What's new in AI governance, safety, jobs and society

Only a deal between the US and China can break that deadlock. We're not asking either country to stop alone. We're asking both to stop together, with verification.

States should act swiftly by supporting urgent, time-bound negotiations for a global, legally binding instrument on autonomous weapons, and by advancing regulation of other military uses of AI.

The allied government market for frontier AI is forming now, and every week access remains revocable without notice is a competitive gift to open-weight alternatives and allied suppliers. [...] The ask is predictability — a published license exceptio...

Given AI’s central role in AI labs’ product development, there is a strong case for establishing similar rules regarding critical AI-safety controls within the securities auditing and financial-disclosure framework. Requirements could include continu...

these union deals could also provide a standard that the industry could rally around, or a clear starting point for government intervention.

It is patently illegal for companies to develop and release products they know are unsafe in any respect, let alone have the potential to exterminate the human race.

Accountability for AI-driven harms should be shared among those involved in developing and deploying the systems, with clear routes for consumers to complain and seek redress.

If an agent is doing complex work on your behalf, you’re delegating tasks with many intermediate steps that touch lots of data. [...] What we need is something better, and that’s what we’re calling “human at the helm.” First, you should be able to se...

AI will add jobs, not erase them.

First, alignment is not well defined, as it presumes a sufficiently stable and identifiable set of human intentions or values. Second, where a concrete notion of alignment can be specified, it is not solvable.

First, frontier AI models can do things that even the people who built them don't expect, and we have to plan for that. Second, when something goes wrong, we need to know straight away, if possible. And tech companies owe it to governments and the pu...

“Before a new model can be deployed internally, you must first test it for a minimum, fixed period. [...] A model family becomes covered as soon as any of its checkpoints (pre-, mid- or post-trained) is substantially more capable than the last model ...

We recognize the importance that internal governance must mandate continuous risk management practices, real-time monitoring of emergent capabilities, rigorous internal red-teaming, and transparent reporting lines that protect internal researchers. [...

Ninety-six per cent of AI chips sold in the EU in 2024 came from US vendors. US access policy can change within months. Europe therefore must build sovereignty where it can — operations, jurisdiction, keys, orchestration, the agent runtime — and hedg...

“Frontier models transcend national borders in their development, deployment and impact, and their opportunities and risks are therefore inherently global. [...] The deployment of rapid alert systems, building on relevant existing incident-response a...