AI Frontier · Jul 31, 2026
Product and model updates from OpenAI / Anthropic / Google, with security-relevant items flagged (🛡).
OpenAI
- GPT-5.6 (Sol / Terra / Luna)released — the first frontier model to pass per-customer U.S. government review before public release; Sol reaches roughly 750 tokens/sec on Cerebras; GPT-5.6 Sol Ultra reportedly produced a proof of the 50-year-open Cycle Double Cover conjecture using 64 subagents. (government pre-review process)
ThursdAI — July 2026
Anthropic
- Claude Opus 5(released Jul 24) — Within 0.5% of Fable 5's peak on CursorBench 3.2 at half the per-task cost; pricing unchanged at $5/$25.
ThursdAI — July 2026
Google DeepMind / AI
- Gemini 3.5 Prodelayed to July 17 due to a ground-up architecture rebuild.
BigGo Finance
Security cross-view
- RufRoot (CVE-2026-59726, CVSS 10.0)is a reminder: as agent orchestration frameworks such as Claude Code / Codex proliferate, MCP bridges / tool endpoints are becoming an attack surface in their own right — unauthenticated default exposure can lead to unauthorized RCE, API key theft, and learning-store poisoning. Self-built/self-deployed agent infrastructure should close external ports by default and enforce authentication.
The Hacker News
Source configuration: see intel/sources.yaml