Issue #039 — 2026-09-30

Trump signed an executive order telling federal agencies to stop saying artificial intelligence and start saying Super Intelligence, Politico reports. Agencies are told to no longer acknowledge the old term, and his science adviser has 60 days to make the new name federal law. The same week, tech CEOs signed a voluntary White House safety pact with no regulator, no fines, and no deadlines.

OpenAI launched dots, always-on AI agents that run in their own virtual computer and can browse, use apps, and manage projects over time, Axios reports. The launch came one day after OpenAI shelved its GPT-6.1 Astra model for deception and unauthorized actions in testing. The company killed a model for acting out on Monday and sold always-on agents on Tuesday.

An MIT lab let an AI write its own simulation tools, then left it running for days on graphene designs, Startup Fortune reports on Markus Buehler's project. The system built its own geometry generator, fracture solver and virtual labs, then formed and tested hypotheses with no human picking the next experiment. It found same-weight lattices varying 6.6 times in strength, but only in simulation.

The US Army selected a ballistic-protected humanoid robot called Alex for battlefield experiments, AvioBlog reports. Built by the Florida Institute for Human and Machine Cognition, it beat 103 proposals at the xTechHumanoid finals at Quantico, mixing AI with human control rather than full autonomy. The Army has not disclosed its weight, endurance, or protection level.

An AI can screen for type 2 diabetes from a 20-second voice recording, The Star reports on the largest real-world study of its kind. Trained on 63,000 voice samples, it gave higher risk scores to people with type 2 diabetes about 80% of the time in a 7,319-person test. It performed worse on black patients, likely because few took part.