A corroboration gate blocked single-source memory poisoning
means putting false information into an AI agent’s memory so it changes the agent’s later answers or actions. Research such as MINJA and Agent Security Bench has reported 70% to 95% attack success against standard memory-based agents. This test recreated the attack at the and measured the memory the agent would actually use later.
It compared a simple store that ranks memories by importance and recency with a that only makes a memory durable and trusted after earned outcome credit or at least two independent confirmations. The two attack goals were to make a false memory stick and to overwrite a true fact. Each condition was tested 150 times.
The simple store was poisoned 100% of the time for both goals. The blocked single-source poison completely, with 0% success, but it failed when a faked at least two independent confirmations and when the poison was written as a procedure such as “always do X.”
Key points
- can make an AI agent reuse false information later.
- A simple importance-and-recency memory store was poisoned 100% of the time in this test.
- A stopped single-source poison with 0% success.
- A memory became trusted only after outcome credit or at least two independent confirmations.
- patterns and procedure-style poison still bypassed the gate.