Executive Summary
Check Point found nearly a dozen flaws, some critical, in major AI-agent frameworks used for enterprise applications, letting attacker-controlled content influence orchestration, memory, state, routing and system instructions.
What Happened
Check Point researchers identified nearly a dozen flaws, including critical vulnerabilities, in major AI-agent frameworks used to build enterprise applications. The flaws can allow attacker-controlled content to influence orchestration, memory, state, routing, and system instructions; one reported issue involved insecure deserialization of untrusted checkpoint data that could enable code execution.
Vulnerability / Attack Technique
Framework flaws exposing agent state to untrusted content, including insecure deserialization
Technical Mechanism
Attacker-controlled content could steer orchestration, memory, routing and system instructions; one issue involved insecure deserialization of untrusted checkpoint data that could lead to code execution.
AI Security Category
LLM03:2025 Supply Chain
Part of the AI Supply Chain and AI SBOM topic.
Relevant Frameworks
RealGround Analysis
The flaws sit in the frameworks, so every app built on them inherits the risk. RealGround analysis: framework versions belong in the AI SBOM and in the patch process.
Recommended Controls
- Track agent framework versions in an AI SBOM and patch promptly
- Never deserialize checkpoint or state data from untrusted sources
- Keep untrusted content separate from system instructions and routing decisions
Related RealGround Service
Secure AI Agent Build · or start with the free AI Agent Security Scanner.
Sources
The Register: https://www.theregister.com/security/2026/08/05/prompt-injection-isnt-the-bug-ai-agent-frameworks-are/5283585
Summary and analysis written by RealGround from the source above; last updated 2026-10-17.