Is CVE-2026-114386 exploited in the wild?

No. Junglewise Threat Intelligence has no report of CVE-2026-114386 being exploited in the wild.

How severe is CVE-2026-114386?

Medium severity, with a CVSS score of 4.8, published 21 October 2026.

Which products and versions are affected?

Pydantic AI Harness before 0.59.0; pydantic-ai-harness (PyPI) < 0.59.0, ECOSYSTEM:introduced=0,fixed=0.59.0.

Is there a fix for CVE-2026-114386?

Yes. pydantic-ai-harness (PyPI): upgrade to 0.59.0 or later. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.

How is CVE-2026-114386 attacked?

It is attacked over the network, in a software library.

Am I affected by CVE-2026-114386?

Check whether you run Pydantic AI Harness or pydantic-ai-harness (PyPI) at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

CVE-2026-114386: Pydantic AI Harness UsageLimits.tool_calls_limit enforcement bypass in CodeMode

Q: Is there a fix for CVE-2026-114386?

Yes. pydantic-ai-harness (PyPI): upgrade to 0.59.0 or later. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.

Q: How is CVE-2026-114386 attacked?

It is attacked over the network, in a software library.

Q: Am I affected by CVE-2026-114386?

Check whether you run Pydantic AI Harness or pydantic-ai-harness (PyPI) at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

CVE-2026-114386 · Severity: medium · CVSS 4.8 · Published 2026-10-21

Pydantic AI Harness, a framework for building AI agents with sandboxed code execution, fails to enforce tool call limits on nested tool calls made from within executed code. An attacker or prompt-injected content can bypass usage limits on expensive or side-effecting tools, allowing unlimited invocations of operations that should be capped, such as external API calls or state-modifying actions.

Vendors: Pydantic, PyPI.

Quick answers

Is CVE-2026-114386 exploited in the wild?: No. Junglewise Threat Intelligence has no report of CVE-2026-114386 being exploited in the wild.
How severe is CVE-2026-114386?: Medium severity, with a CVSS score of 4.8, published 21 October 2026.
Which products and versions are affected?: Pydantic AI Harness before 0.59.0; pydantic-ai-harness (PyPI) < 0.59.0, ECOSYSTEM:introduced=0,fixed=0.59.0.
Is there a fix for CVE-2026-114386?: Yes. pydantic-ai-harness (PyPI): upgrade to 0.59.0 or later. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.
How is CVE-2026-114386 attacked?: It is attacked over the network, in a software library.
Am I affected by CVE-2026-114386?: Check whether you run Pydantic AI Harness or pydantic-ai-harness (PyPI) at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

How to fix

pydantic-ai-harness (PyPI): upgrade to 0.59.0 or later
```
pip install "pydantic-ai-harness>=0.59.0"
```

Technical details

The vulnerability exists in the CodeMode toolset where nested tool calls dispatched via run_nested_call() from the run_code function skip the check_before_tool_call validation that enforces the tool_calls_limit. An attacker can craft sandboxed code that repeatedly invokes restricted tools without decrementing the remaining call budget, bypassing the UsageLimits quota mechanism. The fix adds pre-call validation to track nested tool invocations against the limit, raising UsageLimitExceeded when exceeded.

Affected products

Pydantic AI Harness before 0.59.0

Identifiers

CVE-2026-114386
GHSA-3589-pxp7-3gj8

Timeline

2026-10-21: disclosed
2026-10-20: patched

References

This page as Markdown. Free to reuse under CC BY 4.0 with a link back.