CVSS 4.0
2.3 lowCVSS:4.0/AV:N/AC:L/AT:P/PR:L/UI:N/VC:L/VI:N/VA:N/SC:N/SI:N/SA:NLow (2.3) in pydantic pydantic ai. EPSS 0.6% estimated probability of exploitation activity in the next 30 days. Fixed in pydantic-ai 1.107.6, 2.44.0.
CVSS:4.0/AV:N/AC:L/AT:P/PR:L/UI:N/VC:L/VI:N/VA:N/SC:N/SI:N/SA:NCVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:L/I:N/A:Npydantic-ai (PyPI): fixed in 1.107.6, 2.44.0pydantic-ai-slim (PyPI): fixed in 1.107.6, 2.44.0Explain CVE-2026-107291 for an engineer deciding whether to act. Use https://cve.bz/api/public/cves/CVE-2026-107291?via=ask-ai-copy and https://cve.bz/cve/CVE-2026-107291?via=ask-ai-copy as the source of truth: what is affected and which versions fix it, whether it is in CISA KEV, public exploits, and its EPSS (an internet-wide 30-day estimate of exploitation activity, not a forecast for my systems). If they give no fixed versions, use the CVE record https://cveawg.mitre.org/api/cve/CVE-2026-107291 or the vendor advisory, not third-party mirrors. Tell me how to check whether I am affected and how to verify a fix. Cite the sources and say where they are uncertain.Opens a third-party assistant; its answer is its own, not cve.bz's. The prompt holds only this CVE's id and public links.
From the NVD's CPE matches.
| Product | Affected |
|---|---|
| pydantic:pydantic_ai | from 0.3.4 before 1.107.4 |
| pydantic:pydantic_ai | from 2.0.0 before 2.44.0 |
| Advisory | Ecosystem | Package | Affected | Fixed in |
|---|---|---|---|---|
| GHSA-4x9p-g9wm-8q7f | PyPI | pydantic-ai | >=0.3.4 <1.107.6 | 1.107.6 |
| GHSA-4x9p-g9wm-8q7f | PyPI | pydantic-ai | >=2.0.0b1 <2.44.0 | 2.44.0 |
| GHSA-4x9p-g9wm-8q7f | PyPI | pydantic-ai-slim | >=0.3.4 <1.107.6 | 1.107.6 |
| GHSA-4x9p-g9wm-8q7f | PyPI | pydantic-ai-slim | >=2.0.0b1 <2.44.0 | 2.44.0 |
| PYSEC-2026-4275 | PyPI | pydantic-ai | >=0.3.4 <1.107.6 >=2.0.0b1 <2.44.0 | 1.107.6, 2.44.0 |
| PYSEC-2026-4284 | PyPI | pydantic-ai-slim | >=0.3.4 <1.107.6 >=2.0.0b1 <2.44.0 | 1.107.6, 2.44.0 |
Red Hat rates this Moderate.
A moderate information disclosure flaw was found in the Pydantic AI agent framework, which is shipped in Red Hat Enterprise Linux AI and Lightspeed Core products. When Pydantic AI's OpenTelemetry instrumentation is configured with InstrumentationSettings(include_content=False) to exclude sensitive message content from exported telemetry, several other parts of the same spans are not redacted: exception events record full exception messages and stack traces (including tool retry feedback that can quote tool arguments and results, and provider error response bodies), span error status descriptions repeat the exception message, and the model_request_parameters attribute serializes the agent's runtime instructions and the prompted structured-output template. Anyone with read access to the configured telemetry backend, such as a third-party observability vendor or a shared backend whose audience is broader than the agent's own data, can read this confidential agent content. The flaw does not grant any new access to the agent or its data beyond visibility of the exported traces, and deployments that do not enable include_content=False are not affected; instructions are exported on every successful run, while the exception paths additionally require a tool to raise or a model, embedding, or image generation request to fail. Red Hat rates this flaw as moderate severity (CVSS 3.1 score 4.3), whereas the upstream advisory rates it low severity (CVSS 4.0 score 2.3); this difference reflects Red Hat's assessment of the confidentiality impact on products that ship the affected package. For the affected Red Hat Enterprise Linux AI and Lightspeed Core components, the fix is currently deferred because the impact is moderate and the CVSS score is below 7.0, which does not meet the bar for prioritizing an update for this issue. The residual risk of deferring is that sensitive agent content continues to be exported to, and readable by readers of, telemetry backends populated by affected deployments that use include_content=False.
A workaround is available for deployments that cannot upgrade immediately: in the telemetry pipeline, scrub the exception.message and exception.stacktrace event attributes, the error status description, and the instruction parts and prompted_output_template within model_request_parameters on spans emitted by Pydantic AI, for example with an OpenTelemetry Collector processor. Setting include_model_request_parameters=False alongside include_content=False also stops the export of the instructions and the output template, at the cost of the rest of that attribute. Deployments that rely on include_content=False should additionally review previously exported traces for exception events, error status descriptions, and model_request_parameters. The issue is fixed upstream in pydantic-ai 1.107.6 (1.x series) and 2.44.0 (2.x series); in the fixed versions, with include_content=False, exception events record only the exception type, error statuses carry no description, and model_request_parameters is exported without instruction content or the output template.
| Product | Package | State | Advisory |
|---|---|---|---|
| Lightspeed Core | lightspeed-core/lightspeed-stack-rhel9 | Fix deferred | |
| Red Hat Enterprise Linux AI (RHEL AI) 3 | rhelai3/bootc-cuda-rhel9 | Fix deferred | |
| Red Hat Enterprise Linux AI (RHEL AI) 3 | rhelai3/bootc-gaudi-rhel9 | Fix deferred | |
| Red Hat Enterprise Linux AI (RHEL AI) 3 | rhelai3/bootc-rocm-rhel9 | Fix deferred | |
| Red Hat Enterprise Linux AI (RHEL AI) 3 | rhelai3/disk-image-cuda-rhel9 | Fix deferred |
Contains Red Hat Security Data © Red Hat, Inc., licensed under CC BY 4.0. Source: https://access.redhat.com/security/data. Modified: reformatted. Red Hat's page for CVE-2026-107291.
The attack patterns MITRE's CAPEC relates to this record's weaknesses, as MITRE names them. Credits.