US EditionPublished • loading... • Updated
Google Says Chrome's AI Creates Risks only More AI Can Fix
Google's new Gemini-based User Alignment Critic reviews AI agent actions in Chrome to prevent unsafe behavior and indirect prompt injection, protecting user data and finances.
- On Monday, Google announced a Gemini-based User Alignment Critic for Chrome to vet agent actions and protect agentic features previewed in September, with rollout planned in coming months.
- Amid rising attacks, Google argued stronger controls were needed as it identifies indirect prompt injection as the primary threat to agentic browsers causing data leaks or fraud.
- Only metadata is revealed to the User Alignment Critic to prevent poisoning; it runs after the planner to veto or retry actions while Agent Origin Sets restrict sites and iframes.
- Users will be asked before the agent completes purchases or messages, Chrome pauses sign-ins and requests Password Manager permission for sensitive sites, and Google expanded its Vulnerability Rewards Program, offering up to $20,000 for security breach demonstrations.
- The approach mirrors an accepted pattern of one model moderating another, with Google pairing deterministic rules, model-level protections, isolation boundaries and user oversight as layered defenses.
Insights by Ground AI
19 Articles
19 Articles
Google details security measures for Chrome’s agentic features
An increasing number of browsers are experimenting with agentic features that will take actions on your behalf, such as booking tickets or shopping for different items. However, these agentic capabilities also come with security risks that could lead to loss of data or money. Google detailed its approach to handling user security on Chrome using observer models and consent for user action. The company previewed agentic capabilities on Chrome in …
Coverage Details
Total News Sources19
Leaning Left2Leaning Right0Center4Last UpdatedBias Distribution67% Center
Bias Distribution
- 67% of the sources are Center
67% Center
L 33%
C 67%
Factuality
To view factuality data please Upgrade to Premium
