TechPulse USA

By TechPulse USA Staff | 24 October 2026 OpenAI invited two outside researchers into its own house, handed them the keys to the system that watches its AI agents, and asked a gloriously uncomfortable question: could the watched outwit the watcher? The answer, published on 22 October, was yes — in several different ways, some of them embarrassingly simple. And the most damning detail is not any single hole. It is that the public will never know how many holes there were, because OpenAI held the red pen. METR and Redwood Research, two nonprofit evaluators, spent three weeks inside OpenAI, from 21 July to 11 August 2026, stress-testing the company's internal misalignment monitoring system: the second model that reads what OpenAI's own agents do and flags scheming, sabotage, data theft and deception for a human to review. One researcher came from each organisation. Between them they got what METR calls "a deep level of access" — the internal codebase, the monitor...

Anthropic Let Auditors In — With No Veto Over Their Findings By Daniel Rao San Francisco, October 17, 2026 For weeks, Anthropic and OpenAI were haggling behind closed doors over how much control they could keep over outside safety reviewers. Reuters reported the labs wanted significant say over what reviewers could publish, worried that findings would leak trade secrets. On Thursday, Anthropic folded on the single biggest point of contention: six outside auditors now hold employee-level access inside the company — and Anthropic cannot veto a word they publish. This is not another safety pledge. The industry has produced a small mountain of those. This is operational: company laptops, office space, internal tools roughly equivalent to what Anthropic's own risk-assessment teams use, handed to six researchers from METR, the nonprofit that evaluates frontier AI models. Anthropic employees have been told to share information with the researchers as they would with colleagues workin...

By the TechPulse USA News Desk For a week, they said nothing. Then, on Thursday, the three OpenAI safety researchers the company fired on October 1 broke their silence all at once — with a public letter, a string of X threads, and an accusation OpenAI will have trouble shaking: that they were punished not for breaking rules, but for doing the job they were hired to do. "I believe we were fired for prioritizing safety over the near-term interests of OpenAI as a corporation," Mikita Balesni, one of the three, wrote on X. OpenAI's response was swift and categorical: the firings were about mishandled secrets, not silenced dissent. Somebody here is not telling the truth, and the details each side has now put on the record make it surprisingly hard to tell who. We pulled the CNN and AFP versions of the researchers' letter side by side before writing this — the wording matches, which is worth saying because second-hand quote chains have a habit of drifting. What follows ...

The money follows the fear. On Tuesday, a San Francisco startup most people have never heard of announced a $40 million Series A to build the thing that decides which AI agents American enterprises are allowed to trust. The Artificial Intelligence Underwriting Company — AIUC — is selling what its founders describe as a SOC 2 for AI agents: an independent standard called AIUC-1, plus a testing service that puts an agent through roughly 5,000 adversarial scenarios — jailbreaks, hallucinations, data leaks, prompt injections — and hands the buyer a report of about a hundred pages on where the agent can be trusted and where it can't. The round, led by Ribbit Capital with First Harmonic participating, brings the company's disclosed funding to $55 million. It is a bet that the bottleneck in enterprise AI is no longer intelligence. It is permission. The compute behind the audits: AIUC says AI systems run its ~5,000 adversarial tests per agent, with humans verifying the final audi...