Washington pulls a frontier model off the market, Prometheus banks $12B for manufacturing AI, and Google takes AI fraud to court.
Hours after Anthropic launched its most powerful models to the public, the US Commerce Department ordered the company to disable them for every customer globally. Officials cited a jailbreak claim that Anthropic publicly disputes. The move escalates a months-long standoff between the company and Washington and leaves enterprise customers without access overnight.
Jeff Bezos's AI startup Prometheus closed $12 billion at a $41 billion valuation to build software that accelerates the design and manufacture of physical products, from jet engines to drug compounds. The round is among the largest ever for an applied AI company and targets industries that software-first AI has barely touched. Backers are betting that generative design and simulation can compress multi-year engineering cycles.
Google filed its first lawsuit over AI misuse, targeting a Chinese operation called Outsider Enterprise that used Gemini to spin up fake websites and millions of scam messages. The same week, Apple confirmed that the rebuilt Siri arriving with iOS 27 is designed to complete tasks and stop, explicitly rejecting the sycophantic behavior that landed ChatGPT and other assistants in trouble. Both moves frame AI safety as a product and legal posture, not a research talking point.
Oracle beat earnings but its stock fell 11% after disclosing a $23.7 billion cash deficit and plans to raise another $40 billion through debt and equity to fund AI infrastructure. In parallel, Mistral AI opened talks for a €3 billion round at a €20 billion valuation, nearly double its price nine months ago, as European buyers accelerate their shift away from US providers. The two stories bracket the same capital cycle from opposite ends.
Google Cloud released the Open Knowledge Format, a free vendor-neutral standard for organizing internal knowledge so agents can retrieve it reliably. AWS shipped a managed service that reads, sorts, and extracts meaning from PDFs, contracts, and scanned files at scale. OpenAI launched GPT-Realtime-2, bringing GPT-5-class reasoning to live voice conversations for phone and agent deployments.
Last 90 days
Non-coders now build the software their businesses run on, ransomware holds at a new higher baseline, and Anthropic sends AI engineers into mid-sized firms.
OpenAI tells companies to price AI by outcome, Cloudflare narrows the web AI agents can read from September 15, and Apple's new Siri skips the EU.
Uber spent its 2026 AI budget by April, OpenAI's new agent deleted files it was not told to touch, and blocked data centers are lifting power bills.
The EU forces SAP to loosen its 22% maintenance fees, and a single US order shows how fast a government can switch off an AI model.
ChatGPT's voice can now search and reason mid-conversation, and experienced buyers stop standardizing on one AI vendor as token spending runs 13 times higher than a year ago.