Claude in Chrome hits GA on every paid plan, with autonomous mode
The browser agent that clicks, types and fills forms leaves preview, gated by a per-action safety check — memory now spans chat and Cowork, o3 exits ChatGPT.

Copy markdown
GA on every paid plan
Claude in Chrome has left preview and is now available on all paid plans — Pro, Max, Team and Enterprise — not just Max. The extension reads pages, clicks, types, fills forms and pulls data out of dashboards, so it's a browser agent you can point at real, multi-step web work today.
Autonomous mode, gated per action
You can now let it complete tasks without approving every step. Before each action a separate safety classifier checks it for risk and for instructions hidden inside the page, and Claude hard-stops before purchases and other one-way doors. New prompt-injection defenses ship alongside, and Team/Enterprise admins get site allow/blocklists.
Memory now spans chat and Cowork
Separately, Claude's memory now carries across both regular chats and Cowork sessions, so context you gave in one surfaces in the other. Topics are editable in Settings, a new sensitive-topics control keeps health and beliefs out by default, and memory is on by default for Free/Pro/Max (off for Team/Enterprise).
Elsewhere: o3 leaves ChatGPT
OpenAI retired o3 from the ChatGPT model picker on Aug 26 after its 90-day sunset; if you leaned on it in the app, move to the GPT-5.x line. The API is unaffected — o3 is still callable there — so only your in-app workflows need to change.