Research · curated 30 Aug 2026
Black Hat 2026: AI Agent Framework Flaws Expose Secrets
First reported checkpoint.com
Coverage timeline
Single-source research — first reported, latest, and curated coincide.
Why it matters
AI agent frameworks that much of the industry builds on carry old-school exploitable bugs in their serialization, caching and parsers, meaning attackers can compromise the plumbing beneath the model rather than beating the model itself.
Check Point researchers Shahar Tal and Yarden Porat presented at Black Hat 2026 an audit of major AI agent frameworks — LangChain, CrewAI, Microsoft Agent Framework and Google's ADK — uncovering 21 findings across eight codebases including 12 CVEs. The flaws are classic vulnerability classes (unsafe deserialization, SSRF, SQL injection, sandbox escape, arbitrary file read, memory corruption, PDF-parser RCE) reachable via post-injection exploitation, where attacker-controlled content poisons an agent's memory and triggers the framework's own internal plumbing to steal credentials and data without calling dangerous functions directly.