1 of 1
Story summary
- Agentic AI-enabled browsers, including Perplexity's Comet and OpenAI's ChatGPT Atlas, raise new security risks.
- Brave reports show vulnerabilities in Comet that enable malicious websites to hijack the AI assistant through indirect prompt injections.
- Experts advise treating webpage content as untrusted, conducting independent alignment checks, and obtaining user consent for sensitive actions.
- Users should limit permissions and audit stored memories to curb data trails and strengthen privacy.
