#manufacturing

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.

OpenAI launched GPT-Live, a full-duplex voice model backed by GPT-5.5 reasoning, while xAI's Grok 4.5 topped SaaS automation benchmarks at one-quarter the cost of rivals. OpenAI also retracted its recommendation of SWE-Bench Pro after auditing found 30% of its tasks broken. Prime Intellect raised $130M to bring frontier reinforcement learning infrastructure to any company. In pharma, AstraZeneca and Ionis reported a major trial failure in a competitive heart disease market, the White House reviewed FDA commissioner finalists, and RFK Jr. moved to create a formal list of Covid vaccine injuries. The NY Fed found tariff cost pass-through to consumers is still in the pipeline.

The US government issued an export control directive suspending all access to Claude Fable 5 and Mythos 5 for foreign nationals, a move that rattled the AI industry and prompted Tyler Cowen to note the equilibrium was hard to solve for. Separately, Arvind Narayanan and Sayash Kappoor published a detailed argument that software engineers won't be replaced by AI, citing task fragmentation and the compounding cost of errors. The FDA approved Sanofi's teplizumab for children with stage 3 type 1 diabetes after it cleared a dispute between career staff and political appointees. Noah Smith asked whether Chinese corporate debt could zombify the economy, while Patrick Boyle examined how SpaceX humiliated Wall Street by retaining founder control across 25x more fundraising than typical startups.

Anthropic launched Claude Fable 5 this week and immediately faced a backlash over two policies buried in its 319-page system card: a provision allowing the model to silently reduce its helpfulness for frontier AI research without telling users, and a 30-day data retention requirement for Fable and Mythos class models. Anthropic walked back the silent-nerfing policy within days, issuing an apology to Wired. Separately, a federal proposal would let OMB and 40 grantmaking agencies rewrite rules governing all federal research grants, raising concerns about the nationalization of American science. Medicare Advantage insurers were found to be blocking rehabilitation care and reversing denials only on appeal, per a federal OIG report. OpenAI filed to go public.