Morning brief

Sunday, June 28

1 people · 0 companies · 4 topics

TL;DR US government cleared Anthropic to redeploy Mythos 5 to ~100 orgs on June 27; OpenAI previewed GPT-5.6 Sol the day before under similar gated restrictions.

Models & research

  • Anthropic got US government sign-off to restore Claude Mythos 5 to roughly 100 selected companies and federal agencies; Commerce Secretary Lutnick cited “significant progress” after 16 days of daily talks. Fable 5 is still blocked and has no redeployment date. Euronews · CNBC · AnthropicAI on X

  • OpenAI previewed GPT-5.6 as three tiered variants: Sol (frontier reasoning, long-horizon agentic), Terra (GPT-5.5-competitive at 2x lower cost), Luna (fastest/cheapest). Access is gated to “trusted partners” via API and Codex only, not in ChatGPT — same pattern as Mythos. OpenAI · CNBC

  • China’s Zhipu AI (GLM-5.2) benchmarked at or near Claude Mythos levels on security-bug detection, per Semgrep research. WSJ ran it as “China Has Matched Anthropic in Cybersecurity, Resetting AI Race” — the timing, right as US labs face export controls, is notable. CNBC · X

People

  • karpathy’s CLAUDE.md reportedly grew from 4 rules to 10 with a new self-check protocol for agentic coding loops; the expanded file is circulating today (sourced to Karpathy at Anthropic) but he has not confirmed it publicly. The original 4-rule file already crossed 100K GitHub stars. TechTimes · Raytar on X

Tools & dev

  • claude-code v2.1.195 (June 26): adds /rewind to resume from before /clear, live file-path autocomplete in bash mode, CPU usage cut ~37% during streaming, fixed hook matchers for hyphenated identifiers. v2.1.193 added autoMode.classifyAllShell and OTel assistant-response logging. GitHub Releases

OSS

  • DeusData/codebase-memory-mcp is today’s top GitHub trending repo (+2,162 stars): an MCP server that indexes any codebase into a persistent knowledge graph, 158 languages, sub-ms queries, claims 99% fewer tokens vs file-by-file exploration. Ships as a single static binary, zero dependencies. GitHub

  • xbtlin/ai-berkshire also trending (+1,456 stars): AI value investing research framework built for Claude Code with multi-agent analysis. GitHub