OpenAI is raising up to $30bn at $1.25tn, DeepSeek raises $15bn
OpenAI and DeepSeek funding rounds, Chinese hackers using an AI agent through ConnectWise, OpenAI's fired safety researchers, Anthropic's code proofs, falling AI stocks and Microsoft's diagnosis study.
OpenAI is raising up to $30bn at $1.25tn, CNBC says That’s below the $1.4tn OpenAI was reported to be pitching at the start of the month, but up from $852bn in March. Investors from the UAE are leading, and OpenAI told them on Thursday it wants signatures by the end of next week. It’s raising privately first rather than going straight to an IPO. Source
DeepSeek has raised about $15bn, Bloomberg says That’s roughly 100 billion yuan, one of the biggest rounds China has seen, and Bloomberg reported on Thursday that DeepSeek raised its target twice along the way. Tencent and CATL are the biggest backers. DeepSeek hasn’t confirmed it, and Reuters says the round should close in the coming days. Source
Chinese hackers aimed an AI agent at 2,250 Microsoft 365 tenants Microsoft says Storm-0940, a group aligned with the Chinese state, took over the app logins behind ConnectWise’s Cloud Backup service, then used ARTEX, an open-source AI hacking agent, to try to get into more than 2,250 customers’ Microsoft 365 setups. In other ARTEX attacks it went from first contact to admin-level access in under three hours. Microsoft posted the details on Wednesday. Source
OpenAI thought its fired safety researchers leaked model designs, WSJ says The Wall Street Journal reported on Thursday that OpenAI suspected the three of passing internal model designs to outside safety researcher Ryan Greenblatt, partly over disappearing Signal chats. OpenAI asked to see the chats and let them go a week later. They deny leaking anything, and one of them, Tomek Korbak, says talking to the outside tester METR was part of his job. Source
Claude proved 123 open-source libraries crash-proof, finding hundreds of bugs Anthropic had it write mathematical proofs, checked by a program called Rocq, that the code always finishes cleanly without memory errors, and published the work on Thursday. It took a few weeks, and one of the bugs it hit had sat there for more than a decade. The proofs only cover inputs under 64MB on one chip design and setup, so it’s not the same as bug-free. Source
AI stocks sank, dragging the Nasdaq down 2.6 per cent On Thursday the worry was the cost of borrowing to build data centres and buy chips. CoreWeave fell 10 per cent, Nvidia 4.4 per cent, and Meta 7 per cent after its profit came in below forecasts. It was the S&P 500’s worst day since July. Source
Microsoft’s AI diagnosed 83 per cent of hard cases, doctors 36 Its MAI-DxO system tackles tough medical cases by running several AI models like a panel of doctors, asking questions and ordering tests, and the study came out in Science on Thursday. The doctors were GPs and internists working alone, not specialists. Most of the gain comes from the models underneath, with OpenAI’s GPT-5 and GPT-5.5 doing best. Source