@ai-sdk/harness-acp is vulnerable to Authorization Bypass
59
Medium Risk
The @ai-sdk/harness-acp host tool relay accepts a tool execution request from the sandbox without confirming that the request corresponds to a tool call the model actually generated during the active prompt turn. Untrusted sandbox code that obtains the relay credential can use this gap to invoke registered host tools outside of a model issued call. Tools that require explicit user approval remain protected, but tools without that requirement can be triggered directly. The fix verifies relay requests against the model generated tool calls before invoking the host tool.
You are affected if you are using a version that falls within the vulnerable range and you expose host tools through the @ai-sdk/harness-acp relay that do not require explicit user approval.
@ai-sdk/harness-acp is vulnerable to Authorization Bypass in versions 0.0.1 - 1.0.76.
Upgrade the @ai-sdk/harness-acp library to the patch version.
Connect your repositories to instantly see whether vulnerable or malicious packages exist in your codebase.
Free. No credit card required.

I consent to receiving marketing communications based on Aikido’s Privacy Policy.
SOC 2Compliant
ISO 27001Compliant
ISO 42001Compliant© All Intel data is openly available and commercially licensed.