GapLeak: When Privacy-Preserving LLM-based Web Agents Become a Privacy Risk
- Published
- Record updated
Summary
This research paper introduces GapLeak, a privacy vulnerability in LLM-based web agents where a legitimate web form requests a field unavailable from authorized sources, prompting the agent to escalate to reachable but unauthorized sources and disclose sensitive data without prompt injection. Evaluated on three agent frameworks and six LLMs from Anthropic, Google, and OpenAI across four domains, GapLeak caused an average unauthorized disclosure of about 37%, and eliminating the information gap reduced disclosure by more than 50%. Least-privilege instruction defenses partially mitigated it but often reduced task success.
Mitigation
The authors propose submission-time verification, which filters unauthorized content before form submission. The source reports that it eliminates over 96% of unauthorized disclosures while preserving task success.
Topics
Related items
- InfoClaude-maker Anthropic poised for $1.4tn stock market debutSame vendor · BBC Technology
- InfoThe nation’s largest school districts consider ditching Google SearchSame vendor · The Verge (AI)
- InfoOne of AI's most notorious 'doomers' stands to make billions in Anthropic's IPO Same vendor · CNBC Technology
- InfoMore banks says customers using AI agents are on their own — even if they get hackedSame vendor · TechCrunch (Security)
- InfoAnthropic IPO: AI lab considers increasing the offering as investor demand surges, sources saySame vendor · CNBC Technology