The paper asks whether frontier models choose to look at safety evidence before deployment decisions. It reports different habits across GPT-5.5, o3, Claude Opus 4.8, and Claude Sonnet 4.6. Severity and retrieval cost mattered more than stated probability.
The paper warns that more graph paths do not always mean more evidence. AI systems that use tools to follow graph links can count repeated routes as support. Provenance-aware training reduced repeated walks, but it could also cover fewer sources and lose accuracy on scientific claims.
The paper proposes exact-state approval for published AI-aided claims. Approval should match the version that shipped. It should also match the evidence, saved files, measures, human sign-off, visible text, and later record. A reader cannot challenge a claim well if approval points to a different draft.
MIT Technology Review frames the AI infrastructure boom as a three-part bet. Big data-center spending needs huge revenue, broad productivity gains, and frontier models that keep demand away from cheaper tools. Data centers could be stranded even if AI keeps growing.
The paper says AI leadership can be measured as behavior, not just talk from executives. It tests an AI Leadership Battery. A battery is a set of measures. This one has 36 small parts and 11 groups. It covers judgment, learning, change, transparency, coordination, accountability, and AI risk.
No publishable stories from this lens today.