Topic

AI agent memory security

Security of the persistent memory stores that AI agents read from to shape their behavior, including tampering, injection and provenance of stored experience.

Current clusters

security1 publisher

Coding agents are writing API keys into plain-text memory, Vectorize CEO says

Vectorize CEO Chris Latimer says coding agents he used wrote API keys, credentials and sensitive documents into plain-text long-term memory. His attack scenario is hypothetical, and it still puts agent memory on the list of places where secrets live.

Publishers:helpnetsecurity.com

Reality

Evidence30
Adoption: Insufficient
Hype gap+30
Incentives: Insufficient
Confidence35

build1 publisher

A hand-written memory graph matched 600 rewarded trials at 1.00 accuracy

The substrate reads keys, traces, signs and strengths and takes all four at face value, so write access to the graph is authorship of behavior, including a transplant that scored 0.17 against 0.33 chance.

Publishers:dev.to

Reality

Evidence35
Adoption: Insufficient
Hype gap+20
Incentives: Insufficient
Confidence35