What happened
LangGraph's task-result caching mechanism generates cache keys using a weak hash function, which could allow crafted inputs to collide with cached results, potentially causing stale or incorrect cached task outputs to be served.
Why it matters
While LangGraph is widely deployed for building stateful LLM agents, this specific flaw is low severity (CVSS 3.1) and requires local access — worth tracking given the framework's popularity but not an urgent risk on its own.
Attack vector
The _freeze function in libs/langgraph/langgraph/_internal/_cache.py uses a weak/non-cryptographic hash for the default_cache_key argument in the Task Result Cache, which a local attacker could potentially exploit to cause cache-key collisions.
Affected systems
langchain-ai langgraph, up to version 1.2.4
Mitigation
Upgrade to a patched LangGraph release using a stronger hash function for cache keys, or supply a custom strong-hash cache-key function.