Morning brief

Tuesday, June 23

1 people · 0 companies · 5 topics

TL;DR Claude hit a broad service outage today starting 14:19 UTC with all services affected and a fix in progress; OpenAI launched the full GPT-5.5-Cyber model with automated patch generation for vetted defenders, and Fable 5 simultaneously flipped to usage-credit billing across all paid plans.

Models & research

  • model-releases Fable 5 is no longer bundled in Pro, Max, Team, and seat-based Enterprise plans as of today; further use bills at API rates ($10/$50 per M tokens in/out). The export-control suspension June 12-18 meant subscribers got roughly 4-5 effective days out of the advertised 13. Anthropic · Developers Digest

  • model-releases GPT-5.6 may be silently rolling out: some ChatGPT Pro users report noticeably sharper outputs when selecting GPT-5.5 Pro, and an independent researcher pegs June 25 as the public drop date. Claimed: 1.5M-token context, reasoning budget raised from 768 to 960. No official announcement yet. TechTimes

  • model-releases OpenAI launched the full GPT-5.5-Cyber model (June 22) for vetted defenders: CyberGym 85.6%, ExploitGym 39.5%, SEC-bench Pro 69.8%. Ships alongside an updated Codex Security plugin and “Patch the Planet,” an open-source vuln-patch initiative. Access requires advanced account security verification. Axios · OpenAI

People

  • simonw published two posts today: “A pelican for GPT-5.5 via the semi-official Codex backdoor API” documenting an undocumented API path in Codex, and “Extract PDF text in your browser with LiteParse for the web.” simonwillison.net

Tools & dev

  • claude-code Claude hit a broad outage today starting 14:19 UTC, with elevated error rates on Claude Opus 4.8 across claude.ai, Claude Code, Claude Cowork, Claude Console, and the API. Anthropic confirmed “a fix is being implemented” by 14:25 UTC; 8,000+ Downdetector reports in the US. TechRadar · Claude Status

  • claude-code Latest release ships agent teams (spawn directly via Agent tool name parameter), nested skills loading from subdirectory .claude/skills/, Tool(param:value) permission rules, and a subagent safety fix closing the gap where subagents could request blocked actions before classifier review. Claude Code Docs

OSS

  • mukul975/Anthropic-Cybersecurity-Skills (+1,040 stars today): 817 structured cybersecurity competencies mapped to MITRE ATT&CK, NIST CSF 2.0, and four other frameworks. GitHub

  • jamiepine/voicebox (+1,042 stars today): open-source AI voice studio for cloning, dictating, and creating, TypeScript. GitHub

  • calesthio/OpenMontage (+3,590 stars today): open-source agentic video production system, 12 pipelines, 52 tools, 500+ agent skills. GitHub

  • DeusData/codebase-memory-mcp (+1,299 stars today): MCP server indexing codebases into persistent knowledge graphs across 158 languages. GitHub

Other

  • US officials and intelligence partners warned that some AI models are months from being capable of launching cyberattacks powerful enough to overwhelm governments and major companies. CNN