NewApp Monitoring is liveSign up free
AgentbeamBeta
Open-source package vulnerability · PyPI

CVE-2026-115419: Strands Agents SDK: Exposure of data element to wrong session in the A2A Server

highCVSS 4.0 8.6 · 3.1 8.1GHSA-4v6f-7vjh-6x4m

The Agent2Agent (A2A) server in the Strands Agents SDK shares one conversation history across all callers, whatever their `context_id`. An authenticated user can read another caller's messages and steer the model for other callers. Python `strands-agents` is fixed in 1.43.0 and TypeScript `@strands-agents/sdk` in 1.5.0. Any A2A server reachable by more than one user or tenant is affected.

Published Oct 23, 2026 · Analysis Nov 2, 2026

What an attacker can do

The A2A executor reused a single Agent object for every request. The agent's message history therefore belonged to the server, not to the A2A context. A caller with a different context_id still landed in the same history, so the model saw everyone's earlier turns.

An attacker needs only a valid login to the A2A endpoint (the CVSS vector says low privileges, network access, no user interaction). They can ask the agent to repeat or summarise earlier conversation and so read other tenants' prompts and tool results. They can also plant instructions in the shared history, which then shape answers for other callers. The PR notes the shared history also grew without bound.

Single-user deployments are not exposed in practice. The risk is multi-tenant setups, or one A2A server shared by several teams or agents.

Who is affected

Python: every strands-agents release that contains the multiagent.a2a module and is older than 1.43.0. TypeScript: @strands-agents/sdk older than 1.5.0. The flaw matters only if you run an A2A server and more than one user or tenant can reach it.

Check your installed version:

pip show strands-agents
npm ls @strands-agents/sdk

Search your code for A2AServer( and StrandsA2AExecutor to find servers you run.

strands-agents

Ecosystem
PyPI
Affected
< 1.43.0
Fixed in
1.43.0
Check
pip show strands-agents

@strands-agents/sdk

Ecosystem
npm
Affected
< 1.5.0
Fixed in
1.5.0
Check
npm ls @strands-agents/sdk

How to fix it

Upgrade strands-agents to 1.43.0 or @strands-agents/sdk to 1.5.0.

pip install --upgrade "strands-agents>=1.43.0"
npm install @strands-agents/sdk@^1.5.0

The fix isolates state per context_id. In Python, the PR adds an agent_factory argument: a callable that takes a context_id and returns an Agent. This is the recommended mode and lets different contexts run concurrently. Passing a single agent= still works but is deprecated, emits a DeprecationWarning and runs requests one at a time. An agent that already has a session_manager now raises ValueError. A new max_contexts argument (default 1000) evicts the least recently used context.

No workaround fully fixes older versions. Until you upgrade, do not expose an A2A server to more than one user or tenant. Run one instance per user or tenant.

How to tell if you were hit

There is no log signature in the advisory. Review A2A server access logs for one caller sending requests under many context_id values, or asking the agent to recall earlier conversation. Look in stored conversation history for instructions or content that no one in that context wrote. Also check which deployments run an A2A server shared by several tenants, since those are the ones to treat as exposed until upgraded.

Original advisory

From the GitHub Advisory Database (CC BY 4.0).

Summary Strands Agents SDK is an open-source SDK, available for Python (strands-agents) and TypeScript (@strands-agents/sdk), for building and running AI agents. An issue exists where, under certain circumstances, the Agent2Agent (A2A) server shares conversation history across callers with different context\_id values, allowing one caller to read and influence another caller's session.

Impact The A2A server did not properly isolate conversation state between callers. In multi-tenant deployments, an authenticated remote user could access another caller's conversation history or influence model behavior for other callers (CWE-488: Unintended Disclosure of Data Element to Wrong Session). This affects any deployment where an A2A server is reachable by more than one user or tenant.

Impacted versions: strands-agents (Python): all versions containing the multiagent.a2a module, < 1.43.0 @strands-agents/sdk (TypeScript): < 1.5.0

Patches This issue has been addressed in strands-agents version 1.43.0 and @strands-agents/sdk version 1.5.0. The A2A server now uses an agent factory so conversation state is isolated per A2A context\_id. We recommend upgrading to the latest version and ensuring any forked or derivative code is patched to incorporate the new fixes.

Workarounds There is no workaround that fully addresses this issue in affected versions. Until you upgrade, do not expose an A2A server to more than one user or tenant. If multiple callers need access, run a separate A2A server instance for each user or tenant.

References If you have any questions or comments about this advisory, we ask that you contact AWS Security via our vulnerability reporting page or directly via email to [[email protected]](mailto:[email protected]). Please do not create a public GitHub issue.

Questions

Does this affect Strands agents that do not use the A2A server?

No. The flaw is in the multiagent.a2a server path. An agent you run directly or through another interface does not share history this way.

Does the attacker need an account?

Yes. The advisory describes an authenticated remote user, and the CVSS vector lists low privileges required. No user interaction is needed.

Is there a workaround if I cannot upgrade?

Only isolation. The advisory says no workaround fully fixes it. Do not expose one A2A server to several users or tenants, and run a separate instance per user or tenant.

Is CVE-2026-115419 exploited in the wild?

No source reports exploitation and it is not in CISA's KEV catalog. The EPSS score is 0.0043.

References

How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. The analysis is written by an AI model from the sources listed above and checked by automated rules before it is published. Check versions against the vendor's advisory before you act. Methodology

More PyPI vulnerabilities

All PyPI entries