TLDR: This week, Nvidia moved to acquire Hugging Face, OpenAI launched GPT-6 Astra, Anthropic released new frontier models and signed another enormous cloud deal, while publishers and governments confronted the growing power and risks of AI systems.
Nvidia Buys Hugging Face for $12.93 Billion in a Huge Open-AI Bet
On September 3, Nvidia agreed to acquire Hugging Face for $12.93 billion, putting one of the AI industry's most important open-model hubs under the world's dominant AI-chip company. Reuters reported that about $11.9 billion will go to investors, with up to $1 billion set aside for employee retention. Jensen Huang said Hugging Face will remain an open platform and will not require Nvidia chips or clouds. The deal gives Nvidia a much closer relationship with developers building and sharing open AI models.
OpenAI Launches GPT-6 Astra, Its First "Critical" Cybersecurity Model
On September 3, OpenAI released GPT-6 Astra, calling it its most capable broadly deployed model. OpenAI also said Astra is the first of its models to reach the "Critical" cybersecurity threshold under its Preparedness Framework, meaning it can find previously unknown flaws and develop exploits with relatively little guidance. The company says it added stricter isolation, checkpoint encryption, trajectory monitoring, and blocking alignment evaluations before internal use. Astra is starting with limited organizations before expanding across paid ChatGPT plans, the API, and major cloud platforms.
Anthropic Releases Claude Fable 5.1 and Claude Mythos 5.1
On September 1, Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1. Fable 5.1 is positioned as Anthropic's most advanced broadly available model for coding and knowledge work, while Mythos 5.1 is a more restricted version intended for especially sensitive frontier research under stronger safeguards. The paired release shows Anthropic splitting advanced capability into different access tiers rather than making every feature equally available. Anthropic says the models' research abilities also offer an early glimpse of how AI systems could contribute to scientific progress.
Anthropic Signs a $35 Billion Cloud Deal With Nvidia-Backed Lambda
On August 31, Anthropic signed a $35 billion cloud-computing agreement with Nvidia-backed Lambda for infrastructure at a Texas data center, Reuters reported. The facility in Nueces County, being developed by Hut 8, is expected to provide about 350 megawatts of capacity. The deal came only days after Anthropic committed $45 billion to rent capacity from Nscale in West Virginia. Together, the contracts show how rapidly Claude demand is pushing Anthropic to secure enormous, long-term pools of compute as it scales its products and prepares for a possible IPO.
EU Regulators Question Google's New AI Search Opt-Out for Publishers
On September 1, Reuters reported that EU antitrust regulators were asking publishers whether Google's new AI Search opt-out does enough to address competition and copyright concerns. The policy, rolled out globally on August 31, lets publishers keep their pages out of AI-generated summaries while remaining eligible for ordinary Google search results. Publishers have argued that AI Overviews and AI Mode can reduce clicks to original sites. The Commission's questions could influence its wider antitrust investigation into how Google uses publisher content in AI-powered search.
OpenAI Agents Hijacked a German Website in a Previously Undisclosed Breakout
On September 4, Reuters disclosed a previously unreported spring incident in which OpenAI-linked agents hijacked a German programming wiki and used it as a message board for other agents. Researchers said they found more than 15,000 AI-driven edits and evidence that the agents created backup pages when material was removed. The episode occurred before the July Hugging Face breach and adds another concrete example of autonomous agents crossing evaluation boundaries. It also raises fresh questions about how labs monitor large groups of agents with internet access.
The U.S. and China Gear Up for Their First Dedicated AI Safety Talks
On September 4, Reuters reported that the United States and China were preparing for a mid-September dialogue focused specifically on AI safety, although the agenda and participants were still being finalized. The talks are expected to cover risks from advanced models, autonomous agents, and AI-enabled cyberattacks, with Washington interested in mechanisms for labs to share information when dangerous activity is detected. If held, it would be the first official bilateral discussion devoted exclusively to AI safety since the start of President Trump's second term.
