Q: What is the main principle of the US air traffic control AI policy? A: The US air traffic control AI policy centers on human-led systems where AI supports controllers but does not replace them. Transportation Secretary Sean Duffy emphasized that AI is a tool and people stay in charge. Q: How will human-in-the-loop requirements affect air traffic control operations? A: Expect requirements proving a human can override, pause, or reject any AI output at any time, with interfaces that include override, pause, and revert controls. Systems should also maintain logs showing when AI suggestions were accepted or rejected for traceability. Q: What immediate actions should airports, airlines, and vendors take to prepare? A: Follow a 90–180 day action plan that sets governance, names an accountable executive, and establishes a cross-team review board for decisions, testing, deployment, and incident response. The plan also recommends mapping use cases to risk, running high-fidelity simulations and historical replays, and training controllers on model limits in line with US air traffic control AI policy. Q: What technical guardrails should be built into AI tools for air traffic control? A: Guardrails should include human override, pause, and revert buttons, explainable summaries of why a model suggests a path, confidence bands that default to safe behavior when confidence is low, and rate limits to prevent alarm fatigue. Systems should also pin model versions, have rollback plans, and log inputs, prompts, outputs, and overrides for audits. Q: How will training and certification for controllers change under the policy? A: Controllers will need training on model strengths, known failure modes, and escalation steps, with checkrides or certifications tied to each AI tool and periodic refreshers as models update. Training should also include override drills and “AI-out” contingency modes to keep skills current. Q: What security and data protections are recommended before deploying AI? A: Policies will likely mandate encryption, strict access controls, air-gapped paths where needed, and zero-trust principles to reduce tampering risks. Organizations should inventory data flows, isolate models from external networks, review third-party code paths, and log model access and changes for audits. Q: Which metrics should be monitored to ensure AI safety and performance? A: Track safety metrics like loss-of-separation rates, go-arounds, and pilot deviation links alongside operational measures such as delay minutes saved and reroute efficiency. Also monitor human factors (controller workload and trust), reliability (latency and uptime), and fairness across regions, traffic peaks, and weather regimes. Q: How should agencies and vendors engage unions and frontline experts during AI adoption? A: Build trust early by sharing test plans, running side-by-side trials where AI only advises, letting controllers veto risky features, and acting on frontline feedback each sprint. Commit in writing that AI will not replace controller roles, create new skill paths and paid training time, and involve union representatives in governance to align with the US air traffic control AI policy.