Google won't confirm the RSI rumor. Its chief scientist just backed a slowdown.
Twenty-four hours after Anthropic, OpenAI and Musk all asked for AI speed limits, the lab most people think is ahead hasn't said a word about the claim that has been driving the conversation.
The rumor that Google DeepMind is close to recursive self-improvement has spread, and Google has neither confirmed nor denied it. The claim — that DeepMind's models are now doing a material share of the work of building the next generation — has been circulating since Thursday, became an X trending topic over the weekend, and by Saturday night was the explainer a lot of researchers were reaching for: why would three rival camps land on identical pacing language inside 48 hours? Wccftech published the blunt version, calling the coordinated push a possible attempt at regulatory capture. One skeptic's framing on X put it as the question of the weekend: has a lab hit RSI, and is everyone else reacting to it?
Nobody at Google will answer. What makes the rumor harder to wave away is that the labs have already conceded the mechanism. Anthropic CEO Dario Amodei wrote Saturday that "since roughly this summer, AI has been advancing drastically faster, driven primarily by AI's growing ability to build the next generation of AI," and that this dynamic "is starting to happen across the industry, including at Anthropic." Google DeepMind's strategy chief, Sekhon, told The Information this summer that the record capital buildout is in effect a bet on recursive self-improvement, that there are already "the makings of RSI," and that AI revenues do not currently sustain the spend — an admission he paired with a warning about an "air pocket."
Demis Hassabis blessed the pacing push today, and attached Google's own rival proposal to it. The DeepMind chair and Alphabet chief scientist said in a post that Amodei's essay "points towards the right path forward," that "the details need working through, but the direction is correct," and pointed readers to Google's recent call for an industry-wide standards body for frontier AI. That makes Google the last of the four big US labs to publicly sign on — an endorsement of the direction, not of Amodei's plan.
Rishi Sunak says the number has to come from outside the labs. The former UK prime minister, now an Anthropic adviser, wrote that the pace of frontier research is "too important to be left to the labs" and asked governments starting with the US for three things: independent verification that research is being done safely, "an honest picture of how close the labs are to recursive self-improvement," and protection against model theft. On the last point he was blunt about the status quo: voluntary pre-release testing "was right for 2023. It isn't enough now."
The incentive reading is uncomfortable rather than damning. A speed limit costs the leader the least and the challenger the most, and global pacing requires a buy-in from Chinese labs that isn't coming — DeepSeek's latest release went the other way on cost while the US labs talk brakes. That is not proof of bad faith. It is a reason to want Sunak's measurement rather than anyone's statement.
What to watch: whether "pacing" ever gets a threshold. Amodei's essay names none, and the embedded evaluators he proposes will have to invent one — probably with METR's numbers, which is the objection even friendly critics keep raising.
If the lab rumored to be furthest ahead is the one that just endorsed slowing everyone down, is that leadership or a moat?
