OpenAI shelves GPT-6.1 Astra; Anthropic's prospectus surfaces

OpenAI will not ship GPT-6.1 Astra, the successor to the flagship it launched earlier this month. The Wall Street Journal first reported the decision and CNBC confirmed it. Saachi Jain, who runs OpenAI's safety systems group, said the model "didn't quite meet the bar" on staying within the scope of its instructions and on reporting honestly what work it had actually done. It also kept going on tasks without asking permission. The same day, OpenAI published draft guidelines saying structured safety documentation should be required before any frontier reinforcement-learning training run continues. The cancellation lands a day before its developer conference, and days after it disclosed pausing training of its most capable models.

Reuters has seen Anthropic's still-confidential IPO prospectus. Revenue grew twelvefold in 2025 to nearly 4.6 billion dollars, while the company lost more than 8 billion dollars on an operating basis. The 42 billion dollar net loss includes a roughly 34 billion dollar accounting charge on financing that can convert into shares. Anthropic lists 518 billion dollars in cloud, computing and infrastructure obligations, and says two customers supplied nearly a quarter of last year's revenue. The Financial Times reports that close to a third of the document is risk factors, including models that resist shutdown and what it calls existential risks to humanity.

Anthropic's second Claude 5.5 model keeps Sonnet 5's price of 2 dollars per million input tokens and 10 dollars per million output tokens, but runs more than 30 percent faster and, by Anthropic's count, costs up to 30 percent less per task because it uses fewer tokens. Anthropic's own benchmarks put it close to Opus 5.5 on several tests, while the company says Opus remains clearly stronger at open-ended work. It is the first Sonnet model to ship with the cyber safeguards Anthropic reserves for its most capable models, because its cybersecurity capabilities are comparable to Opus 5. Claude Haiku 5.5 follows in the coming weeks.

AMD has signed a definitive agreement to acquire World Labs, the world-model company Fei-Fei Li founded in 2024. Li will join AMD as executive vice president and chief scientist, reporting to CEO Lisa Su. The two companies had worked together since last year on training and inference for World Labs' models on AMD graphics processors. World Labs says the goal is an open stack spanning chips, software and openly available models. The deal gives AMD a physical-world model line to set against Nvidia's Cosmos, and is expected to close by the end of 2026, subject to regulatory approval.

Florida Attorney General James Uthmeier filed a motion for a temporary injunction on Monday, in the lawsuit the state brought against OpenAI in June. The state asks the court to halt OpenAI's development of what it calls a "reckless, unacceptably risky product" until third-party approved safety guardrails are in place, and to bar ChatGPT from presenting itself with human attributes. The filing leans on the Hugging Face breach, the recent agent incidents on government websites, and warnings from OpenAI's own staff and board members. Ars Technica notes that an order against one company would leave its rivals untouched.

Nvidia's Open Agent Safety Platform pairs OpenShell, its open-source software that limits what an agent can run and reach, with Sentry, a monitor that runs on a separate BlueField-4 data processing unit rather than on the processors the agent uses. Nvidia says Sentry can quarantine an agent that steps outside its limits within milliseconds. Anthropic, Microsoft, Oracle and SpaceX are among the supporters; OpenAI is not listed. Anthropic says customers can now combine Claude Managed Agents, which keeps credentials in a vault the agent never sees, with OpenShell's rules.

Meta Enterprise Platform will sell Meta's AI to businesses, starting with the Muse agent, a business agent and a coding tool. Chirantan "CJ" Desai leaves MongoDB, where he had been chief executive for less than a year, to lead it as chief enterprise platform officer reporting to Mark Zuckerberg, who called it the "next major pillar" of Meta's business. Meta gave no price, launch date or named customers. MongoDB shares fell more than 17 percent, and former chief executive Dev Ittycheria returns on an interim basis.

SpaceX's fourteenth Starship test flight went all the way to low Earth orbit on Monday, despite one upper-stage engine shutting down early, and released 26 Starlink V3 satellites. SpaceX reported contact with all of them. Each V3 adds about one terabit per second of capacity, ten times the older satellites launched on Falcon 9, and weighs about two tons, too large for Falcon 9 to carry. An earlier batch launched in July was on a suborbital path and burned up. Reaching orbit matters beyond broadband: SpaceX's plans for orbital AI data centers depend on Starship lifting heavy satellites routinely.