Scope before action
Experiments and security research run in authorised environments. Tools are constrained to the resources and responsibilities of their role.
Decisions leave evidence
Persistent context should make a decision easier to inspect: what was known, what changed, and who authorised it. An agent reporting success is one source of evidence; verification is another step.
People remain accountable
The lab develops mechanisms for approval, interruption and recovery. A system’s autonomy does not remove the operator’s responsibility.
Research stays open to correction
Prototypes, concepts and deployed systems are identified separately. We publish observations and revise our conclusions as the work develops.