Morning brief

Saturday, June 27

0 people · 0 companies · 3 topics

TL;DR The US government gave Anthropic a partial win: Mythos 5 is restored for critical-infrastructure operators on day 15 of the export ban, but Fable 5 is still dark. OpenAI separately previewed GPT-5.6 Sol overnight under its own government-gated rollout.

Models & research

  • model-releases: US Commerce Dept notified Anthropic overnight it can redeploy Claude Mythos 5 to US organizations operating and defending critical infrastructure; Commerce Secretary Howard Lutnick’s letter cites “appropriate safeguards.” Fable 5 remains suspended for all general API, Claude Code, and subscriber traffic. Day 15 of the ban; first permissioned restore. Semafor · Axios · CNBC · MLQ

  • model-releases: OpenAI previewed GPT-5.6 Sol, Terra, and Luna last night in a limited rollout to ~20 government-approved partners. Sol scores 88.8% on Terminal-Bench 2.1 (just ahead of Mythos 5’s 88.0%) and beats GPT-5.5 on ExploitBench using roughly a third of the output tokens. Terra is pitched at 2x cheaper than Sol; Luna targets high-volume economy workloads. General availability promised in coming weeks; OpenAI said the government-gated launch “shouldn’t be the norm.” TechCrunch · VentureBeat · Engadget

OSS

  • oss / agentic-ai: cognee (topoteretes/cognee) is up +729 stars today — persistent long-term memory for AI agents built on a self-hosted knowledge graph plus vector search, running on a single Postgres instance; v1.2.0 preview adds Slack and Google Drive connectors. GitHub

  • oss: xbtlin/ai-berkshire (+686 stars today, Python) is a value-investing research framework that chains Claude Code agents for multi-angle company analysis; niche but heavily starred for an overnight project. GitHub