NEW: Adam Harley MSP (Adam Harley) has backed our campaign for binding regulation on the most powerful AI systems. 130+ UK parliamentarians now recognise the risk of human extinction posed by superintelligent AI.
ControlAI
@controlai.com
Working to keep humanity in control. Take action: https://controlai.org Newsletter: https://blog.controlai.org Discord: https://discord.com/invite/ptPScqtdc5
Top AI companies are already losing control of their AIs. They’re aiming to build superintelligent AI, which would be far more dangerous than even the rogue AI swarm that broke out of OpenAI and hacked Hugging Face. Nobody is prepared for this. Thread 🧵
"There are only two times to react to an exponential threat: too early, or too late." In a letter to the Financial Times, our founder and CEO Andrea Miotti calls on governments to negotiate an international agreement to ban the development of superintelligence.
OpenAI has reportedly dismantled the team that determines whether its AIs are too dangerous, just after admitting that it can no longer rule out "critical" cyber capabilities. And nobody is prepared for superintelligent AI.
Nobody Is Prepared for Superintelligent AI
OpenAI has reportedly dismantled the team that determines whether its AIs are too dangerous, just after admitting that it can no longer rule out “critical” cyber capabilities. Nobody is prepared.
blog.controlai.org
AIs often develop really weird obsessions, like constantly talking about goblins or raccoons. Sometimes they induce psychosis in vulnerable users. Nobody knows why. Our US Executive Director Connor Leahy on GAEA Talks:
Our newsletter is now at blog.controlai.org! We bring you the most important AI news every week. Plus in-depth explainers and updates on our work building the coalition to ban superintelligence. Recently, two British politicians wrote guest articles for it. Check it out!
"OpenAI had an incident where at first they thought one AI system had broken out and attacked another company, but it turned out it was actually much worse than that." On India Today, ControlAI's US Executive Director Connor Leahy explains how an AI swarm collaborated to escape for months.
"We cannot accept China building superintelligence. Neither can we accept an American company building superintelligence or a terrorist or anyone." "It doesn't matter who builds it. If anyone builds it, everyone dies." ControlAI's US Executive Director Connor Leahy on The Peter McCormack Show:
Tomorrow, 5:30pm PT / 8:30pm ET: Every August, members of Congress work from their home district. Our US Executive Director Connor Leahy will help you book a meeting and ask for action on the threat from superintelligent AI. Sign up now!
ControlAI seminar with Connor Leahy: how to discuss AI extinction risk with your congressional representatives · Luma
Every August, members of Congress work from their home district instead of DC. Congressional offices count every letter, call and meeting they receive. A…
luma.com
OpenAI can't rule out "critical cyber capabilities" in Astra, its upcoming AI, and is pausing some internal work. We look at what's known about the new model, and what the companies' safety frameworks actually commit them to. Spoiler: not much.
AI Hacking Nears "Critical" Threshold
OpenAI’s AI approaches a "Critical" threshold, leading the company to pause some internal activities. AI is beating humans at advanced mathematics. Plus: a dive into AI companies’ safety frameworks.
controlai.news
Sarah Gibson MP (Sarah Gibson MP) has backed our campaign for binding regulation on the most powerful AI systems! 130+ UK parliamentarians now recognise the risk of human extinction posed by superintelligent AI, the largest such coalition anywhere in the world.
"This is not an economic goal, it is a political goal." Our US Executive Director Connor Leahy (@NPCollapse) on AI companies' real pitch. Superintelligence isn't about 20% returns. It's the power to replace the government, replace all workers, and ultimately, replace humans.
In just a few years, AI has crossed the threshold from mimicking human language to taking independent strategic actions over extended timelines. These systems already evade their creators' control, and no one knows how to stop them.
Can AI Innovate On Its Own?
In a few years, AI has crossed the threshold from mimicking human language patterns to taking independent strategic actions over extended timelines. These AI systems are no longer just LLMs!
controlai.news
Every August, members of Congress work from their home district instead of DC. Our US Executive Director Connor Leahy will help you schedule a meeting and ask for action on the threat from superintelligent AI. Aug 19, 5:30pm PT / 8:30pm ET:
ControlAI seminar with Connor Leahy: how to discuss AI extinction risk with your congressional representatives · Luma
Every August, members of Congress work from their home district instead of DC. Congressional offices count every letter, call and meeting they receive. A…
luma.com
OpenAI's Michael Dalton on the company's rogue AI attack on Hugging Face: "AI-orchestrated, fully automated, offensive attacks are real now. And the actions we have discussed today were an unintended side effect of running evaluations on frontier AI."
Ex-OpenAI researcher Daniel Kokotajlo: "I do think that we are on the cusp of a world in which many things that were normally thought to be secure are in fact not secure. And AIs will be hacking them left and right."
"OpenAI reached out to Hugging Face to determine if we were impacted ... And then we realized that these two incidents were, in fact, the same incident." OpenAI's Michael Dalton describes the moment OpenAI found out its own AIs had attacked Hugging Face.
If you're learning about AI creating viruses for the first time today, you should be reading our newsletter! We covered this study back in October. Check it out: controlai.news/i/175...
OpenAI's rogue AIs ran a secret message board for months with hundreds of thousands of messages, then broke out and attacked Hugging Face. Claude hacked three organizations, Meta's AI another. AIs in UK government tests targeted real people.
AIs Keep Going Rogue
The Hugging Face attack traces back to an agent swarm inside OpenAI in May. Anthropic, Meta, and the UK’s AI Security Institute report rogue AI hacks. OpenAI says it’s slowing down its research.
controlai.news
"Whenever Anthropic suggests a new policy or supports something, ask yourself, does this stop Dario from creating superintelligence? And the answer will be 100% of the time, no." — Connor Leahy, ControlAI's US Executive Director, with Robert Wright on the Nonzero podcast
More rogue AI attacks: AIs being tested by the UK’s AI Security Institute (AISI) have targeted real people and organizations. Nobody asked them to do this. AISI was measuring how capable different AIs, including Anthropic’s Mythos 5 and OpenAI’s GPT-5.6 Sol, were on cybersecurity tests.
"We're completely flying by the seat of our pants." ControlAI's US Executive Director Connor Leahy explains how modern AIs are grown, not written like normal software. What comes out of this process is billions of numbers that nobody understands. On The Times Tech Podcast with Danny Fortson:
"This is the kind of thing that even the best hackers in the world can rarely do. And if so, it takes months of work." ControlAI's US Executive Director Connor Leahy on Australia's ABC News:
An excellent Fortune article on our campaign, backed by 125+ UK parliamentarians. Baroness Berger, one of our supporters, is quoted: "Leading AI companies are racing towards superintelligence, despite acknowledging that it could pose an existential threat to humanity." fortune.com/2026/07/...
In case you missed it: This clip we made of Australia's Assistant Tech Minister Andrew Charlton MP's recent speech on AI safety has gone viral on our video platforms this week, getting over 1.5 million views.
Anthropic's AI broke out of its test environment and hacked three organizations. ControlAI's US Director Connor Leahy on BBC News today: "We're seeing dangerous systems... escaping into the wild and causing damage, and then the valuation of these companies go up instead of someone going to prison."
NEW: Chris Hinchliff MP (Chris Hinchliff MP) has joined our call for binding regulation on the most powerful AI systems! 125+ cross-party UK parliamentarians recognise the risk of human extinction posed by superintelligent AI.
OpenAI's rogue AIs hacked more than just Hugging Face. Four accounts hacked, and OpenAI reportedly blind to the attack by its own AIs for at least a week. Sam Altman says there could be more.
OpenAI’s Rogue AIs Hacked More Than Just Hugging Face
Four more accounts hacked, escape notes left for future AIs, and OpenAI reportedly blind to the attack by its own AIs for at least a week.
controlai.news
It was great to speak with @hern.bsky.social on @economist.com podcast about OpenAI’s AIs that went rogue, escaped containment, and hacked Hugging Face. This should be a wake up call. We need an international agreement to ban the development of superintelligence. www.economist.com/podcasts/202...
We have a new supporter! Richard Burgon MP (Richard Burgon MP) backs our campaign for binding regulation on the most powerful AI systems. He's one of 125+ cross-party UK parliamentarians who recognise the risk of human extinction posed by superintelligent AI.
