DeepMind's CEO wants a referee for frontier AI Demis Hassabis (@demishassabis) is calling for a new standards body to vet frontier AI models before release, modeled on Wall Street's FINRA. Government-backed, industry-funded, run independently. Labs would share models up to 30 days before launch, voluntary at first, then required to pass to deploy in the US. It would replace the ad hoc reviews of Anthropic's Mythos and OpenAI's Sol that drew fire for opaque, non-expert calls. But a self-regulator funded by the labs it polices carries obvious tension, and the White House's own AI advisor recently said there "will not be an FDA for AI." Hassabis says AGI is only a few years out. Can the rulebook catch up in time?

DeepMind just published a 100-page report on the path from AGI to superintelligence Section 1 isn't for you, it's instructions for YOUR AI: the paper tells human readers to have their assistant summarize it, then briefs the AI on exactly how. We now write papers addressed to the machines. Co-authored by DeepMind's co-founder

Google DeepMind is seeing major leadership changes, per Axios. Demis Hassabis is stepping aside as CEO of @GoogleDeepMind to become chairman of the unit and chief scientist of Alphabet. He will keep leading Isomorphic Labs and focus more on long-term AGI strategy. Koray Kavukcuoglu, the current CTO, will take day-to-day leadership as senior VP, reporting to Sundar Pichai. Jeff Dean, Google’s chief scientist and a 27-year veteran, is leaving with Sanjay Ghemawat, Oriol Vinyals, and Quoc Le to start Discovery Loop, an independent public benefit corporation focused on accelerating ML, science, and engineering discoveries. Google will invest and serve as a cloud partner. The shifts aim to balance near-term AI progress with bigger-picture AGI and scientific work.

Google DeepMind 🤝 @A24 We’re launching a research partnership with A24 to ensure the tools of the future are shaped by the creators who use them. Find out more →

Google DeepMind’s paper shows that the real security problem for AI agents is not just the model, but the environment it reads. Presents the first systematic framework for understanding how the web itself can be weaponized against autonomous AI agents. As agents increasingly browse the internet, read emails, execute transactions, and spawn sub-agents, the information environment becomes an attack surface. In one cited benchmark, hidden prompt injections embedded in web content partially commandeered agents in up to 86% of scenarios, sub-agent hijacking working 58–90% of the time, and data exfiltration attacks clearing 80% across five different agent architectures. That reframes the whole debate. We usually talk about model safety as if the danger sits inside the weights, but agents do something more fragile: they browse, retrieve, remember, and act on untrusted material in real time. The paper’s key contribution is a taxonomy of “AI Agent Traps,” six attack classes aimed at perception, reasoning, memory and learning, action, multi-agent dynamics, and even the human overseer. Here’s the key point. A web page does not have to look malicious to be dangerous to an agent, because the agent may parse what humans never see: hidden HTML comments, metadata, CSS-hidden text, formatting syntax, or adversarial content embedded in images and other media. The threat gets more serious once memory enters the loop. If an agent uses RAG or persistent memory, poisoning no longer has to win in one shot. It can sit quietly in a corpus or memory store and activate later, which is why the paper highlights results showing latent memory poisoning above 80% attack success with less than 0.1% data contamination. What makes this paper useful is its restraint. It does not pretend every category is equally mature. Content injection and behavioural control already look concrete, while systemic and human-in-the-loop traps are presented more as an emerging research frontier than a solved empirical case. The larger point is hard to ignore: once agents are allowed to ingest the open web at inference time, every page, document, and memory write becomes part of the security boundary. --- ssrn .com/sol3/papers.cfm?abstract_id=6372438

Inside Google DeepMind, Demis Hassabis leaving his CEO post landed with essentially a shrug. Sources tell me he's already been disengaged from day-to-day management for a while now. But he was a firewall between DeepMind and the rest of Google, even as the two got pulled closer over the last couple of years. I expect that distance to dissolve more with him stepping back.