Definition
A flaw in an AI tool that fetches web content, where the tool can be tricked into requesting internal network services or cloud metadata that should never be reachable from outside. Agent tools that 'browse the web' or connect to integrations are the classic entry points for this attack in AI systems.
Why it matters
SSRF in an agent tool can leak cloud credentials or internal responses, giving attackers a bridge from a public-facing AI feature into the organisation's private infrastructure.