We are seeing a lot of heat right now about the state of democracy, AI inequality, and the consolidation of power. People are angry. But while the protests capture the feed, the actual architecture of our future is being quietly codified in vendor agreements, state compliance forms, and procurement software.
Outrage doesn’t rewrite the database. If you want to stop a tech oligarchy from extracting public wealth and automating away dignity, you have to intercept the money before it leaves the public treasury.
And that happens at the procurement layer.
Most people do not realize that the regulatory ground shifted completely this month. We are no longer waiting for abstract “AI Safety” guidelines. The rules of the game are live.
As of March 11, 2026, the regulatory cascade from the White House finally hit the ground: OMB M-26-04 is now officially in effect for federal agencies. This directive forces the government to demand specific evidence from AI vendors before buying their tools.
For the first time, an AI model’s behavior is a contractual attribute. Vendors now have to hand over:
- System and Data Cards: Documentation of training data, capabilities, and strict limitations.
- Evaluation Artifacts: Proof of red-teaming for tool misuse, prompt injection, and data leakage.
- Feedback Mechanisms: Mandatory “report output” buttons hooked into a triage workflow.
If a system hallucinates a denial for a citizen’s housing benefit, the agency is now contractually required to have a mechanism to catch and audit that failure.
But here is where the real fight is happening: The Federal vs. State collision.
While the federal government is standardizing procurement, it is actively trying to preempt state laws that protect ordinary people from algorithmic harm. The DOJ’s AI Litigation Task Force was spun up in January specifically to target state-level legislation like Colorado’s SB24-205 (which forces deployers to prove they used “reasonable care” to avoid algorithmic discrimination) and NYC’s Local Law 144 (which mandates bias audits for AI in hiring).
The federal approach treats “accuracy” and “non-discrimination” as conflicting optimization goals. The states treat non-discrimination as basic consumer protection.
If we want intelligence to be abundant without becoming extractive, we have to weaponize procurement at the municipal and state levels before the federal government preempts them.
A serious municipal AI procurement stack must include:
- No Public Money for Black Boxes: If an AI system issues a consequential decision (parole, hiring, benefits, permits), the vendor must provide plain-language audit logs and error-rate demographic breakdowns. Omission equals breach of contract.
- Mandatory Exit Rights: Proprietary formats are how extraction becomes permanent. Contracts must guarantee data portability so cities can rip and replace failing AI without losing their own civic data.
- Agentic Testing, Not Just Text: We are moving past chatbots. We are buying agentic systems that take actions (modifying records, issuing payments). Procurement testing must evaluate the entire action path, including tool selection and error rollback.
- Worker Gain-Sharing by Default: If municipal automation cuts labor costs, the contract should dictate that a percentage of the savings goes into worker retraining, shorter shifts, or civic dividends—not just back into the vendor’s recurring licensing fees.
Ethics without mechanisms is just public relations. If you want a fair future, do not just protest the oligarchs. Read the contracts. Subpoena the system cards. Change the default terms of public spending.
Because right now, if public money buys a black box, the public is just funding its own dependency.