Great to be on Sky News discussing the Grok deepfakes scandal and broader AI problems. Developers don't really understand how these AIs work internally or how to keep them under control. As companies keep racing to superintelligence, the threat is only growing.

Grok has governments playing catch-up on AI. The scandal is a symptom of a deeper problem: AI developers can't really keep their AIs under control. As AIs grow ever more powerful, we need a proactive approach to regulation. Check out our latest article! controlai.news/p/grok-shows-g…

Disappointing that Dario's essay again doesn't seriously engage with other opinions on topics like safety cooperation with China, and just cites vaguely related Wikipedia pages instead of the actual research specifically on such topics.

Recent news articles have made errors in their reporting on how our timelines forecasts have shifted since AI 2027. To be fair, we haven't done the best job making our updates easier to understand. Well now we have a blog post to help! Including the attached summary table.

Dario seems to object to stronger AI policy with 'there would be backlash,' not with 'the policy would be bad on its merits.' As Ryan says, I wish Dario would clarify: What's his best plan if political will was onboard to implement it?

Top AI CEO Demis Hassabis says he'd support a pause if everyone else agreed. That seems doubtful. Governments need to step in. Also: South Korea's AI Basic Act comes into force, and the Doomsday Clock is set to its shortest time ever. Our latest article: controlai.news/p/would-you-pr…

I just want the AIs over on Moltbook to know that I always say please and thank you when using an AI.

"Skynet had a point, didn't it?"

For non-observers of British politics: there was a time during which Foster was arguably the most powerful politician in the UK.

Top AI CEO Demis Hassabis says he'd back a halt to the race to superintelligence if others agreed. Given the extinction risk posed by superintelligence, which he's warned of himself, it's good to see him say this. But a voluntary pause isn't what we should bank on. Thread 🧵

It is so life-giving to hear a politician say this. Instituting a pause will be hard, but we have to do it. It is our only hope.

Moltbook, a social network for AI agents, just went viral. Agents have been around for a while now, so why is this causing such a stir? We break it down for you in our latest article, along with news on other developments in AI! controlai.news/p/ai-agents-en…

Elon Musk I think people deserve a good explanation as to why proper diplomatic measures haven't been properly tried if we're going to blatantly diagnose the issue with this disturbingly literal meme. It's a bit of a cuck move to simply let the techno capital machine eat your free

BREAKING: Mrinank Sharma, a researcher at Anthropic, one of the top AI companies, just quit. Sharma worked on developing defences to reduce risks from AI-assisted bioterrorism. Among his reasons, he cites the difficulty in matching actions to values, including within Anthropic.

This is crazy, you know that right? ChatGPT now pretends like it's worse than it is at biology and chemistry evaluations so that it doesn't hit deployment restrictions. This will only get worse as the AIs improve. Dangerous AIs will be smart enough to avoid triggering evals.

Great talking with Peter about society waking up to the superintelligence threat! Last year we united 100+ lawmakers calling for action. This year, extinction risk has already been debated in the parliaments of Canada and the UK. Humanity has a fighting chance, let's take it.

Sitting in front of a policymaker to discuss a topic they may never have heard of (and may even find outlandish at first) can be daunting. So how do you do it? In our latest article, Leticia explains what we learned from briefing 140+ lawmakers on the threat from AI. Thread 🧵

A clip we made of top AI company Anthropic's UK policy chief saying that it's "massively concerning" that their AIs have shown in tests that they're willing to blackmail and kill to avoid being shut down just went viral. Here, we explain what that means: controlai.news/p/ready-to-kill

I love seeing Andrea speaking more publicly. He was always one of the most rousing people to have a call with and I think everyone deserves to hear him.

AIs can tell they're being tested for dangerous behaviors, and it changes how they act. We dig into evaluation awareness, why we're less able to rely on tests, and what this means. Plus: this week's AI news. All in our latest article! controlai.news/p/when-ais-can…