Note 17
Production that only agents touch
Why closing standing human access to production, and letting agents act on approved, temporary, audited access instead, makes production safer, not riskier.
- AI agents
- Security
- AWS
- Kubernetes
Standing access is the attack surface
Most production breaches and outages involve someone, or something holding someone's credentials, that had more access than the moment required. Administrator access that exists all the time, for many people, is the largest attack surface most companies have.
The alternative
People keep authority, not access. Agents do the operating, and every change they make passes through the same narrow door: a plan, an approval by policy or by a named person, access that exists only for that plan and only briefly, and a record of everything that happened.
- People decide. They write the policy, approve what it does not cover, and own the outcome.
- Agents act. They investigate, propose, carry out and verify, and they never approve themselves or each other.
- Nothing is standing. No human and no agent holds production write access between changes.
- Everything is recorded, in a form an auditor can follow from the alert to the approver.
Getting there
Nobody switches human access off on day one. Agents earn trust level by level, and human access is narrowed as each kind of operation moves to the agents, with an audited emergency path for people that remains. See the six levels of operations autonomy.
Available to enterprises in AI Factory.
Start a conversation
Bring your infrastructure
into focus.
Tell me what you are building, operating, or trying to improve.