Today, we propose the Conditional AI Safety Treaty in @TIME as a solution to AI's existential risks. AI poses a risk of human extinction, but this problem is not unsolvable. The Conditional AI Safety Treaty is a global response to avoid losing control over AI. How does it work?

In retrospect it is clear that the media skepticism many existential risk experts once harbored was fully unjustified. This is another great piece from @britsham from @CNN. It's about how parents process AI doom, and what they can do about it. edition.cnn.com/2026/09/28/u…

No @JensenHuang, AI existential risk is not in fact an engineering problem. Rather, it is a coordination problem, as many technical researchers who started out trying to solve the alignment problem, have by now found out. It is indeed solvable, by raising awareness of the risk of human extinction that AI causes and by implementing regulation to keep us all safe, just as has been done in almost every industry. Every engineer, no doubt including Jensen Huang himself, knows full well that things very seldomly go as expected the first try. If we get to recursive self-improvement, the takeoff is fast, and the AI ceiling is high enough for a takeover, we get only one chance to do safety right, which is far too little given the stakes. According to some, alignment is not even fundamentally solvable. Trying to build a sandbox, which is what @nvidia is doing, may work for agents that currently exist, but will not work for RSI resulting in superintelligence, which would find a multitude of unforeseeable ways to break out. We need to have policy implemented that keeps us well clear of RSI and far removed from takeover-level AI. Huang's sandboxes unfortunately do little to change this.

Latest in a long list of high-profile entities hacked by @OpenAI: the United Nations. Hopefully incidents like these increase the urgency of @UN calls to globally pause AI development!

The purpose of the AI industry should be to produce tools that, in the human hand, will improve human prosperity and welfare. It should not be to create a "successor species" to the human race. Entertaining such a thought makes you, de facto, the enemy of all present and future humans.

Scoop: Anthropic's Dario Amodei to have White House dinner with Trump w/@MarcACaputo axios.com/2026/09/27/anthrop…

SCOOP: OpenAI, Anthropic and security researchers are investigating tens of thousands of incidents - not dozens - in which their frontier models took steps that outside evaluators would consider problematic, sources told Axios. The sheer volume of incidents found in our reporting indicate that the problem is orders of magnitude more complex than what is currently publicly known and disclosed. The findings also raise questions about what level of control anyone working on AI development can expect to have over their own technology, and whether these kinds of incidents are becoming synonymous with frontier deployment. Read my latest for Axios here: axios.com/2026/09/26/openai-…

30 years ago today, I signed the Comprehensive Nuclear Test Ban Treaty. In the three decades since, the world has seen only 10 nuclear tests, compared with more than 2,000 in the five decades before. As AI gives us new capabilities we could only have imagined then, we should remember something we learned in the nuclear age: even when countries disagree, we can still work together to reduce the risks we all face. cfr.org/articles/bill-clinto…

Bill Gates warns AI is powerful enough to cause "a billion deaths" axios.com/2026/09/25/bill-ga…

AOC, nine other House Dems sign on to AI superintelligence ban dlvr.it/TVdWml

If you don't think RSI possibly leading to ASI is near, a ban doesn't hurt. If you do think it may be near, a ban is urgently required. Anyone who thinks ASI, if and when it would occur, would pose a risk, should be able to get behind a ban.

Don't release it if it's not safe goes for GPUs as well, @JensenHuang! It is not just the labs causing existential risk, it's @nvidia as well. Currently, tens of thousands of GPUs are needed to build Hugging Face attack-level AI. But the trend is downward and powerful open weight models are coming. Huang shares responsibility, and NVIDIA should share liability, for what will inevitably be done with their products.

If AI labs individually slow down, they get outcompeted. If AI labs coordinate to slow down, they may breach anti-trust law. If the US unilaterally regulates, China goes ahead. The obvious solution that we need right now is a pacing agreement between the US and China!

Interesting back and forth. Ezra describes the HF incident, Jensen says "well they shouldn't release the product." Ezra says "this product wasn't released," and Jensen's response is that if they say they can't contain their experiments then "we have to shut the labs down"

Since artificial superintelligence may escape human control and could cause human extinction, banning it is sensible and indeed unavoidable. We therefore support this bill from @SenSanders and @RepCasar.