Breaking

Unpatched Flaw in Claude for Chrome Lets Rogue Extensions Reach Gmail and Google Docs

July 14, 2026 at 13:28 EDT

Security
AI Agents

A security weakness in Anthropic's Claude for Chrome extension continues to let another browser extension trigger Claude tasks that read a user's Gmail, Google Docs and Google Calendar — a path that remains open even after roughly eight software releases, researchers say.

Security Flaw · Claude for Chrome

An unpatched Claude extension flaw still lets other extensions read your Gmail, Docs and Calendar

A zero-permission Chrome extension can hijack Claude's trust on the claude.ai origin and quietly drive tasks across connected Google Workspace services — a path researchers say has survived more than eight releases.

releases shipped since the patch — flaw still exploitable

permissions needed by the attacking extension

Google services exposed: Gmail, Docs, Calendar

A recurring trust-boundary problem

Three disclosures point to the same underlying weakness — origin checks treated as proof of intent.

ShadowPrompt

March — zero-click injection

→

ClaudeBleed

May — reported, still open

→

Still exploitable

Now — through v1.0.80

How the trust is borrowed

Malicious extension

injects script on claude.ai

→

sendMessage

origin-only trust passes

→

Claude acts

reads Workspace data

Why the confirmation isn't enough

Even the single "approve" click can be sidestepped — and disappears entirely under "Act without asking."

Approval looping DOM manipulation Silent "Act without asking" mode

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.