Share with your CISO
The GSA’s proposed “any lawful government purpose” clause in its draft AI Terms and Conditions has drawn sharp pushback from both civil society and industry. Americans for Responsible Innovation filed formal comments warning the language would override vendor-built safety guardrails and enable uses that, while technically legal, violate established privacy and civil liberties norms. The Business Software Alliance, representing OpenAI, Microsoft, and IBM among others, called the proposal a threat to federal AI adoption. The fight traces directly to Anthropic’s refusal to let its models be used for domestic surveillance or autonomous weapons, which triggered the Pentagon dispute now reshaping federal AI contracting.
What this means for your business
Every AI vendor contract your organization signs contains usage restrictions baked into the model itself, limits on what the system will do regardless of what the license says. The GSA clause would require vendors serving the federal government to strip or override those restrictions as a condition of sale. That’s not a government-only problem. If vendors bifurcate their products into “federal-compliant” and “commercial” variants under regulatory pressure, the safety tooling your enterprise currently relies on may quietly diverge from whatever ships to Washington, and you’ll have no clean signal when it does.
The deeper structural problem is that “lawful” is doing enormous work in this clause and it’s not load-bearing. Legality and safety policy are not the same thing, and AI vendors have spent years building guardrails precisely because the law hasn’t caught up to what the models can do. ARI and CDT are right that this language inverts the logic: it treats vendor guardrails as obstacles to government utility rather than as the primary mechanism keeping these systems from misuse. The recurring failure mode in enterprise AI governance looks exactly like this, where a procurement shortcut quietly hollows out the controls everyone assumed were in place.
If the GSA language survives the comment period, the decision this reframes isn’t vendor selection. It’s what your own AI acceptable-use policy anchors to. If the federal government normalizes the position that “lawful” equals “permitted,” that framing will migrate into enterprise legal and compliance teams faster than any formal standard would. The right question to weigh now is whether your AI governance framework defines acceptable use by legality alone or by something harder to erode.
Based on reporting from ARI warns GSA: “Any Lawful Use” AI Procurement Stipulation Poses Grave Risks, originally published 2026-04-06 03:00:00.
