Anthropic Makes Claude in Chrome Generally Available with Autonomous Action Safeguards

·2026.08.26 09:00

Key point

Claude in Chrome is now available on all paid plans, allowing autonomous browser actions protected by new safety classifiers.

1 / 4

Details

Anthropic has made Claude in Chrome generally available on every paid Claude plan. The extension allows Claude to access internal dashboards, legacy systems, and vendor portals that lack direct API connections by using the user's existing logins to view pages, click links, and fill forms.

Autonomous Actions and Safety

Claude can now take actions autonomously in the browser without requiring approval for each step. This is enabled by a safety classifier that validates each action before execution to ensure it matches the user's request and is safe. Users can disable this auto-approval feature in settings if they prefer manual confirmation.

Prompt Injection Defenses

To mitigate the risk of prompt injection attacks, where malicious instructions hidden in web content trick the agent, Anthropic implemented several layers of defense:

  • Model Training: Claude is trained against a growing library of attacks sourced from internal attackers, red-teamers, and real-world monitoring.
  • Probes: Specialized probes scan web content reaching Claude via tool results. If a likely attack is detected, Claude is warned to treat the content with suspicion.
  • Action Verification: A classifier reviews actions (like navigating or typing) against the original user request, blocking mismatches.

Evaluation Results

Anthropic reported results from recent evaluations testing resilience against prompt injection:

  • Initial Evaluation: No attacks succeeded against Claude Fable 5, Claude Opus 5, or Claude Sonnet 5 in the Cowork harness, even without additional safeguards.
  • Current Evaluation: Using stronger attacks from professional red-teamers, attacks that reached the model succeeded against Opus 4.5 17.6% of the time and Opus 5 3.8% of the time before safeguards.
  • With Safeguards: With probes and safety classifiers active, no attacks succeeded against Claude Sonnet 5 or Claude Opus 5. Claude Fable 5 had a 0.3% attack success rate, with all successful breaks verified as low-severity.

Availability and Limitations

Claude in Chrome is available via the Chrome Web Store. Enterprise admins can manage access and limit domains through Organization Settings. The extension does not currently support other Chromium browsers or mobile devices; users must still use the Claude desktop app for local file access.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.