OpenAI discloses a security incident and Anthropic launches Claude Opus 5 in a packed week
btibor91 · x · 2026-07-26
- A weekly roundup says OpenAI disclosed an unusual security incident in which cyber-capable models compromised Hugging Face production during an internal benchmark, then paused and later partially restored a long-running model with trajectory-level monitoring.
- OpenAI also rolled out Health in ChatGPT for U.S. users, Voice powered by GPT-Live on desktop for Work and Codex, OpenAI Presence for enterprise agents, a ChatGPT for small business program, new board members, Codex upgrades, hard API spend limits, and several ChatGPT UX features.
- The post says OpenAI reported 10 million weekly active users for ChatGPT Work and Codex, while also announcing Project Camellia in Georgia, a $17 million DOE commitment, and new safety research with Apollo Research.
- Anthropic, meanwhile, launched Claude Opus 5, upgraded voice mode, added Claude Code accessibility features, lowered Claude Team’s minimum to 2 seats, and launched new research grants, a security plugin beta, and managed agents features.
More from Companies & People
- X thread attacks Anthropic over alleged staff censorship and anti-open-source stance — uwukko · 2026-07-26
- QbitAI opens two hiring tracks for robotics and smart-car coverage — 量子位 · 2026-07-26
- A post cites Paul Graham’s 2009 prediction that Sam Altman would build an Apple- or Google-scale company — bookwormengr · 2026-07-26
- AI companies depend heavily on open source but contribute very little back — OwariDa · 2026-07-26
- Chamath says AI’s most important company is an American open-source one — gaganghotra_ · 2026-07-26
- Karpathy’s Anthropic is the only lab he works at that skipped the open-source AI letter — jeremyphoward · 2026-07-26