[
https://issues.apache.org/jira/browse/CAMEL-24607?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Work on CAMEL-24607 started by Andrea Cosentino.
------------------------------------------------
> camel-langchain4j-agent-api: CodeInjectionGuardrail blocks legitimate input
> by counting duplicate same-type matches
> -------------------------------------------------------------------------------------------------------------------
>
> Key: CAMEL-24607
> URL: https://issues.apache.org/jira/browse/CAMEL-24607
> Project: Camel
> Issue Type: Bug
> Components: camel-langchain4j-agent
> Reporter: Andrea Cosentino
> Assignee: Andrea Cosentino
> Priority: Major
> Fix For: 4.23.0
>
>
> CodeInjectionGuardrail.validate() (around lines 221-246) collects matched
> injection patterns into a List<InjectionType> without de-duplication, then in
> non-strict mode blocks the input when detected.size() >= 2. The inline
> comment states the intent is to require multiple DIFFERENT types, but because
> several patterns share the same InjectionType (4 SHELL_COMMAND, 4
> SQL_INJECTION, 3 JAVASCRIPT, 2 HTML_XSS, 2 TEMPLATE_INJECTION), a single
> input matching two patterns of ONE type is blocked.
> Concretely, a normal templating question such as "how do I use {{name}} and
> ${value} in my template?" matches both TEMPLATE_INJECTION patterns and is
> rejected as a false positive (the message even prints the duplicated list,
> e.g. [TEMPLATE_INJECTION, TEMPLATE_INJECTION]). This is reachable via the
> default CodeInjectionGuardrail and Guardrails.strictInputGuardrails().
> The sibling PromptInjectionGuardrail de-duplicates correctly. Fix: only add a
> type when not already present (if (!detected.contains(pattern.getType()))).
> _Reported by an automated functional audit._
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
