CVE-2026-107291: CWE-212: Improper Removal of Sensitive Information Before Storage or Transfer in pydantic pydantic-ai

Pydantic AI, a Python agent framework, improperly removes sensitive information before telemetry export when OpenTelemetry instrumentation is configured with InstrumentationSettings(include_content=false). Despite this setting, sensitive content such as exception.message, exception.stacktrace, error status descriptions, and model_request_parameters can be exported to the telemetry backend. This includes tool feedback, provider error bodies, runtime instructions, and structured-output templates. The vulnerability affects versions >=0.3.4 <1.107.6 and >=2.0.0b1 <2.44.0 and is resolved in versions 1.107.6 and 2.44.0. The issue does not provide new access to agent data and only impacts deployments using the include_content=false setting.

Sensitive agent content may be exposed to readers of the configured telemetry backend, including exception details and runtime instructions, potentially leading to unintended data disclosure. However, this does not grant new access to agent data beyond what is already accessible through telemetry. Deployments not using include_content=false are not affected by this bypass.

Upgrade to pydantic-ai versions 1.107.6 or later, or 2.44.0 or later, where this issue is fixed. If upgrading is not immediately possible, review telemetry backend access controls to limit exposure of sensitive data. Note that this vulnerability only affects deployments using OpenTelemetry instrumentation with InstrumentationSettings(include_content=false).