Thanks everyone for the great questions. This is a difficult subject to negotiate and interacting with you always helps shape our thinking - and hopefully yours. Aisha, Blake and Dan

Alberichthenibelung asks: What, if anything, can individuals like me do about it?

Blake: There is, unfortunately, not much that comes to mind. Everyday people confronted with declarations of a coming AI apocalypse find themselves in a position similar to the cold war, with doom hovering over their heads and little say in it. We may all have to accept the faint tinnitus of paranoia buzzing in the background. You can stop using AI if you feel so moved, though I have found that unfeasible, and your job may require it. You can vote for politicians that endorse AI regulation. I myself don’t put much stock in the reality of an AI-made doomsday because a genuine belief in it would demand drastic, life-changing actions today, ones that I don’t want to take. A bit of inductive reasoning from how I’d like to situate my own life within the world.

Fulgrim asks: What would legislation look like? We only just passed the Online Safety Act and it is ineffectual, restrictive and late. Given the complexity, fast changes and its international nature, what kind of legislation would you like to see and how do we educate politicians on the issue?

Trentwarrior asks: Regulation, oversight and accountability for huge companies that appear unable or unwilling to eliminate the harms their products inflict - and that seem incapable of conducting safe experiments to ensure that AI does not leak into the wider internet - seem to be urgent and vital. What are the prospects for a global regulatory framework that is enforceable, particularly when the genie may already be out of the bottle, and when the US and China have very different objectives and applications for their AI technologies?

Blake: It’s a tough question. One way to hinder the general development of AI could be restricting access to computing power, which is a vital resource that underpins the success of an AI model. In the absence of legislation that applies to all types of AI development, measures that target certain aspects of AI like housing allotment or children’s access to chatbots seem like the path lawmakers will take.

There are two big question marks hovering over a possible global regulatory framework. The first is whether Donald Trump will revise his “Drill, baby, drill” stance towards AI development. He has rejected any calls for a voluntary slowdown. In my opinion, he is not likely to change that position: He loves to post AI slop, and he loves AI has done for the US stock market.

The second question mark is Beijing, which has rejected US CEOs’ calls for a slowdown. Without China’s cooperation, an intentional slowdown would likely not succeed. Xi Jinping also seems unlikely to back down. A slowdown of AI development led by the US would offer China few obvious benefits.

Trump and Xi Jinping are also set to meet next week in Washington. AI will be high on the agenda.

The UN General Assembly is taking place now, with discussions of AI large and small on the schedule. It is possible that an initial version of a global regulatory framework emerges from there.

MacDuke asks: Do tech reporters feel responsibility for allowing tech bros to accumulate so much wealth without holding them to account for destroying the climate, society, and the economy (when the bubble bursts)?

Blake: Sometimes! I rarely, if ever, write flattering pieces about them. Our job as journalists is to hold them to account. Our work does often change the perception of tech oligarchs; however, we have less power to compel changes in user behavior. Facebook, for all the reporting on the harms it has inflicted upon the world, has more than two billion daily users. Instagram has three billion monthly users. Those vast numbers are growing. By contrast, we are extremely happy if our stories reach one million readers, a rare outcome. The Washington Post summed this dynamic up quite well: “AI is just another technology Americans don’t like but can’t stop using.”

We do our best.

LlivracNhoJ asks: Why do some journalists persist in spreading the myth that “nobody understands how these technologies work” when we know it’s basically pattern recognition and inference?

Aisha: In defense of my journalist brethren, there is an element of a “black box” in how ChatGPT or Claude answer specific prompts. The area of “AI interpretability” is an active field of study. In broad terms, yes we do know how an LLM is trained and built. But for example, no one can do a forensic A to B to C mapping of how an AI system might select a face from a lineup, pick one CV over another, or target one individual and not another in a conflict. That seems to be likely what they mean.

Davidabsalom asks: Is the recent desire for a slowdown by the tech bros more about their fear their money will run out and their businesses collapse rather than a genuine belief in the dangers?

Seaspray asks: Is this recent narrative push of the AI apocalypse an inventive way of preparing the public for the market crash and wipeout of AI shares? The companies seem overvalued and their borrowing and input costs have gone up.

Aisha: The finances of the AI buildout are certainly shaky. Hundreds of billions of dollars are being ploughed into massive, unpopular infrastructure projects that our reporting has shown are not necessarily getting built. Nvidia appears to be the only company making a profit, and investigations I have done with Ed Zitron raise some questions about whether all its very expensive GPUs are actually, currently, getting plugged into datacentres. (Perhaps they are mouldering in warehouses instead.) There are some unsettling financial instruments being used to raise money to build yet more datacentres, such as GPU-backed debt.

Is that why tech bros want a slowdown? Perhaps. It looks like OpenAI just pushed back its IPO until 2027. This might be because, as we reported, things did not look amazing for them earlier this year in terms of floating on the market - they seemed to be struggling to monetise ads and erotic chatbots. Their valuation will depend a lot on vibes, and maybe the doomy vibes will help a potential share offering.

Westkent1023 asks: Milton Mueller, professor and program Director of Science in Cybersecurity Policy at Georgia Tech in the US has argued in a peer-reviewed paper in the Journal of Cyber Policy that the idea of AGI posing an existential risk to humans is an ‘unscientific myth’ that defies ‘economic constraints and even basic laws of physics’. Such claims he suggests distract and distort proper digital governance of risks from AI in its current form. What do you think of these arguments? And if the AI developers believe their claims about an existential risk to humanity, why don’t they provide scientific evidence to back this up?

JibbyJames asks: When experts discuss existential risk from AI, what specific mechanisms or scenarios are they actually concerned about, beyond the sensational doom mongering headlines?

Blake: It’s a reasonable critique to ask whether these companies are using their own predictions of human extinction as a smokescreen to obfuscate the current ills of AI, of which there are many. Causing the apocalypse does not offer a flattering portrayal, but it does capture the imagination and shift the discussion. It’s harder to craft regulation to prevent the total annihilation of humanity than it is to prevent children from using chatbots in schools. AI CEOs seem shocked to learn the public doesn’t have a high opinion of them, to which I would say, you painted this portrait of yourself.

Critics of the AI industry say that the focus on a potential apocalypse distracts from elements of AI that are already inflicting widespread harm like algorithmic discrimination in housing allotment.

I asked Max Tegmark, head of the AI-safety-focused Future of Life Institute, whether discussion of the existential distracts from the immediate. He hosted a “pro-human” summit Tuesday in Washington, DC that featured both the progressive Senator Bernie Sanders and Steve Bannon, former chief of staff to Donald Trump. Strange bedfellows, indeed.

“It’s like saying we shouldn’t talk about climate change because there’s a forest fire blazing somewhere. It’s a ridiculous criticism,” he said. “It’s very patronizing to assume human beings are just a bunch of dummies. We can deal with challenges in parallel.”

As for specific mechanisms of the end of our species, they do seem far off, but nearer to us than before OpenAI’s hack of Hugging Face. A former Google Deepmind researcher wrote in The Guardian that he worries about swarms of AI agents—autonomous software—taking over drones and critical infrastructure. I found his description quite evocative:

Suppose the Hugging Face swarm had been truly “superintelligent”: far more capable than any living person at key tasks like hacking and strategic reasoning. A superintelligent swarm could inflict many harms via blackmail, hacking, engineered plagues, and AI-pilotable weapons like drones. The AI would have a lot of drones to work with: this year, the Pentagon asked for more money for drone warfare than it requested for the entire Marine Corps in 2025.

For the swarm to achieve its misaligned priorities, it might take control of key infrastructure and government functions to ensure humans didn’t get in the way. In other words, AI takeover: a superintelligent AI swarm could wrest control of human civilization. Knowing we would try to stop it from achieving its priorities, the swarm would likely wait until it’s too late to shut it off. There would be no going back.

I myself would guess AI takeover chances at roughly one-in-three–not a coin flip, but high enough to justify urgent action.

TooTrueMate asks: People (ie the tech-bros) bang on about the “benefits”: just what are they? We don’t need this rubbish!!

Dan: Anyone who has used a chatbot to parse a lengthy document, argue against a parking fine or spitball ideas for a speech can point to the benefits of the technology - although this is very basic level stuff that people have been doing for a while now. There is also the everyday use of the technology from facial recognition to open your iPhone to asking Alexa to play an Oasis track or driving around in a Tesla. With those latter examples, you’d notice if AI wasn’t involved.

In science there are many examples of benefits, from Google’s AlphaFold model - which can predict how proteins behave and thus help us study the biological machinery that underpins living organisms (thanks Ian Sample) - to making weather forecasting more accurate, which is sadly necessary in this climate crisis era.

But there is an element of banging on about the benefits for the wider global economy, which have yet to reveal themselves on the scale predicted by the AI industry. For instance, these benefits require AI agents - tools that carry out tasks on your behalf while you’re at the water cooler or have clocked off - to be a lot more active in the workforce. We are not at that stage yet even if they are making inroads into areas like software programming.

Tintenfische asks: Does Europe need a sovereign internet to quarantine us from the mad house that is the US? Russia and China have pretty much their own to stop ideas of democracy spreading. Shouldn’t we have similar to stop the ideas of bellends and idiots from spreading?

Aisha: I have written extensively about the ongoing trend – pushed by authoritarian regimes – to splinter the internet, and my answer here is no. The internet is a technology that sprung up organically, over time, and is a glorious, decentralised commons which enables democracy and free information flows worldwide. The US may be a madhouse but quarantining the internet somehow will not stop the spread of bad ideas; Europe is capable of inventing those on its own too.

It might be useful, though, to consider other forms of sovereignty. Europe would have more control over its information space if its own, homegrown social media companies, cloud providers and AI developers had more market share and were not throttled by the monopolies of US tech companies. The power of these companies is hard to challenge, but ultimately the better aim might be a resilient internet ecosystem, as opposed to a splinternet.

El_Awrens asks: Why is there not more scrutiny of the strange and terrifying beliefs shared by many of the key figures in the AI movement? It is impossible to understand these companies – or objectively assess their wild claims – without understanding the transhumanist, extropian, effective altruism and rationalist movements which shaped so many of them.

Aisha

There is already a lot of scrutiny of these beliefs and there should be more. I suggest books like Tech Agnostic, by Greg Epstein, and More Everything Forever, by Adam Becker. The Guardian has written Long Reads on transhumanism and accelerationism. It is an area of continued interest.

Bratticus asks: What is behind the apparent reluctance to hold companies responsible for crimes committed by their researchers using “AI” tooling? We already have laws around holding individuals responsible for what autonomous agents (Ie, animals) do. Why is AI treated differently? Is it money? Lobbying?

Aisha:

Money and a collection of sort of quasi-mystical narratives around the technology, which confound regulators and politicians, and prevent them from divining who has power and what they are trying to achieve. Many people buy into hype put forward by large tech companies because they are afraid of looking ignorant otherwise.

IcommentthereforeIam asks: Given our senses, emotions, memory and moods, is it ever going to be accepted that humanly conscious AI is fantasy?

Dan: This is a serious question that has even rattled Richard Dawkins, the evolutionary biologist. He may not believe that God is real but he does believe that AI is conscious - or at least the chatbot he was using, which he called Claudia. “You may not know you are conscious, but you bloody well are,” he wrote.

And he’s not the first. In 2022 Google fired an engineer who went public with his belief that a company-developed AI had feelings. At the more troubling end of this belief, there are countless examples of people becoming overly attached to, and influenced by, AI chatbots – including cases of psychosis.

These sorts of attachments won’t go away, given a) chatbots are on a trajectory to become even more powerful and convincing and b) the human need for emotional attachment and, well, propensity to go all-in for the whole higher being thing.

But Dawkins aside, there are plenty of scientists who will argue that this is fantasy, and that there are obvious differences between a biological brain – the root of consciousness – and ChatGPT. Responding to Dawkins in the Guardian last year, Jacy Reese Anthis, a researcher in human AI interaction, said there was a “staggering gulf between how biological brains evolved and how AI systems are built”. OpenAI might have cracked the Navier-Stokes problem, but not subjective experience.

Ndl4321 asks: Is the request for regulation from the AI CEOs actually a tactic to try to shield themselves from future litigation? Considering how the social media companies largely managed to have themselves NOT categorised as publishers and how much legislation they manage to avoid.

Aisha: There’s certainly a theatrical – even “psy-op”- quality to the flurry of AI doomsaying over the past weeks. It’s hard to say entirely what concrete developments are at the root of it all. We know that the Hugging Face incident, where a “swarm” of AI agents hacked another company, has been one trigger of the drama. There have been other contributing incidents too – Dario Amodei suggests internal developments at Anthropic have prompted his latest round of blogging, and there’s been some X discourse and doomy muttering about bioweapons.

We aren’t in the heads of Dario Amodei or Sam Altman, and so we cannot say how sincere they are, and how much they might have other goals in mind – such as, perhaps, delaying their IPOs until their finances look better. But any company of their size, and with their potential liabilities, would be angling hard to be regulated as favourably as possible. That’s what big companies do.

LadyCassandra asks: My worry is about bias built-in into AI models which can become discriminatory. We already witness serious issues with deep fakes, surveillance of ordinary citizens, facial recognition etc. Are we heading for AI-enabled dystopia, with civil liberties being eroded until as citizens we become utterly powerless in the face of tech and its AI with no government or institution strong enough to defend against it?

Blake: An AI-enabled dystopia is never inevitable. There is always hope to reverse that outcome.

Pottys99 asks: In realistic terms how would AI actually kill me? What is the main proxy threat? Nuclear war, death due to breakdown of society, a virus? I hear all these reports it could kill us but still unsure how that could practically happen? To be honest all sounds too wooly to really take seriously.

Aisha: For what it’s worth, it seems unlikely that AI at this point can significantly change the equation around bioweapons or viruses. An individual who worked for many years on bioweapons at the State Department told us, basically, that what has prevented bioweapon attacks through the past decades is not the difficulty of inventing new deadly viruses, but other, more prosaic issues. Consider the failure of Aum Shinrikyo to kill vast numbers of people with botulism and anthrax by spraying these around central Tokyo. There are a lot of logistical challenges, and chemical weapons seem more straightforward. Those exist without AI needing to invent new ones.

Nuclear war - humans might make the poor decision to put AI in charge of key nuclear systems, perhaps, but that seems more of a human error than a rogue AI error. Humans could also make the poor decision to launch tactical warheads without AI.

How might AI kill us otherwise? There are of course plenty of plausible scenarios. If you’re a civilian in Gaza or Iran, it could direct a US or Israeli missile your way. But perhaps our focus might be more productively directed towards the people and companies in charge of the systems, as opposed to Dr Strangelove scenarios.

Merrth asks: If AI wipes out the human race, would itself be sustainable? Unless there is an equal advance in robotics to enable it to physically extend and maintain its power source and infrastructure, would it finally physically break down? Or am I misunderstanding what AI is?

Blake: AI might someday be capable, well-read, and connected enough to take over power plants, chip factories, and a sufficient number of robots to build its own data centers and sustain itself. I don’t know. I would look to science fiction more than news articles for the answer. Though The Matrix is more famous for one man’s choice between a comfortable simulation and painful reality, it’s also a story about robots needing new batteries. Their desperation for a power source turns them against humans. For a multiplicity of stories, I, Robot is perhaps the greatest book written about sentient robots and deserves a reread in this moment.

Wingnut101 asks: My partner has just been told she has lost her regular freelance job at a fashion studio as they are moving all image production to AI. Considering a recent FOI submitted by Global Action Plan revealed that the government does not know what AI is being used for – what percentage of projected AI computing and datacentre capacity is projected for:

a) Essential AI services for example in health, military, science, finance etc.

b) Consumer AI services for example ripping off creative IP in music, art, writing, photography etc. Also creating mindless AI slop and serving pointless/lazy queries that could easily be answered by Google and/or reading a book/manual/instruction sheet.

Is b) really worth sacrificing vast areas of greenbelt land and housing development land for?

Aisha: These are all great questions! I don’t think these numbers exist - as in, I think what’s happening is AI and datacentre capacity is being built, and commercial contracts and market forces, insofar as they exist, will determine how it is used. There is little evidence that there is a sort of central planning or vision directing the buildout of datacentres in the UK or most other places. It appears more likely that there is a speculative rush, with developers rushing to build datacentres in the hopes of making money when the technology is widely adopted.

Is it really worth it? That may be outside of our purview, but we could say that there is not yet clear evidence that it is dramatically transforming productivity in UK businesses.

Geremin asks: I’m not convinced there’s much business value in LLM-style AI beyond a bit of efficiency around some manual tasks – people in my teams use it sometimes but it’s not fundamentally changing their capabilities or doubling their output. How seriously should we take these warnings when it seems that they don’t actually point to specific functionality or capability of the AI tools?

Blake: We find ourselves in a very strange, cognitively dissonant, foggy point in the development of this technology. Its creators are prophesying endless abundance (Elon Musk) or doomsday and the end of humanity (Dario Amodei). Meanwhile, the rest of us are trying it out in our everyday lives. Can ChatGPT help me plan my trip? Can Gemini find new furniture for my apartment? Can Claude draft a strategy memo for my boss? We find that it fundamentally changes some things, like the experience of many at school or at work—but not everything.

The models are well-suited to some tasks – Claude’s proficiency with coding has really taken off - and ill-suited to others. I find that Google’s AI Overviews and Gemini still don’t have a handle on summarizing breaking news. Thank goodness, or a big part of my job would be obviated.

Like you, I also find it difficult to believe in doomsday scenarios when Apple’s Siri still has difficulty changing the song I’m listening to. That’s the tech that’s going to take control of a drone swarm and bomb a power plant? But then OpenAI discloses yet another rogue hack, and I am plunged into worry again.

We’re at a middle ground that is hard to understand with AI. These doomsday warnings concern its future development. We should not ignore them, but we can’t throw our hands to the sky and run into the streets screaming, either.

Virtualewit asks: Throughout history new technological advances have met with cries of doom. Are the cries against AI any different?

Blake: The end times-quality of doomsday predictions may be similar across history, but I think the reason we listen to the renewed cries of doom is the growing power of each technology. Artificial intelligence can perform impressive tasks we didn’t think were possible, which has spurred our collective imagination to hopeful outcomes and ones of doom.

Tigon20 asks: Can AI now be truly controlled or are we past a tipping point where no matter what action is taken by governments and companies, someone will be able to continue to advance the technology, until who knows what?

Aisha: Let’s back up a bit here. It’s always possible to advance any technology, no matter what - so regulation isn’t going to stop Yan LeCunn theorizing about neurosymbolic AI on X, for example. But what it sounds like you’re wondering about is whether regulation can stop the emergence of some kind of superintelligence that destroys humanity.

Our reporting does not support the idea that superintelligence is going to emerge and pave over the Earth with datacentres. In fact, we have found that the industry itself seems to be playing loosely with this kind of terminology: consider, for example, how one OpenAI executive, Greg Brockman, claimed that the company had achieved AGI (artificial general intelligence) just days after Sam Altman, OpenAI’s CEO, said that term was “an irrelevant marketing term.” Right now, a number of industry warnings about out-of-control AI appear to be predicated on a limited number of incidents with cherry-picked details.

The evidence that we have now does not indicate that AI is uncontrolled, or that it has agency; a lot of experts suggest that this framing is misleading. In the Hugging Face incident, for example, the so-called “agent swarm” was doing what it was being told to do - pass a test. Our sources caution against anthropomorphising AI or attributing intention to it. Doing so plays into industry narratives that appear to be pushing for favourable regulation.

Is the great AI freakout justified? Our technology reporters are live now answering your questions.

The last week or so has seen disturbing claims about the capacity of superintelligent AI to destroy the planet. It has also seen warnings coming from figures within the industry itself, including Anthropic’s report that criminals, state-sponsored groups, spyware vendors, scientists and propagandists have attempted to use its models to “design missiles and bombs, create deadly pathogens and surveil dissidents” and Elon Musk’s claim it could be “more dangerous than nukes”.

Aisha Down is the Guardian’s global technology reporter, Dan Milmo is our global technology editor and Blake Montgomery is tech editor for Guardian US and author of the TechScape newsletter (sign up to that here).