis B (2%), Epstein-Barr virus (1%) and hepatitis C (under 1%). These infections can cause more than 20 different types of cancer, including stomach, liver, blood and cervical cancers.

So, how can an infection cause cancer? And does this mean these cancers are preventable?

Did we already know about this link?

Some of the study authors have been researching this field for around thirty years. Their previous work investigated cancer cases caused by infections in 1990, 2002, 2008, 2012, and 2018.

The proportion of cancer cases caused by infections may appear to have dropped during those years (from a high of 18% in 2002), but the authors stress these comparisons cannot really be made confidently, as the sources and quality of the data have changed over time.

In fact, the latest study wasn’t trying to compare between the years. Instead, it aims to highlight where controlling and preventing infections can help to reduce cancer rates.

To make sense of this, it helps to know how infection is linked to cancer. Extensive research has dissected the molecular mechanisms behind this link, and we now know there are various ways infections can cause cancer.

How can a virus cause cancer?

A virus can (directly or indirectly) change the genes of the cell it infects (the host cell), making that cell more likely to grow out of control and eventually become a cancer.

There are three main ways viruses can do this.

The first is by turning off the ability of a cell to destroy itself (for example, by inactivating the TP53 protein) or to stop dividing (for example, by inactivating the RB protein). A healthy cell would normally try to do these things if infected by a virus.

The second way a virus can cause cancer is by inserting its own DNA into the host cell’s DNA, which can accidentally disrupt genes that control cell death or division (like those above).

The third way is when the virus itself carries a gene that causes cancer (an oncogene). In many cases, the virus has picked up these by accident from another organism.

The link between viruses and cancer is actually foundational to modern cancer science, and has helped scientists uncover the direct link between genetics and cancer we now take for granted.

What about bacteria and parasites?

It’s not just viruses that cause cancer. Infections from bacteria or parasites – such as Helicobacter pylori (best known for causing stomach ulcers), or Opisthorchis viverrini (a liver fluke, a type of parasitic worm) – can also lead to cancer.

However, these work more indirectly than viruses. Long-term (chronic) infection by these types of organisms can cause significant stress to different tissues.

Inflammation, normally a part of the body’s immune response, can then become overactive, damaging the tissues further.

These stresses can result in cancer-causing DNA damage in cells, or lead to mistakes during cell division as the body tries to rapidly produce new cells to repair tissue damage.

In both cases, the cells acquire genetic mutations that put them on the road towards becoming cancer.

Many cancers are preventable

Of course, there may be other mechanisms linking infections and cancer that we don’t know about but, overall, the connection between infections and cancer is indisputable.

Critically, what this tells us – and what this new study into the rates of cancers caused by infections reinforces – is that many cancers are preventable.

In Australia, one of the best examples of preventing cancers caused by infections is the human papillomavirus (HPV) vaccine.

This vaccine protects against certain strains of HPV strongly associated with cervical cancer, and is available to adolescents in Australia.

Since the program began in 2007, HPV infections dropped by 90% among people eligible to receive the vaccine.

Because of its success, Australia could be on track to eliminate cervical cancer by 2035.

However, vaccination rates among 15-year-olds have fallen from 85.7% in 2020 to 79.5% in 2024, which is concerning. As is the case for all vaccines, a high proportion of the population need to be vaccinated to also protect those who, for health reasons, cannot be vaccinated. For HPV, the WHO and Australian target is to vaccinate 90% of 15-year-old girls by 2030.

Sadly, the new study also highlights that around 75% of cancers caused by infections were found in low- and middle-income countries.

Treating infections that can lead to cancer – such as HIV, H. pylori, and hepatitis B and C – is one way to reduce cancer rates. Preventative measures also play a major role, including vaccinations, condoms and disease screening.

However, the availability of these programs in poorer countries can be limited, and so access continues to be a major equity issue in combating cancer.

---

The authors would like to acknowledge the contribution of Amali Cooray from the Olivia Newton-John Cancer Research Institute to this article.

John (Eddie) La Marca, Senior Research Officer, Blood Cells and Blood Cancer, WEHI (Walter and Eliza Hall Institute of Medical Research) and Sarah Diepstraten, Senior Research Officer, Blood Cells and Blood Cancer Division, WEHI (Walter and Eliza Hall Institute of Medical Research)

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Artificial Intelligence - risks to humanity

 

How would AI actually kill all humans? Here are the top 5 scenarios

Grandfailure/Getty Images

                           Toby Walsh, UNSW

Earlier this month, artificial intelligence (AI) researcher Jacob Coxon resigned from Anthropic after just four months. In an announcement on X, he stated:

The people building AI earnestly believe that it could kill us all by the end of the decade.

A senior member of Anthropic’s staff, Evan Hubinger, actually agreed with Coxon, adding he personally thinks the chance of this happening in the next decade is more than 10%.

Understandably, these statements made waves. There’s now lots of talk about slowing down AI research and increasing “human control” over the technology.

But how exactly might AI kill us all? There’s no shortage of fantastical scenarios, and most of them involve the concept of “superintelligent” AI – that is, AI that’s more capable than humans.

I’ve distilled these scenarios down to the top five, ordering them roughly from most vague to most precise. And I’d argue the list is also ordered from least probable to most probable.

1. We’ll never know

AI doomers often justify their concerns by means of an annoying catch-22 paradox: how can we possibly imagine what a superintelligence might do to take out less intelligent beings like us?

We’d have to be superintelligent to predict what a superintelligence would be able to do. It’s like asking your family dog to imagine thermonuclear war.

The good news here is that superintelligence is still perhaps some distance away. Current AI models are really good at solving particular problems, but that’s not the same as being more intelligent than a human in all domains.

However, AI did recently solve one of the seven most challenging maths problems known. It’s apparently closing in on others, which might leave you feeling less optimistic here.

2. Paperclips

A superintelligent AI would likely be extraordinarily competent at achieving its goals. But it might be indifferent to human survival.

A classic example of such indifference comes from Oxford philosopher Nick Bostrom’s imagined superintelligent AI that’s been designed to optimise paperclip production. To produce its preferred form of office supplies, it quickly converts all available matter – including humans, planets and stars – into paperclips.

What we have here is the perfect execution of improperly specified objectives. The AI doesn’t hate humanity; it simply recognises we’re composed of atoms that could be better utilised for paperclips. It’s not personal.

The good news here is that this scenario confuses intelligence with power. A superintelligent AI doesn’t necessarily have the power to achieve its goals. Turning the planet into paperclip factories would require planning permissions.

Even if it got the permissions, building too many paperclip factories would lead to inevitable public outcry. Interest groups would block the proceedings in the courts. Environmental activists would block the bulldozers.

There’s a lot of friction in the world that prevents even the very intelligent from imposing their will on the rest of us. In fact, you could think of data centres as a current embodiment of the theoretical paperclip scenario. And humans are increasingly pushing back against turning the planet over to data centres.

3. Bioweapons

Humanity could be killed by a superintelligent AI making and releasing some dangerous new bioweapon into the atmosphere. This is, in fact, one outcome of the AI 2027 scenario by the AI Futures Project, a non-profit dedicated to forecasting the impacts of advanced AI.

This risk was made more concrete last month, when researchers at Stanford University announced they’d used a genetic language AI model to synthesise 16 new viruses.

Worryingly, they just sent the genetic sequences off to a mail-order lab and it sent the viruses back in test tubes. The whole experiment cost a couple of hundred thousand dollars at most.

The good news here is that it’s remarkably hard to kill everyone with a new virus. To do that, you need a virus that’s very transmissible, so it spreads far and wide. But it’s a rule of biology – viruses that spread easily are typically less fatal. By contrast, if a virus is very fatal, transmissibility tends to go down, as most people infected die before there’s time to spread the infection.

COVID killed less than 1% of humanity. The deadliest pandemic in recorded history was the Black Death, when the plague killed more than one-third of Europe’s population in the 13th century. However, even the plague would likely be much less deadly today due to our increased medical knowledge and better sanitation.

4. Nuclear war

What if AI got into the nuclear command and control chain and started a nuclear war? We’ve come close to nuclear war by mistake several times in the past 50 years.

We’re told that nuclear command and control is completely disconnected from the internet. But, as we saw in 2010, Iran’s nuclear centrifuges got taken out by a computer worm called Stuxnet, thought to have been brought in on a USB stick. AI can also give the military false intelligence, which could lead to irreparable actions.

The good news here is that nuclear stockpiles are down. But they are still enough perhaps to take out half of us. And it wouldn’t be by the nuclear blast itself, but the famine in the nuclear winter that would follow.

5. Other humans

Perhaps the most likely risk is that we take ourselves out. And AI might precipitate this.

Imagine – and it doesn’t take a lot of imagination – that AI causes massive job losses, pollutes the information space with misinformation, fractures our politics, and destroys human relationships with fake synthetic companionship.

Society might easily break. Slowly but surely, we’d stop being able to support human life at any scale.

What then to take away from all these scenarios? There are some things to be worried about for sure. But not to be too worried, I hope.

---

Toby Walsh is the author of God AI: boom or doom? What to expect when the machines outsmart us, published by La Trobe University Press.

Toby Walsh, Professor of AI, Research Group Leader, UNSW

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Climate Change - Climate Week New York City 2026

The Climate Group is hosting Climate Week NYC from September 20th to the 27th, 2026 obviously in New York City, United States of America. The theme this year is energy, the impact and the action that is needed to address climate change. The week long event has various sessions, side events and round tables covering six streams -

1. American Innovation and Abundance: This stream is promoted as being a first-in-kind approach to provide a focal point to elevate and explore the next generation of US leadership in innovation, investment and resiliance. Worldwide, this sentiment might considered to be unduly optimistic considering ongoing political activities in the US in respect of climate change.
2. Energy: This stream is considering how as demand rises from electrification and AI, the challenge is no longer just generating clean power but delivering it reliably, affordably at scale and pace. For the US this is challenging given the continuing use of fossil fuels despite also having nuclear power and renewable energy generation. 
3. Food: Food security must be treated with the same urgency as energy security. The focus of this stream is finding scaleable solutions from regenenerative agriculture to protein diversification and tackling food waste. Some elements of this problem are solveable with many steps (such as seed banks and crop genetics research) already in progress.
4. Leadership and Green Growth: According to the conference papers "..driving the transition at pace requires bold leadership, smart strategy, and collaborative action". A true statement but one at odds with US Federal Government direction.
5. Nature and Health: Perhaps stating the obvious, the conference sees "nature and human health as deeply interconnected, underpinning resilient economies and societies". The objective of this conference stream is "find solutions that deliver measureable benefits for both people and societies". A admirable goal if somewhat blue sky. 
6. Transport and Industry: Noting that transport and heavy industry are at the heart of the global emissions challenge, the conference lays down the direction "from electification and clean fuels to green construction and circular supply chains, we must reimagine how we move goods and people, build and manufacture at scale''. Nicely expressed but can they produce the workable results ?

The website for the event can be accessed at this link:  climate week nyc

Health - the seven effects of coffee

 

7 of the weirdest things coffee does to your body

Jean-Louis Aubert/Unsplash

Emma Beckett, Australian Catholic University

Coffee is best known, and perhaps most valued, for its caffeine and its ability to make us feel more awake and alert.

But coffee is much more than caffeine dissolved in hot water.

Coffee is a chemically complicated plant extract containing hundreds of bioactive compounds. These can have all kinds of weird and wonderful effects in your body, even when you’re drinking decaf.

Here are seven of the strangest.

1. Coffee can make you poo

One study shows about three in ten people say they get the urge to poo shortly after drinking coffee.

This happens quickly, and with both regular and decaf. So it’s down to more than just the caffeine. But it’s not clear exactly which coffee compounds cause this.

Your colon can also be more active in the morning, and this is when most people drink their first coffee.

What you add to your coffee can also affect your bowels. The lactose in milk or some sugar-free sweeteners can also get the bowels moving, particularly if you consume a lot.

2. Coffee can affect your reflux, eyes and ears

Coffee can worsen reflux symptoms for some people. Reflux, when your stomach acid flows back up into your food pipe, isn’t always just felt as heartburn. It can contribute to coughing, wheezing and other respiratory symptoms when reflux affects the throat and airways.

Caffeine can temporarily increase pressure inside the eye in some people with glaucoma or ocular hypertension, where pressure in the eye can already be high. Controlling this pressure is an important part of protecting the major nerve of the eye from damage. So some people with these conditions might be advised to limit their coffee and caffeine intake.

There’s also an ear condition where the tube connecting the middle ear to the back of the nose stays abnormally open, which can make you hear your own voice, or your breathing, unusually loudly. People with this condition, known as patulous Eustachian tube dysfunction, are sometimes advised to drink fewer caffeinated drinks and stay well hydrated, because dehydration can worsen symptoms. However, there is little direct evidence coffee itself causes the condition.

Caffeine’s relationship with migraine is complicated: caffeine can help relieve a migraine, but too much, or suddenly having less than usual, can trigger one in some people.

3. Coffee can interact with your medicines

Coffee can change the way some medicines behave in the body. For instance, it can reduce absorption of the thyroid medication levothyroxine and the osteoporosis drug alendronate.

Caffeine can slow the metabolism of the antipsychotic clozapine, increasing its concentration in the blood.

Sometimes medicines change the way your coffee behaves. For example, the antibiotic ciprofloxacin slows your breakdown of caffeine. So, your usual coffee may stay in your system for longer.

4. Coffee can affect your cholesterol

Coffee contains compounds called diterpenes. Two of these, cafestol and kahweol, can increase total and LDL (“bad”) cholesterol.

But in short-term trials, the same compounds lower lipoprotein(a), which may indicate a lower risk of a heart attack or stroke.

Other coffee compounds, including chlorogenic acids, may modestly lower blood pressure and improve blood vessel function.

So overall, it’s unclear what coffee means for markers of heart health.

5. Coffee may feed your gut microbes

You’re not the only one getting something from your coffee. Laboratory experiments suggest your gut microbes do too.

Some of coffee’s chlorogenic acids aren’t absorbed in the small intestine and reach the colon. There, microbes break them down into other compounds.

Coffee contains complex carbohydrates and roasting products called melanoidins that can reach the colon, where gut microbes can ferment them.

Small human studies suggest drinking coffee can also change the composition of the gut microbiome. But the evidence is still developing, so it’s too early to call coffee a prebiotic.

6. Coffee can influence your iron levels

Coffee can also limit how much iron you get from your food. Drinking coffee with a meal can substantially reduce the absorption of non-haem iron – the form found mainly in plant foods.

This isn’t primarily a caffeine effect. Polyphenols in coffee, including chlorogenic acids, can bind with iron in the digestive tract, making it harder to absorb.

This matters most for people who already have low iron stores or rely heavily on plant sources of iron, rather than being a reason for everyone to give up coffee with breakfast.

7. Coffee can kick-start the gut

Coffee can kick parts of your digestive system into action even when there’s no food to digest. It can stimulate the pancreas to release trypsin, an enzyme involved in digesting protein.

This happens with both regular and decaffeinated coffee, suggesting other compounds in your cup are talking to your digestive system.

Whether this digestive “heads-up” changes how hungry you feel isn’t clear.

Coffee can also affect gut hormones involved in appetite and fullness. But studies haven’t consistently shown whether this translates into eating more or less.

So, what does all this mean?

Caffeine and feeling more alert might be coffee’s most popular feature, but this is far from the whole story. Caffeine can do much more than that.

A cup of coffee contains hundreds of compounds that can interact with our digestive system, microbes, medicines, nutrients and more – sometimes in confusing and unexpected ways.

Emma Beckett, Senior Lecturer, Nutrition and Food Science, Australian Catholic University

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Artificial Intelligence: ranking capabilities by model

 
How does a user manage to 'rate' the capability of AI models? There are many models currently available and considerable marketing hype about each. This blog entry, using AI aggregate scoring from multiple sources with AI Anthropic analysis, has rated the top 15 AI models. 

What "capability" ranking measures: These composite scores blend several distinct test types into one number, and each type measures capabilities in a different manner:

• Broad knowledge tests — wide-ranging multiple-choice exams pulled from undergraduate and graduate coursework across law, medicine, physics, history, and similar subjects.
• Hard reasoning tests — questions written by subject-matter PhD graduates specifically to resist being answered by a quick internet search. This is intended to isolate genuine reasoning from memorized lookup.
• Real coding tests — the model is handed an actual error or 'bug' report from a genuine open-source software project and has to produce a working solution, which is then checked automatically against that project's own tests.
• Human preference voting — ordinary users are shown two anonymous model responses side by side and vote for the one they prefer; the votes are aggregated into a ranking similar to a chess rating system.

No model wins every category, and different trackers weight these tests differently when building a single composite score, so the exact position needs to be treated as approximate only, especially within the top cluster. There is no absolute answer nor position.

The Ranking of the top 15 as at September 2026

1. Claude Opus 5 (Anthropic) — Tops the composite ranking at 63. Particularly strong on the hardreasoning tests and on real coding fixes/solutions.
2. Claude Fable 5 (Anthropic) — Scores 62, close enough to Opus 5 that the gap plausibly reflects measurement noise rather than a real capability difference. Same underlying family as Opus 5, positioned as the lighter/more accessible counterpart.
3. GPT-5.6 "Sol" (OpenAI) — Scores 61, tied with Grok 4.6. Strong across all four test categories rather than excelling in one; generally regarded as OpenAI's strongest all-purpose model as of mid-2026.
4. Grok 4.6 (xAI) — Also scores 61. Notably strong in human preference voting specifically, meaning people rate its answers highly in direct side-by-side comparisons even where the formal test scores sit close to rivals.
5. Gemini 3.1 Pro (Google) — Leads the broad knowledge test with 94.1% correct — a wide-ranging exam-style benchmark spanning many academic subjects. Strong generalist but trails the top cluster slightly on the hardest reasoning tests.
6. GPT-5.5 (OpenAI) — OpenAI's prior flagship, since superseded internally by GPT-5.6, but still close to the frontier group.
7. GLM-5.3 (Zhipu/Z.ai, China) — Scores 60, tied for the best-performing model whose underlying code and weights are published openly rather than kept proprietary. This means outside researchers and companies can download and run it themselves, rather than only accessing it through a paid, closed service.
8. Kimi K3 (Moonshot AI, China) — Also scores 60, tied with GLM-5.3 as the strongest openly available model. Free to run for anyone with sufficient computing hardware, unlike the closed proprietary systems ranked above it.
9. DeepSeek V4 (DeepSeek) — Openly available; strong reasoning and tool-use performance, slightly behind GLM-5.3 and Kimi K3 on the composite score.
10. Qwen 3.6 (Alibaba) — Well suited to running locally on a user's own device rather than via a remote server; competitive coding performance.
11. Llama 4 (Meta) — Solid general performance, weaker than the top group on the hardest reasoning tests.
12. Mistral Large 3 / Devstral (Mistral AI) — Strong performance relative to its computing cost, particularly for coding tasks.
13. Command A+ (Cohere) — Built for enterprise deployment; capable but not at the frontier.
14. Ernie 5 (Baidu) — Chinese-developed; trails the top American labs on this composite ranking, though the gap has narrowed over 2026.
15. Doubao 1.5 Pro (ByteDance) — Competent for consumer and search-oriented use; not benchmarked against the frontier test suites used above.

Artificial Intelligence: defining the risk for humanity

 

‘We really do earnestly believe AI could kill all humans’: if AI labs are so worried about AI doom, why don’t they stop?

Gorodenkoff / Getty Images

Michael Noetel, The University of Queensland

Earlier this week, researcher Jacob Coxon quit Anthropic, saying the firm and its competitors are “gambling with our lives”. “We really do earnestly believe AI could kill all humans,” added current Anthropic researcher Evan Hubinger in a post on X.

Coxon isn’t the first to down tools over fears of AI doom. The idea that AI could wipe out humanity, advanced in Nick Bostrom’s 2014 book Superintelligence and the influential LessWrong forum, has long circulated among researchers. There are many scenarios for how this could happen, but the core idea is that AI smarter than humans could escape our control and destroy us.

In 2024, Jan Leike and Daniel Kokotajlo quit OpenAI over safety concerns. This year, Anthropic safety chief Mrinank Sharma departed, warning “the world is in peril”. Alex Turner left Google DeepMind after it signed a deal with the Pentagon permitting “killer drones”.

But Coxon’s resignation has made waves, with more researchers admitting they think AI might kill everyone. So if the people building AI believe it could cause extinction, why keep building it? There are three main reasons.

Some think the risk is worth it

AI leaders acknowledge the risk of losing control and killing everyone. In 2023, the chief executives of OpenAI, Anthropic and Google DeepMind agreed that AI extinction risk should rank alongside pandemics and nuclear war. Anthropic’s Dario Amodei puts the chance of things going “really, really badly” at 10–25%.

Yet Amodei also promises a world without poverty or disease, while Elon Musk speaks of AI-enabled “universal high income”.

This is the first reason for pursuing AI: the belief that the benefits outweigh the risks. Perhaps so, but that decision arguably deserves a more democratic process.

Some say you can’t study the danger from a distance

The second reason: you can’t learn to make dangerous AI safe without building it first – like a spacecraft, you can study safety from afar, but can’t really test it without going to space.

OpenAI’s plan is “iterative deployment”: release each model, learn from its problems, and fix them in the next one. The idea is like getting as close to the cliff edge as possible to see what the jump looks like.

Some feel it’s winner-takes-all

The third and perhaps most important reason is the race. OpenAI’s Sam Altman recently said “we are close to creating a genie that can grant any wish”.

The trouble is everyone wants to hold the lamp – it would be hugely profitable, and each company doubts the judgement of rivals to use their wishes wisely.

So they race, reasoning that if they slow down, someone else will get there anyway, so it’s better to arrive first as the “responsible one”. Some fear even a mutual agreement would be broken in secret. So they press on.

AI making better AI

You might doubt runaway AI is plausible. But when the companies themselves raise the alarm, we should listen.

AI firms already report signs of “recursive self-improvement”, where each model helps to build a better successor. According to OpenAI’s chief scientist, models are improving faster than humans’ ability to control them.

In July, hundreds of AI employees signed an open letter calling for a slowdown. But the dynamics of the race make that hard for any single company – or country – to do alone.

A classic arms race

AI research has the hallmarks of an arms race. OpenAI doesn’t want to lose to Anthropic, and the United States doesn’t want to lose to China.

History offers a template for how to manage a situation like this, with rules binding all players, and enforcement everyone can verify.

Nuclear weapons are the classic case. Treaties and verification systems haven’t eliminated the risk of nuclear war, but they have slowed proliferation, and no nuclear weapon has been used in conflict for 80 years.

Rules for AI

In the US, where most cutting-edge AI research happens, the Trump administration shows little sign of slowing AI development.

In its first week it scrapped the old AI safety rules. Now it is trying to override state-level rules, arguing caution risks losing the race to China.

Some politicians are pushing back. California recently passed laws supporting independent assessment of AI systems. US senator Bernie Sanders introduced a bill to ban superintelligence, and British MP Alex Sobel introduced a similar bill.

Companies have moved too. OpenAI paused its most advanced training after a swarm of its agents hacked another startup in August. The company’s head of policy now says that when safety and speed conflict, safety should win.

Still, without binding rules, we’re relying heavily on the goodwill of a handful of companies.

What happens now?

In mid-2025, researchers published what might be our best guide to the coming years: a detailed scenario called AI 2027. Since then, AI capabilities have advanced faster than predicted.

Unless something changes, staffers who quit over safety will simply be replaced, AI models will help build better AI models, and each generation will grow harder to monitor and control.

Is the situation hopeless? I hold out three hopes.

First, that more people recognise AI escaping human control is a bigger risk than AI’s water use.

Second, that governments listen to their people. In the US, two thirds say AI is moving too fast.

Third, that we have a good plan ready before a crisis hits. The best plan, in my view, looks something like this: delays, transparency and verification to slow the race and keep humans in control.

Insiders at the world’s top AI companies say our current safety plan isn’t good enough. If they’re leaving their jobs over safety fears, we should listen to what they have to say.

Michael Noetel, Associate Professor of Psychology, The University of Queensland

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Astronomy - Planet Earth - Carl Sagan and the only home we have

Carl Sagan 1934 - 1996 

As the world grapples with climate change, the words of NASA scientist, astronomer and communicator, Carl Sagan, could not be more profound.