Call It the Prompt Privilege

The federal government has discovered a new way to hide how it makes policy. Call it the prompt privilege: a claim that any input fed to an AI model during agency deliberations is exempt from public disclosure, on par with a draft memo or pre-decisional email. The category does not exist in FOIA case law. Agencies are trying to invent it.

WIRED's reporting on HUD's FOIA response shows the first test. A document labeled "GPT defined Econ Analysis approach 11 10 25.docx" was withheld because it was classed as a "deliberative AI input" — a phrase invented for the occasion. The pattern is portable. A team uses AI to draft a regulatory analysis, flags rules for elimination, or runs a GPT-defined economic model. The prompt becomes a deliverable. When a requester asks for the trail, the agency labels the prompt a deliberative AI input and withholds it. The exemption is bounded only by what agencies can credibly assert.

The repeated move: bounded disclosure rules expand through category invention, then harden into precedent by default. Once a court accepts the category, every agency that adopts AI inherits the same shield. The case that loses on deliberative process today is the case that defines the privilege for every other department tomorrow.

What changes: if the category holds, prompts join the protected inner circle of federal policymaking. If it does not, FOIA rules finally catch up with the tools agencies already use.