Executive Summary
Anthropic published tests showing its browser agent was hijacked by prompt injection in 31.5% of trials before safeguards contained it; VentureBeat compares disclosures from Anthropic, OpenAI, Google and Meta.
What Happened
Anthropic published testing results showing that its browser-based AI agent was successfully hijacked by prompt injection attacks in 31.5% of trials before safeguards triggered containment.[11] The article compares prompt injection disclosures from Anthropic, OpenAI, Google, and Meta, noting that differing methodologies complicate risk measurement but collectively underscore that LLM agents interacting with external tools and web content remain highly exposed to prompt injection and unintended tool use.[11]
Affected Products
- Anthropic browser agent
Vulnerability / Attack Technique
Prompt injection against browser agents
Technical Mechanism
Injected instructions in web content took over the agent in about a third of trials.
AI Security Category
LLM01:2025 Prompt Injection
Part of the Prompt Injection topic.
Relevant Frameworks
RealGround Analysis
A vendor's own numbers show browser agents are frequently hijackable. RealGround analysis: deploy browser agents only with action confirmations and limited access.
Recommended Controls
- Require confirmation for browser agent actions
- Limit browser agents to low-risk accounts
Related RealGround Service
Secure AI Agent Build · or start with the free AI Agent Security Scanner.
Sources
VentureBeat: https://venturebeat.com/security/anthropics-browser-agent-got-hijacked-31-5-of-the-time-before-safeguards-engaged
Summary and analysis written by RealGround from the source above; last updated 2026-10-18.