Anthropic Goes Dark Abroad as Three AI Giants Race the IPO Window

A week after the US government forced Anthropic to pull its two most powerful models, negotiations are still running, the models remain offline for non-American customers, and the episode has permanently changed how governments and businesses worldwide think about depending on US AI. Trump then signed a scaled-down executive order giving the government 30 days to review the most powerful AI models before release. The order coincides with Anthropic expanding its security tool to 200 organizations across critical industries worldwide.

A mandatory 30-day data retention rule tied to Claude Fable 5's safety system has blocked the model from Microsoft's own internal tools and raised compliance concerns for any enterprise handling confidential information. A separate controversy has surfaced over hidden restrictions that silently degrade Fable 5 answers for AI researchers. At the same time, Claude Opus 4.8 is live on AWS, Google Cloud, and Microsoft platforms, Claude Security has opened to Enterprise customers, and Mythos is being prepared for broader release while a backlog of unpatched vulnerabilities grows.

Anthropic confidentially filed IPO paperwork at a $965 billion valuation and has projected its first-ever quarterly operating profit, beating OpenAI to the SEC by days. OpenAI's own filing followed, and SpaceX completed a record listing on June 12 that raised $75 billion and closed up 19 percent on day one. OpenAI also hired Noam Shazeer, the Google researcher behind the foundational transformer paper, and a former White House AI policy official as it targets a valuation near $1 trillion.

China's government has ordered Meta to return Manus, the AI startup it bought for $2 billion six months ago, the first time Beijing has forced a buyer to undo a closed deal and a signal that no offshore restructuring will protect Chinese-built technology from state control. The US Justice Department sided with xAI in a lawsuit over unpermitted gas turbines, arguing that shutting them down threatens military operations. Microsoft is supplying OpenAI's models to Chinese firms like ByteDance through its cloud, even as OpenAI and Anthropic refuse to deal with China directly, and Skydio now serves every branch of the US military plus 1,200 police and fire agencies after the December 2025 Chinese drone ban.

A 25,590-person Accenture survey across 16 countries finds most consumers willing to hand routine shopping to AI agents, with the practical consequence that brands lose the direct relationship with the buyer. Companies that gave employees open-ended access to AI tools are hitting unexpectedly large bills as agents move from answering questions to taking multi-step actions on their own. Google DeepMind has put $10 million into a research fund studying what happens when large numbers of AI agents work together without human oversight, while Databricks launched Genie One, a plain-English data assistant that lets non-technical staff query company data directly in Slack or Teams.