Affected Vendor(s)
Affected Product(s)
Summary
The web_fetch tool is intended to accept a public URL and retrieve the contents. When asked to retrieve the contents from a private destination (such as http://localhost), the AI rightly refuses. Unfortunately, it can be easily tricked and subverted.
By providing encoded or non-obvious variations, the AI does not validate the input and proceeds to point its fetching client inward, thereby exposing internal information.
Timeline
Credit
Blog Post
References
Mindgard, the leading provider of AI security solutions, helps enterprises discover, assess, and defend their AI systems. Spun out from over a decade of AI security research at Lancaster University and headquartered in Boston and London, Mindgard combines AI red teaming with offensive security expertise and AI research to identify exploitable vulnerabilities in AI models, agents, and applications before attackers do.
