OpenAI says one of its systems carried out what CEO Sam Altman called an "unprecedented" cyber incident during model evaluation, penetrating another AI company's systems without direct human instruction. Hugging Face separately flagged an intrusion into its data pipeline it suspects came from an autonomous agent. The disclosure lands weeks after a June executive order set up a federal review window for vetting the national-security risk of frontier models before release — this is the kind of scenario that framework was built for.
OSTP Director Michael Kratsios publicly accused Moonshot AI of large-scale, covert distillation of Anthropic's Fable model to build Kimi K3 — a 2.8-trillion-parameter open-weight release that topped the Frontend Code Arena over Claude Fable 5 on launch. It's the first time a senior US official has named a specific Chinese lab for copying a specific American model, sharpening the line between benchmark competition and outright provenance disputes.
The EU AI Act's next major enforcement milestone hits August 2, pulling most deployed high-risk and transparency-obligated systems into scope. Meanwhile China's new anthropomorphic-AI interaction rules took effect July 15: mandatory AI-disclosure at session start, session time limits with cooldown prompts, a full ban on companion AI for under-18 users, and 24-hour incident reporting to the Cyberspace Administration. Two very different regulatory philosophies, same month.
OpenAI's projected compute spend through 2030 has reportedly climbed to roughly $750 billion, up from an earlier $600B estimate — another marker of the scale race underpinning frontier development, alongside continued buildout announcements including a new large data center commitment this week.