Sunday: export controls lifted, new models gated — safety is the new launch strategy
The week's throughline is access — who gets the most capable models, under what conditions, and why. From Anthropic's Fable 5 reinstatement to OpenAI's government-only GPT-5.6 preview, the frontier is increasingly shaped by what labs choose to withhold as much as what they ship.
Anthropic redeploys Claude Fable 5 after US export controls lifted; details safeguard framework and government collaboration
A jailbreak triggered real export controls — and lifting them required a new industry-wide severity framework, setting a template for how labs and governments will negotiate access going forward.
OpenAI Previews GPT-5.6 Family (Sol, Terra, Luna) with Government-Only Access and Advanced Safety Guardrails
GPT-5.6's government-only rollout and layered biosecurity guardrails show OpenAI treating access tiers and safety evals as core release infrastructure, not afterthoughts.
Microsoft reveals MAI-Thinking-1, a from-scratch reasoning model with MoE architecture
Microsoft building a frontier reasoning model from scratch — not distilled from OpenAI — signals a strategic decoupling that could reshape the lab's model supply chain.
Get the next edition in your inbox
Midweek and Sunday. Primary-sourced AI news paired with the guides that explain it.