DHS AI policy keeps its ban on mass tracking but one officer can waive it

The US Department of Homeland Security (DHS) released a new AI policy for its use of AI. It started in September. The policy keeps the ban on mass monitoring of persons. But a new provision lets the chief AI officer waive each requirement, and this includes banned uses. Two policy experts tell that the waiver makes the ban weak.

What changed

The ban on mass surveillance and tracking stays, with small changes. The chief AI officer can waive each rule in the directive. Tom Bowman of the Center for Democracy and Technology tells that the waiver has no inspection by the privacy or civil rights offices, no notice to Congress and no end date.

Other points

The policy removes language on equity and on underserved communities. The chief privacy officer must at this time make sure that AI “complies” with the guardrails. Before, the text said that AI “does not erode” privacy. DHS must also share its AI models and publish its AI code as open-source software.

What is not known

In September, ICE gave a $125 million contract to Thomson Reuters Special Services to monitor 1 million persons. An ICE spokesperson tells that ICE will examine each use of AI in the contract against the policy. The source names no waiver that DHS has used.

Insight: The off switch is with the user of the tool

In October 2026, three US agencies told AI companies to use external monitors. They also told them to write who can stop a test. DHS gives the stop authority to one internal officer. This reads as a rule that the same office can lift, and the ICE contract can have no external check. Leaders who give AI to government must know that limits can change.