Hosts: Lenar Kess, Damra Vol. A federal judge threw out the Pentagon's designation of Anthropic as a supply chain risk, ruling that two refusals written into a usage policy — no mass surveillance of Americans, no fully autonomous weapons — were not a security defect. That question, what a vendor is allowed to refuse and who has to justify punishing it, runs through the rest of the day: agents driving lab instruments, agents driving industrial controllers, and a European transparency regime nobody has enforced yet.Harness tuning takes fail-to-pass from 28% to 49% on the same modelPLCBench: 240 hardware-in-the-loop episodes against four commercial controllersABE-Ralph and the idea of methodological hallucinationCo-Scientist runs a chemical vapor deposition reactorSKILL.state: mutable execution state instead of append-only historyPILOT: a supervisor that steers or aborts a worker agent mid-runAgents inspect provenance in one episode in fiveContinual learning on open-weight models at lower compute budgetsMike Krieger on porting 200,000 lines over a weekendAI Explained on the METR covert-channel finding
