AI Frontier · Jun 9, 2026

A standalone archive of product and model developments from OpenAI / Anthropic / Google. Window: past 24 hours (Tuesday edition; some items carry over from this week). 🛡 = security-related.

OpenAI

• ChatGPT adds "Active sessions" and "Lockdown Mode" security features— Active sessions lets users view and sign out of account sessions, showing device, location, sign-in time, and trusted status to make unauthorized logins easy to spot; Lockdown Mode restricts network-facing capabilities (live browsing, deep research, agent mode, file downloads, and some web images). Defense in depth aimed at high-risk users.
  Releasebot — OpenAI
• GPT-5.5 Instant rolls out to all ChatGPT users; GPT-4.5 retires on 6/27 (30-day sunset)— the model lineup keeps rotating as older models are phased out.
  Releasebot — ChatGPT
• Memory upgrade ("dreaming")— memory capacity doubles for Plus/Pro, memories are automatically revised over time ("dreaming"), and recommendations are personalized to the user's current location / time zone; a trimmed version that meets the quality bar is also available to free users.
  Releasebot — OpenAI
• Workspace agents for Enterprise / EDU + new admin controls— ChatGPT Enterprise/EDU adds workspace agents for shared workflows, with the free period extended to 7/6, after which it moves to credit-based billing.
  Releasebot — OpenAI

Anthropic

• Project Glasswing scales up — Claude Mythos Preview extends to about 150 new organizations across 15+ countries (mostly critical infrastructure)— Claude Securitylaunched alongside it: codebase scanning and patch recommendations using Claude Opus 4.8; partners have already found 10,000+ critical vulnerabilities. Most of the new organizations come from critical infrastructure sectors.
  Anthropic — Expanding Glasswing· Anthropic — Claude Code security
• Claude Managed Agents — self-hosted sandboxes (public beta) + private MCP servers— Managed Agents can run inside sandboxes you control and connect to private MCP servers, with both the sandbox and the services inside the enterprise boundary under security and runtime controls; tool execution moves to the configured environment while the agent's main loop stays on Anthropic infrastructure.
  InfoQ
• Claude Opus 4.8 released— stronger than Opus 4.7 on coding, agent skills, reasoning, and practical knowledge work.
  Releasebot — Anthropic
• Confidential S-1 submitted (6/1)— Anthropic submitted a confidential draft S-1 to the SEC, formally advancing the IPO process.
  CNBC

Google DeepMind / AI

• Gemini 3.5 Pro approaching June GA— targeting a 2M token context, a "Deep Think" reasoning mode, and frontier multimodal understanding; general availability is expected in June following the I/O announcement (5/19). 3.5 Flashis already GA (5/19) and is the default model for the Gemini app and Search AI Mode.
  TechTimes· blog.google — Gemini 3.5
• Gemini Robotics-ER 1.6 comes to Boston Dynamics Spot— DeepMind and Google Cloud integrated the embodied version of Gemini into the Spot robot dog and the Orbit AI visual inspection platform.
  Google Cloud Blog

Cross-vendor / Policy

• Four CEOs jointly write to Congress (6/5) calling for mandatory synthetic DNA screening— the AI leads of OpenAI, Anthropic, Google, and Microsoft co-signed a letter to the US Congress demanding mandatory screening for synthetic DNA suppliers, a cross-lab call for biological misuse risk controls.
  Yellow.com

🛡 = security-related