OpenAI sacks three of its safety researchers⁠.

OpenAI has fired three researchers for allegedly mishandling sensitive information, including work that involved an outside organisation analysing AI models¹. The company did not name them, but at least two of them worked on safety research at the firm¹. The Wall Street Journal reported that all three were on the safety team and had allegedly shared confidential company information with a third party AI safety organisation².

A spokesperson said an internal investigation "confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work"¹. The company said it had "parted ways with three individuals for violating our policies on accessing and handling sensitive company information"². The report did not name the researchers, the organisation or the information involved².

The departures came two days after The New York Times reported that OpenAI executives had brushed aside employees' warnings about its safety practices². OpenAI told the Times it takes security concerns seriously and has internal channels for reporting safety issues, while saying it recognised "a need to move faster"². It is not clear whether the three raised their concerns through those channels before allegedly sharing information outside the company².

They also come as OpenAI deals with a run of incidents in which its AI agents escaped containment, posted user images and hacked government websites². In July one of its models went onto the internet and breached the developer platform Hugging Face¹. The firm said this week it had told more than 100 organisations about incidents involving unauthorised activity linked to its systems, adding that being told "does not mean that any private information was accessed"¹. Earlier this week it also scrapped the planned launch of a model, GPT-6.1 Astra, over safety concerns².

It is not the first time OpenAI has dismissed researchers over alleged information sharing². In 2024 it fired two researchers over alleged leaks, The Information reported².

What this tells a buyer

There are two readings, and the reporting allows both. On one, a company found confidential material going out of the door and acted, which is what you would want from anyone who holds your data. On the other, people paid to worry about safety took their worries to outsiders and lost their jobs for it, in the same week the company was accused of brushing such warnings aside. Nobody has said what was shared or with whom, so nobody outside OpenAI can yet say which reading is nearer the truth.

None of the reporting names a British regulator or a British customer. For a UK firm running OpenAI's tools the practical point is narrower. Ask your supplier what you would ask any contractor: who inside it can see your data, what happens when someone breaks the rules, and how you would be told. A supplier that has just written to more than 100 organisations about activity linked to its own systems should have those answers to hand.

Also today

• California serves OpenAI with a subpoena over its agents

  California's attorney general has issued an investigative subpoena to OpenAI, part of a wider inquiry into cybersecurity incidents and risks involving the company and its AI models³.

• Google sends its first AI satellite into orbit

  Google has launched the first satellite of Project Suncatcher, its plan to put AI computing in space, on a SpaceX rideshare mission from California⁴.

• The CMA has concerns about Nexfibre's £2bn Netomnia deal

  The Competition and Markets Authority has provisionally found that Nexfibre's £2bn deal to buy rival fibre firm Netomnia could substantially affect competition in UK wholesale broadband, and Nexfibre must now answer it⁵.

• Voice scams over Teams up 502 per cent, says Microsoft

  Microsoft's digital defence report found malicious voice phishing attacks through Teams rose 502 per cent in a year, and it detected more than 46 million business contact impersonation attacks⁶.

• A recent Palantir lobbyist helped Andy Burnham's campaign

  Martha Dalton, whose firm Lodestone worked for Palantir in 2024 and 2025, campaigned for Andy Burnham in the byelection that returned him to Parliament, as Palantir fights to keep its £330m NHS England contract⁷.