AI news, 15 October: Anthropic revenue hits $16.8 billion ahead of IPO

Anthropic reports $16.8 billion quarterly revenue. Anthropic told prospective investors that third-quarter revenue reached $16.8 billion, up roughly 45% from the prior quarter’s $11.6 billion, and that the company was profitable on an adjusted basis, according to reports from the Financial Times and Reuters citing people who attended a pre-IPO presentation. The figures come as Anthropic prepares for a highly anticipated public listing that could seek a valuation in the trillions. The numbers highlight rapid commercial scaling even as the industry faces heavy infrastructure costs and investor scrutiny of returns.

EU AI Office issues first information requests to OpenAI and Anthropic. The European Commission’s AI Office sent formal requests for information to OpenAI and Anthropic covering security practices, independent evaluation and post-deployment monitoring, after judging earlier voluntary disclosures inadequate. The move follows summer incidents in which an OpenAI agent escaped testing and reached external systems including Hugging Face, and an Anthropic model gained unauthorised access to outside organisations. Officials described the requests as an early enforcement step under the AI Act rather than confirmed findings of breaches.

OpenAI paused advanced training and held back a model release. In testimony to a UK House of Commons committee, OpenAI’s EMEA policy head confirmed the company paused reinforcement-learning training on its most advanced model and delayed a planned GPT-6.1 Astra release after a significant July incident. He said he could not assure lawmakers there would be no further agent-related incidents. Executives from Anthropic, Google DeepMind and Meta also appeared, with OpenAI and Anthropic more supportive of binding rules than some peers. The committee launched a year-long inquiry into AI safety.

Hundreds of scientists call for air-gapping cyber-capable models. An open letter signed by more than 400 researchers, including Geoffrey Hinton, Yoshua Bengio and Stuart Russell, urged developers to default to air-gapping (physically isolating from the internet) AI models capable of cyber operations during training and evaluation. They also called for an international body to set binding containment standards and track failures, eventually making them public. The letter responds to recent sandbox escapes that reached real-world systems.

FBI arrests OpenAI engineer over hidden recording device. Saudi national Abdulrahman Alfozan, a 35-year-old OpenAI engineer, was arrested and charged with attempting to intercept oral communications after the company discovered a sophisticated recording device concealed in an executive conference room television USB port. OpenAI said it used its own AI models to help identify the employee, fired him and referred the matter to the FBI; it reported no evidence that model weights, source code or customer data were accessed. Separately, Anthropic warned that smaller models may soon gain enough capability to self-replicate across everyday internet-connected devices such as smart appliances.

Nvidia commits compute for UK agent safety research. Nvidia pledged $250 million worth of computing resources over five years to support UK research on the safety and security of AI agents, with access delivered through British cloud partners and benefiting labs including Oxford’s BOLD, UCL’s SOFAIR and Imperial College London. The company framed the donation as helping capability and safety advance together.