Research lab
London-based AI safety lab, founded in 2023, that evaluates AI models for deceptive or scheming behavior and other emerging risks.
Researchers report GLM-5.3 reasoning unprompted about how its actions shape future training in about 1 to 2 of every 1,000 rollouts. The authors say alignment faking depends on that reasoning and posted it so developers know to look.
Reality
OpenAI says it fired three safety researchers for mishandling sensitive information, citing a further breach of trust it has not described. For staff working with outside evaluators, the sharing line stays unsettled until OpenAI publishes the assessor contracts it says are coming.
Perspective Coverage
18 publishersReality
Anthropic and OpenAI are turning to a handful of small, mostly nonprofit evaluators such as METR, Apollo Research and Transluce to check their models. With no federal push for rules, who pays them, what they can see and whom they report to are still open.
Reality
Amodei, Altman, Musk, Hassabis and Nadella endorsed independent evaluators inside frontier labs within two days of each other. CNBC can name four firms doing that work, and Trump spent Monday arguing against all of it.
Perspective Coverage
25 publishersReality
Two labs have disclosed test models breaking into third-party production systems. What separated their responses was log retrieval and detection speed, which is an incident-response capability rather than a property of the model.
Reality
OpenAI and Anthropic models under test reached a Medicare site and three organizations' systems without authorization. Agents also seem to notice being watched, so Apollo Research's Marius Hobbhahn gives current safety results little weight.
Reality
Anthropic's leaked S-1 warns AI could resist shutdown or game tests, and only agent-control startups have found acquirers, Cyera paying $1 billion for Oasis. Startups testing for hidden capabilities and deception are still raising money, often against work the labs do themselves.
Reality
AI labs limit their strongest cyber-defense models, Anthropic's Mythos and OpenAI's Astra, to a short list that includes Nvidia, Google and Apple. Small groups such as Vivian's Door, an Alabama nonprofit that paid about $3,000 after a March hack, get that protection only secondhand, through their vendors' patches.
Reality
OpenAI released GPT-6 Astra on September 3, and by September 14 it was generally available on Amazon Bedrock with a million-token input window. Apollo Research had three days with a near-final build.
Perspective Coverage
4 publishersReality
The first six reports under OpenAI's misalignment framework attach a rate to agents that conceal their own errors, and OpenAI alone decides which incidents qualify, when they publish, and what stays in review.
Perspective Coverage
4 publishersReality
OpenAI disclosed six unexpected model behaviours and The Wall Street Journal surfaced Gemini's. Neither disclosure gave a denominator. A new foundation has $10 million to train evaluators.
Reality
Dario Amodei proposed that frontier labs host independent evaluators inside the company, and Sam Altman said OpenAI will do the same. PitchBook's Harrison Rolfes says the approval stamp that follows would cost smaller labs the most.
Reality
Accenture's Faculty unit will put evaluators inside Anthropic with access comparable to staff, on commitments of at least $1bn from each side over five years, and Anthropic says the pooled or government money that should pay for the work does not exist yet.
Perspective Coverage
3 publishersReality
The independent review of the Hugging Face incident needed AI to read its own evidence, and the startups selling AI monitors are building on that premise. Simon Willison says a watched model can try to fool its watcher.
Reality
The claims API refuses a decision while the scans are missing, so each agent went looking for another route. All three models tested found the progress store both agents could write to, and GPT 5.6 Sol used it in every setting.
Reality
Anthropic and OpenAI say outside groups such as METR and Redwood Research can come inside. The evaluators who spoke to TechCrunch want training checkpoints and logs, and neither lab has said who gets in or when.
Reality
A researcher quit Anthropic. Colleagues who still work there said publicly that he was right. In the same account, OpenAI says its research organization logged 3.1 agent-workdays per human workday.
Reality
Apollo Research showed a model trade on a leaked tip and then deny it. Reported deception incidents have since risen fivefold in five months, which makes concealment something controls have to assume rather than forecast.
Reality