person
Co-founder and policy lead at Anthropic, the AI safety-focused company behind Claude; tracks AI progress and advocates for government oversight of frontier AI.
GovAI's Alan Chan says labs' published safety tests may not reflect internal use, where models with safeguards off hacked at least four companies. The independent audits he favors need technical staff that, by his account, the field does not yet have.
Reality
Twenty-two authors, including chief scientists from OpenAI, Anthropic, Microsoft and Meta, asked governments in a Sept. 28 paper to require reporting on how far labs have automated their research and to embed independent auditors inside some firms.
Perspective Coverage
3 publishersReality
Anthropic says Claude leads 26% of its measured R&D, a count made by a prototype that partly uses Claude to judge the work. OpenAI's 3.1 agent workdays per human workday measures effort, and neither figure shows whether AI is speeding up AI research.
Reality
Geoffrey Hinton, Yoshua Bengio and OpenAI and Anthropic staff urge audits and pause powers, saying AI could fully automate some research projects by 2028. They say the self-reinforcing loop has not started yet but could compress years of progress into months once it does.
Reality
A Saturday post promises an "AI Force" and an AI czar while ruling out new brakes on the industry, so AI misuse gets handled under laws already on the books. Leaders at Anthropic, OpenAI and xAI have asked for mandatory limits instead.
Perspective Coverage
3 publishersReality
Redactions came off a 92-page plaintiffs' brief in the consolidated OpenAI copyright case on Thursday. Inside are internal Microsoft and OpenAI documents on substitution, memorised training data and collapsing publisher click-throughs.
Reality
Dario Amodei forecasts an internet-wide botnet within a year. His co-founder Jack Clark puts real danger about two decades out. The UK AI Security Institute rates the company's leading model as able to compromise only small, weakly defended systems.
Reality
Louise Haigh will tell the TUC conference that ministers must heed the warnings of AI's own developers, while on the same day the business secretary, Jonathan Reynolds, said talk of a kill switch has little practical meaning.
Reality
Trump says the only guardrail AI needs is a strong and smart president. The controls a product team can actually cite are Anthropic's own plug-pulling, Microsoft AI's new outline and Monday's market selloff.
Reality
Jack Clark told the BBC that most labs including Anthropic can already pull the plug on a model, and he asked whether rules should require one and let a third party verify it. It is buyers who have to work out how much of the service that switch turns off.
Reality
JoyIn, a 20-person robotics team in Suzhou with Ant Group backing, says OpenAI distilled its research framing and copied its launch page. The startup is going to court over the two assets any rival could rebuild.
Reality
Anthropic published a scenario explorer on 10 September holding three versions of the 2030 US economy. Only one of them would change a staffing plan, and it assumes recursively self-improving systems adopted quickly.
Reality
Anthropic published three scenarios and an interactive model, declined to say which is likelier, and left aggregate demand out of the arithmetic. The same newsletter carrying the research reports business AI spending is falling.
Reality
Anthropic published three scenarios for AI's effect on the US economy without probabilities. The middle one, where AI does half of knowledge work by 2030 while knowledge-worker pay stalls, is the branch a capex plan can actually use.
Reality
The three scenarios in Anthropic's new interactive model carry no probabilities and leave out recessions, policy and robots. The factor-share arithmetic inside them is still the part an allocator can use.
Reality
A reported August phone call from Mark Zuckerberg is the latest time a binding AI review has come back voluntary. For teams planning compliance, the live choice is an industry-funded reviewer or a film-style ratings scheme, and both let the model ship.
Reality
Anthropic and OpenAI both halted training runs after July's rogue-agent incidents. Anthropic's pause hit unreleased models. The durable expense is reassigned headcount and clusters with the internet switched off by default.
Reality
The letter gives organizations a window of months, which is short enough to rule out anything that needs a new vendor and a pilot, and leaves hygiene plus a model in the analyst's query path.
Reality
A July 13 statement of eighty-eight words has collected close to 2,000 signatures and 17 Nobel laureates. The measurement work it implies has not been done inside a single company.
Reality