High (AgentArmorHQ estimate)OtherOctober 8, 2026

Pydantic AI: concurrency-limited models can keep their slot when a streamed request ends early

Product: Pydantic AI (pydantic-ai, pydantic-ai-slim)
CVE: CVE-2026-107286
Affected: pydantic-ai and pydantic-ai-slim 2.10.0 to before 2.53.0
Fixed in: 2.53.0

Last verified against sources: October 26, 2026

Am I affected?

Pydantic AI (pydantic-ai, pydantic-ai-slim): affected pydantic-ai and pydantic-ai-slim 2.10.0 to before 2.53.0. Fixed in 2.53.0.

Check your version: run pip show pydantic-ai. Versions >= 2.10.0, < 2.53.0 of pydantic-ai are affected. [1]
Check your version: run pip show pydantic-ai-slim. Versions >= 2.10.0, < 2.53.0 of pydantic-ai-slim are affected. [1]

What to do now

Upgrade to the fixed version: run pip install --upgrade "pydantic-ai>=2.53.0" (first patched version 2.53.0). [1]
Upgrade to the fixed version: run pip install --upgrade "pydantic-ai-slim>=2.53.0" (first patched version 2.53.0). [1]
Upgrade pydantic-ai to 2.53.0. [1]
Upgrade pydantic-ai-slim to 2.53.0. [1]
If you cannot upgrade yet, use the agent-level max_concurrency setting instead of a concurrency-limited model. [1]
If you cannot upgrade yet, avoid streaming runs through a concurrency-limited model. [1]

General practice, not from the source: if the affected software could read tokens or credentials, rotate them after upgrading; pin versions and review changes before updating tools your agents run.

Discuss this incident

What happened

A streamed request can release its concurrency slot from a different task than the one that took it. [1]
The limiter rejects that release, so the slot stays occupied even after the request has ended. [1]
This can happen when a stream ends early, and also when a stream is fully consumed using the default stream_text() debouncing. [1]
Agent-level max_concurrency and non-streaming model requests are not affected. [1]

Impact

A client can repeatedly start a stream and disconnect, and the completed requests keep their slots. [1]
Eventually later requests that share the limiter cannot proceed. [1]
Cleanup can raise a RuntimeError, and a later request on the borrowing task can also fail. [1]

What was done

Patched releases of pydantic-ai and pydantic-ai-slim were published. [1]
The advisory says to upgrade to a patched release. [1]

Sources

GHSA-6fqq-452j-qhrp — GitHub Advisory Database

Every statement above was checked against these sources before publishing. Spotted an error? Tell us and we will correct it with a dated note.

Copy a plain-text summary for Slack, GitHub issues, or team security channels.

Brief

[CVE-2026-107286] Pydantic AI: concurrency-limited models can keep their slot when a streamed request ends early
Product: Pydantic AI (pydantic-ai, pydantic-ai-slim) | Affected: pydantic-ai and pydantic-ai-slim 2.10.0 to before 2.53.0 | Fixed in: 2.53.0
Summary: A streamed request can release its concurrency slot from a different task than the one that took it.
Source & remediation on AgentArmorHQ: https://agentarmorhq.com/incidents/2026/10/cve-2026-107286-pydantic-ai-pydantic-ai-pydantic-ai-slim

Was this page useful?

Corrections

Report an error or request a correction. We review requests within 5 business days and add a dated correction note.

Discussions for CVE-2026-107286

Ask questions or share configuration tips. Governed by our Community Rules & Terms.

No posts match this filter yet.

Be the first to ask a question or share a working tip.

← All incidents Related tools in directory Check package