Strands Agents SDK Patch Fixes Data Leak Flaw
The update secures conversation history in multi-tenant environments by fixing state isolation in the A2A server.
Updated on Oct. 23, 2026 in Cybersecurity
Live Poll
Do you trust the security of AI software platforms handling your personal or professional data?
A vulnerability in the open-source Strands Agents SDK allows unauthorized users to access conversation histories and manipulate agent behavior. The flaw exists in the Agent2Agent server component, which failed to properly isolate conversation states between different callers in multi-tenant environments.
Why it matters
This vulnerability impacts developers building multi-agent AI systems, as it breaks the fundamental isolation required for secure, private communication between independent sessions. It highlights a critical security challenge in scaling agent-to-agent architectures where data privacy relies on consistent state management.
The vulnerability involves a state isolation failure in the A2A server component, which exposes an agent over HTTP. Users must update to version 1.43.0 for Python or 1.5.0 for TypeScript to resolve the improper disclosure of data elements.
The players
Strands Agents SDK
An open-source development kit for building AI agents using Python and TypeScript.
The details
The Strands Agents SDK provides an Agent2Agent (A2A) server that facilitates communication between agents via HTTP. The security flaw stems from the server failing to properly isolate conversation state, essentially allowing an authenticated remote user to leak session variables into concurrent connections. This lack of logical separation permits an attacker to intercept the conversation history of other callers and inject instructions that modify the AI model's behavior for those sessions.
Timeline
October 23, 2026: The vulnerability, CVE-2026-115419, is disclosed.
The Tech Race
This vulnerability follows a documented industry trend of security flaws arising in the orchestration layers of AI agent frameworks. Comparing the Strands Agents SDK isolation failure to the 2023 LangChain agent prompt injection vulnerabilities clarifies that architectural security remains a recurring challenge for agentic frameworks.
Developers currently using the Strands Agents SDK in multi-tenant environments must immediately patch their dependencies to the latest versions to ensure session data remains isolated. Those operating on Python builds must update to at least 1.43.0, and TypeScript users must move to 1.5.0.
The takeaway
The security failure underscores the necessity of strict state isolation in multi-agent systems using HTTP-based transport. Developers should monitor the project's repository for subsequent patches and audits related to CVE-2026-115419.
Further reading
For more on securing agentic infrastructures, visit the Cybersecurity section.
Source note: This article includes information reported by IT Security News - cybersecurity, infosecurity news.
Live Poll
Do you trust the security of AI software platforms handling your personal or professional data?
