Intel

AIKIDO-2026-480854

anthropic is vulnerable to Improper Authorization

Improper Authorization Pre-CVE
Found by Aikido Intel before public disclosure or CVE publication.
Published 6 days ago

61

Medium Risk

This Affects:

PYTHONanthropic
0.103.0 - 0.116.0
Fixed in 0.117.0
Are you affected? Scan for Free

TL;DR

The beta Sessions SessionToolRunner dispatched every agent.tool_use and agent.custom_tool_use event immediately on arrival and ignored the server-supplied evaluated_permission verdict. A tool configured with an ask or always_ask permission policy therefore executed before, or entirely without, the user confirmation the policy required. The fix routes tool-call events through a confirmation gate that holds ask-gated calls until a matching user.tool_confirmation allow verdict arrives, never runs denied calls, and fails closed on unrecognized permissions or verdicts. This restores the human-in-the-loop approval the permission policy is meant to enforce.

Who does this affect?

You are affected if you are using a version that falls within the vulnerable range and you run the beta Sessions SessionToolRunner with tools configured under an ask or always_ask permission policy.

Background info

anthropic is vulnerable to Improper Authorization in versions 0.103.0 - 0.116.0.

How to fix this

Upgrade the anthropic library to the patch version.