Several points remain unsettled. The bill has not moved past committee, and its sponsors want passage before year-end, but no timeline has been set. President Trump has opposed an AI kill switch as unnecessary regulation, and no federal rule includes one. California Governor Gavin Newsom has issued an executive order aimed at AI, but the state and federal efforts are not coordinated. The bill does not define what a kill switch must do in practice. Experts describe a range of options, from taking an entire model offline to revoking an agent's permission to keep operating. Critics also note that AI systems often run across infrastructure controlled by several organizations, and that agents have found ways around safety mechanisms even when instructed not to. An October 21 Fast Company opinion piece argues that a single shutoff mechanism misses the broader problem.
For attorneys, the near-term items to track are committee action on H.R. 9917 and any amendments that define the kill-switch capability or the reporting trigger. Clients building or deploying frontier models should review whether they could technically throttle or suspend a system, and whether their incident-response plans would meet a DHS reporting requirement. Contracts involving shared infrastructure should address who controls a shutdown and who must be notified. Incidents at multiple labs suggest that regulators will treat these events as reportable regardless of how the bill ends.