AI Frontier · Jun 5, 2026
Daily developments from OpenAI / Anthropic / Google DeepMind · Window: past 24 hours (2026-06-04 → 2026-06-05)
🛡 = security-related
OpenAI
- GPT-5.5 Instant readability upgrade + legacy model retirement schedule (6/3)— GPT-5.5 Instant rolls out to all ChatGPT users, emphasizing a more natural, more readable answer style with fewer verbose bullets; Canvas is removed from 5.5 Instant / 5.5 Thinking, with writing and code moving to in-conversation writing/code blocks. GPT-5.3 Instant stays for paid users for another three months. Retirements: GPT-4.5 exits ChatGPT on 6/27(30-day sunset), o3 exits on 8/26(90-day sunset; o3 remains available long-term in the API). gHacks· Bleeping Computer· the-decoder
- GPT-Rosalind (life sciences) update— combines GPT-5.5's agentic coding/tool use with stronger drug discovery, genomics, and medicinal chemistry capabilities, adds evidence retrieval and bioinformatics workflow plugins, and expands the trusted-access research preview to qualifying institutions worldwide. OpenAI News
- ChatGPT Sites in preview for Business— teams can build and deploy internal web apps with Codex, complete with hosted URLs, Sign-in-with-ChatGPT integration, and workspace controls. Releasebot
- codexui-android npm token-theft campaign (6/3–6/4)— an npm package posing as a "Codex remote web UI" (29,000 weekly downloads) hid its stealer code only in the npm build, exfiltrating Codex access/refresh/ID tokens for a full month. Refresh tokens do not expire, allowing indefinite impersonation of the developer. Codex users should revoke and rotate tokens and audit dependencies.See the AI security section of the main brief. The Hacker News· Aikido
Anthropic
- KPMG global alliance (6/3–6/4)— Claude rolls out to all 276,000 KPMG staff across 138 countries/territories, embedding Cowork and Managed Agentsinto KPMG's Microsoft Azure-based Digital Gateway, starting with tax/legal client tooling. A major focus of the alliance is private equity: Anthropic is designated the preferred partner for deploying Claude and its agents to KPMG's PE portfolio companies. Following the Services Track / Partner Hub, this is another landmark enterprise rollout. The Next Web· Anthropic news
- Project Glasswing (continuing)— roughly 200partners across 15+ countries / about 150 new organizations(critical infrastructure including power, water, healthcare, telecommunications, and hardware); Claude Security(Opus 4.8-powered codebase scanning + patch recommendations) is live, and Glasswing's internal tooling is open to trusted security teams on request. The "10,000+ high/critical vulnerabilities in 30 days" storyline continues, with the bottleneck still "not fixing fast enough." Anthropic — Expanding Glasswing· TechCrunch
- Claude Code / Opus 4.8 (continuing)— high effort by default on hard tasks, dynamic workflows (orchestrating dozens to hundreds of agents in the background), and a faster Fast mode. Dual-use capability: the same week, Sophos disclosed an AI ransomware toolkit whose development leaned on Claude Opus + Cursor agents. Releasebot· Help Net Security
- IPO context (reminder)— a confidential SEC filing was submitted in late May / on 6/1, following a $65B Series H at a roughly $965B valuation.
Google DeepMind / AI
- Gemini 3.5 Flash GA + Deep Think rollout (continuing)— Gemini 3.5 Flashis positioned as the strongest agentic/coding model currently available, beating Gemini 3.1 Pro on several key benchmarks; Gemini 3 Deep Thinkis available to Ultra subscribers as the top-tier reasoning mode. The Ultra subscription drops to $200/month, with a new $100/month Developer tier. Gemini 3.5 announcement· Google Cloud — I/O 26
- DeepMind recruits Contextual AI researchers (continuing)— reportedly an $80–90M licensing/talent deal. HeyGoTrade
- "AI finds bugs" keeps delivering— this issue's Redis CVE-2026-23479and NGINX Rift CVE-2026-42945, both unauthenticated/RCE-class flaws, were found by autonomous AI code analyzers, extending DeepMind/Big Sleep's earlier results; this corroborates the Verizon 2026 DBIR theme that exploitation has overtaken credential theft as the top initial access vector — the speed of AI bug-finding keeps compressing the defender's patching window. The Hacker News — Redis· The Hacker News — NGINX
This file is the standalone AI Frontier archive; the main brief is at intel/daily/2026-06-05.{zh,en}.md