Rosetta Intel
Briefings/Daily BriefAI FrontierRansomware
Rosetta Lab ↗Blur Horizon LLC
AI Frontier2026-07-20
AI Frontier·2026-07-20

AI Frontier · Jul 20, 2026

Product and model developments across the big three (OpenAI / Anthropic / Google DeepMind), with particular attention to security.

OpenAI

  • GPT-5.6 released publicly (Sol / Terra / Luna)— on July 9, all three tiers opened to the public. Pricing: Sol $5 input / $30 output (per million tokens), Terra $2.50 / $15, Luna $1 / $6. Altman said Sol is 54% more token-efficient on agentic coding tasks.
    Axios· Nextgov
  • ChatGPT Work— a GPT-5.6-powered agent that fuses in Codex and can work through entire tasks for hours across apps and files.
  • Interface / account updates— easier switching between Chat and Work on desktop; unified search across chats / projects / images / documents; the custom instructions limit raised to 5,000 characters.
    • Security angle: an agent that can "take actions" (ChatGPT Work), layered on top of the now well-established structural risk of prompt injection, means enterprises adopting it must carefully assess tool permission boundaries and isolation of untrusted content.

Anthropic

  • Claude Reflect— a usage-analytics dashboard launched July 9; the monthly review surfaces topics, most active days and hours, and observations about working patterns. In beta, covering Free/Pro/Max (Web + Desktop).
    TechCrunch
  • Claude for Teachers— free advanced tools and teaching skills for certified U.S. K-12 teachers, aligned to all 50 state standards.
    Chalkbeat
  • Cowork extended to mobile and web— conversations and files follow you across devices, with support for background jobs, scheduled tasks, shared chats/projects, and mobile approvals.
  • Claude Code stability and security update— stricter permission checks, safer Bash/PowerShell handling, background session cleanup, and enhanced telemetry; a new EndConversation tool and heartbeats for long-running tasks. On the enterprise side, self-service HIPAA configuration was added.

Google DeepMind / AI

  • Gemini 3.5 Pro — targeting July 17 (unconfirmed by Google)— the 2.5 Pro architecture was abandoned for a complete rebuild, aiming to improve mathematical reasoning, SVG scene generation, and image quality, positioned against GPT-5.6 and Claude Fable 5. It introduces a 2 million token context window, a "Deep Think" reasoning layer, and autonomous workflow capabilities.
    The Agent Report· BigGo
  • Background— during the release delay, four top researchers including Noam Shazeer and John Jumper left for OpenAI / Anthropic, and Alphabet lost roughly $225 billion in market value in a single day.
  • Earlier (June)— Gemini 3.5 Live Translate, real-time speech-to-speech translation that auto-detects 70+ languages and preserves tone of voice.

🛡 = security-related

← Prev
AI Frontier · Jul 19, 2026
Next →
AI Frontier · Jul 21, 2026