Rosetta Intel

Daily developments from OpenAI / Anthropic / Google. 🛡 = security-related.

OpenAI

GPT-5.3-Codex released— OpenAI says the model "played a key role in creating itself," continuing the trend of self-improving coding models. Worth watching for its double-edged effect on automated vulnerability discovery / remediation.
AI update tracker

Claude Fable 5 (Mythos-class) released publicly (around 6/9)— already generally available (GA) on Amazon Bedrockand GitHub Copilot.
Security / compliance points: Fable 5 mandates 30-day data retention + human review(for trust and safety purposes), overriding existing zero data retention (ZDR) agreementsfor this model class. The other Claude models (Opus 4.8 / Sonnet 4.6 / Haiku 4.5) can still run under ZDR — only Fable 5 cannot, so regulated industries need to assess the compliance impact before adopting it. Bloomberg reports that this public version "does not include cyber capabilities."
Anthropic announcement· Cybernews — data retention· TechCrunch
"When AI builds itself" (6/4)— Anthropic published a piece calling for globally coordinated slowdown / pause of frontier AI development; Fable 5 shipped days later, prompting discussion of "calling for a slowdown while releasing the most powerful model."
IPO— Anthropic filed confidentially for an IPO around 6/1.
Project Glasswing scales up— the Claude Mythos preview opens to about 150 more organizations; it is said to have helped partners find 10,000+ high-severity vulnerabilities.

Gemini 3.5 Pro expected this month(Pichai: "give us one more month"; no firm date yet); Gemini 3.5 Flash(announced at I/O on 5/19) is now the default for the Gemini App and Search AI Mode, at API pricing of $1.50 / $9.00 per million tokens.
Gemini Robotics-ER 1.6integrated into Boston Dynamics' Spotrobot dog and the Orbitvisual inspection platform (a Google Cloud + DeepMind collaboration) — embodied AI enters industrial inspection.
June 2026 AI launch wave

Configuration is in the ai_frontier category of intel/sources.yaml