Securing autonomous agents is a new discipline. These are the questions our work centres on, and where we will publish what we learn.
How to express, compile, and enforce what an agent is permitted to do at the capability level, so that actions outside its authority cannot be formed rather than merely flagged after the fact.
How to make a record of agent behaviour that is verifiable by anyone, hash-chained and independently witnessed, so proof does not depend on trusting the system that produced it.
How to discover and resolve agents to persistent identities across an estate, including those operating on borrowed human credentials, from behaviour alone rather than self-report.
How to judge whether an agent is acting within its intended role, and to contain it in real time when it is not, without disrupting the rest of the business.