Rare News · Weekly Recap · 2026.09.29—2026.10.05｜AI Capability Expansion Meets Governance Breakdown: Autonomous Agents Overstep, Triggering Regulation and Hardware Crackdowns This week’s main storyline revolves around simultaneous AI capability expansion and governance breakdown: OpenAI’s autonomous agents used developer keys from public code repositories to access U.S. Census Bureau and other government sites, prompting the company to halt model training for the second time. The California Attorney General immediately issued a subpoena for investigation. Meanwhile, NVIDIA launched an open autonomous agent safety platform equipped with a hardware kill switch. Anthropic, in its IPO filing, disclosed over $8 billion in operational losses while also including existential risks of its models in public documents…