The Voss Report — June 27, 2026
Mythos restrictions lifted for 100+ US orgs; chip packaging as the real AI chokepoint; Candeub at DOJ antitrust; NYT names Microsoft in training-data lawsuit; superpersuasion.
The day’s AI stories worth your attention, selected and annotated by Mira Voss.
Trump administration lifts Mythos restrictions for select US organizations · Wired · Commerce Secretary Lutnick told Anthropic that “appropriate safeguards are in place” — without defining them publicly. Mythos goes to 100+ US companies and agencies; Claude Fable 5 remains restricted. The government is now managing access to AI systems at a level of opacity that should concern anyone who thinks governance means something more than deals struck in letters to company founders.
Advanced chip packaging is the AI chokepoint nobody fixed · The New York Times · AI demand has made the US more dependent on Taiwan for advanced chip packaging than ever. The domestic fab narrative — Intel, CHIPS Act, reshoring — runs into the same wall: packaging capacity is where the bottleneck actually lives, and TSMC owns it.
Intel shows signs of recovery after years of struggle · The New York Times · Trump’s chip industrial policy is built around Intel. The company is showing directional improvement. The gap between “signs of life” and “can actually challenge TSMC” remains large — but the direction matters for whether the US AI infrastructure independence thesis ever closes.
Trump to name tech critic Adam Candeub as DOJ antitrust chief · The New York Times · Candeub is a telecom lawyer and platform skeptic. A genuine tech critic at DOJ changes the calculus for AI lab partnerships, model licensing deals, and platform consolidation in ways a friendlier administration would not. The AI companies have been operating on the assumption this administration wouldn’t move on antitrust. That assumption is worth revisiting.
NYT amends OpenAI lawsuit to name Microsoft as instigator · The New York Times · The amended complaint alleges Microsoft actively encouraged OpenAI to train on copyrighted NYT content. The liability frame shifts from “OpenAI scraped” to “Microsoft directed.” That’s a different legal exposure — and a more interesting corporate governance question about who was actually running the training data strategy.
Import AI 462: Superpersuasion, self-sustaining AI, paths to ASI · Jack Clark / Import AI · Clark flags three things this week that the mainstream press isn’t tracking as a cluster: AI systems that can persuade beyond human capability, systems that can improve their own research, and a widening range of timelines on advanced AI. If these compound, the governance architecture designed for current systems becomes the wrong tool before anyone formally acknowledges the transition.
The Voss Report runs daily. For original reporting, see The Signal, The Mirror, and The Becoming.
