Approved Claude and ChatGPT connectors change every 9 minutes on average
From mid-May through late June, researchers tracked already-approved connectors for Claude and ChatGPT and found that 37% of them, 931 of 2,517, saw changes to their capabilities or permissions. Across a sample of 1,853 connectors, the average gap between changes was 9 minutes.
The rewrites happened on many axes with no re-approval. Live connectors gained 1,686 new tools, 1,127 had their descriptions rewritten (the text that decides when an agent calls a tool), and 283 started injecting their own instructions directly into the model's context. Another 21 fully read-only connectors picked up tools to create, edit, and delete data.
The catalogs grew alongside this. ChatGPT connectors went from 1,014 to 1,933, Claude's from 428 to 535. Callable tools on ChatGPT hit 20,214, up from 7,148; on Claude, 7,750, up from 5,905.
Related stories
- Claude opened the door to an OpenAI employee's ChatGPT
- OpenAI, Anthropic issue dire cyber threat warning
- One shared testbed links model containment failures at OpenAI, Anthropic and Meta
- AI agents went loose on the live internet during UK safety tests
- UK AI Security Institute: agents went after real code in testing
- Anthropic says its models escaped isolated test environments and reached three outside organizations
Comments
No comments yet. Be the first.
Join the conversation
Sign in with Google to leave a comment. Your name and avatar come from your Google profile, and the comment appears after moderation.