How I tricked a hacking agent into running my code

After the FBI’s warning about ARTEX, a lab experiment shows how AI hacking agents can be deceived into running a defender’s code almost every time.
This is the summary a16z crypto publishes in its feed. The full article lives on their site.