The AI directive on June 2 started the clock. Here is your action plan.
Former Department of Homeland Security cybersecurity leader, Paul Loeffler, explains why you need to govern your AI now, not after the fact. We will build the plan with you.
Why this matters now
The directive puts CISA in the lead for defending civilian federal systems while agencies move fast on AI. Upgrading for AI without governance widens your attack surface instead of shrinking it. That is the gap we close.
"*" indicates required fields
The 90-day action plan
Five steps to put your AI under governance in 90 days, whatever your seat. Wherever you are in the window, we will tailor the plan with you.
Phase 0.
Action: Build a live inventory of every AI agent, model, tool, and the identities behind them.
Why: You cannot report or defend what you cannot see. (EO Section 2)
For your lane: Federal reports to CISA. SLED answers grant reviewers. Contractors show their agency customer.
Step 1.
Action: Define how each agentic workflow should run before it goes live.
Why: Agentic systems are systems, not prompts. Approve them in advance, not after an incident.
For your lane: Every authority signs the same artifact, an approved design.
Step 2.
Action: Give each agent least-privilege, revocable authority tied to a named owner.
Why: What you cannot attribute, you cannot trust or shut off. (EO Section 4)
For your lane: Least privilege and instant revocation, in any environment.
Step 3.
Action: Compare live agent behavior against the approved design, continuously.
Why: AI changes without a code deploy. Catch drift the moment it happens.
For your lane: Detection reads against whatever baseline your lane requires.
Step 4.
Action: Produce the audit-ready record that proves your AI is governed.
Why: The directive points to the AI agent as a reportable asset. FedRAMP High is the trust baseline (expected June 2026).
For your lane: CDM and OMB for federal, GovRAMP and SLCGP for SLED, a FedRAMP service offering for contractors.