Researchers Say Claude for Chrome Flaw Lets Rogue Extensions Trigger Gmail Reads

Summary: Manifold Security reported that any browser extension able to run a script on claude.ai can dispatch a synthetic click on the #claude-onboarding-button element, and the Claude for Chrome content-script handler reads its data-task-id and opens the side panel with one of nine allowlisted tasks - including usecase-gmail, usecase-gdocs and usecase-calendar - without checking event.isTrusted. Manifold rated the flaw CVSS 7.7 High in the default ask-before-acting mode and 9.6 Critical with 'Act without asking' enabled, and found the handler and side-panel initialization byte-for-byte identical in v1.0.80 to the v1.0.72 it reported on May 21, with no patch as of July 14.
Why it matters: Anthropic closed the forged-click report by folding it into the earlier ClaudeBleed trust-boundary issue, which it said 'remains open pending a complete fix,' and eight releases later the vulnerable code has not moved - so every paid subscriber running the beta extension alongside any other extension that can touch claude.ai is still in scope.

Read the original at The Hacker News ↗