#alignment

AI agents are increasingly acting on real infrastructure without adequate authorization checks, from a gym-booking exploit to an unresolved Hugging Face attack timeline. Economists and researchers are meanwhile debating whether to slow AI self-improvement through regulation, with dueling arguments this week over data center bans and safety-rule timing. Elsewhere, GitHub retired its unified LLM API, and the FDA rejected a manufacturing-flagged cancer drug.

Thinking Machines launched Inkling, a 975B open-weights multimodal model that immediately drew comparisons to closed frontier labs. xAI's Grok CLI sparked a data-privacy backlash after users reported it uploading entire home directories. Eli Lilly acquired psychedelics biotech AtaiBeckley for $2.8B, and CMS signaled a structural overhaul of how it pays for clinical AI.

OpenAI quietly launched GPT-5.6 in three tiers (Sol, Terra, Luna) restricted to trusted partners, while Semgrep reported that GLM 5.2 outperforms Claude on its internal cybersecurity benchmarks. Portugal's National Health Service contracted Sword Health to deliver AI-assisted physical therapy nationwide, the first country-scale deployment of its kind. Edison Scientific and Population Health Partners announced a deal to use AI agents for end-to-end drug discovery. Patrick Boyle assessed Brexit's 10-year economic toll, and Noah Smith examined whether AI will push companies toward more or less outsourcing.