← sources

How to check whether your AI agent's security guard actually blocks what it flags

source: reddit (reddit:AI_Agents)published 2026-10-28 11:48ingested 2026-10-29 02:36

retained text

A security dashboard showing "prompt injection detected" doesn't necessarily mean the agent was stopped from using that text.

StackOne's open-source Defender has a useful example. Its README says the \`blockHighRisk\` option defaults to false. Detection can flag risky content without making the result "not allowed" unless blocking is enabled.

Before relying on a guard around emails, documents or other tool results, check:

\- Does a high-risk finding actually produce a block verdict?

\- Does your application enforce that verdict, or just log it?

\- What happens if an optional LLM reviewer refuses to answer or returns malformed output?

\- Are you trusting a cleaned-up version of the text instead of checking the verdict?

Defender's README calls its text cleanup best-effort and recommends gating on the allow/block result. It also acknowledges that an LLM reviewer can itself be influenced by the text it's reviewing.

The practical test is the full path: submit a known malicious tool result, inspect the verdict, and confirm that your application stops it from reaching the agent. A warning in a log isn't the same as enforcement.

document fetched 2026-10-29 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1xhlvnb/how_to_check_whether_your_ai_agents_security