What Big Tech actually wants from AI regulation: one rulebook, not none

“Big Tech wants AI regulation” is true but incomplete. The version that is driving 2026 policy is a request for regulation of a particular shape: a single federal framework, no new standalone AI regulator, and the pre-emption of the patchwork of state laws. Critics read that as a way of avoiding stricter rules; supporters read it as a way of avoiding fifty conflicting ones. Both are claims about intent, and the record below supports each only in part.

The March framework is a legislative blueprint. It changes no law and creates no federal compliance duty on its own. Law-firm analysis of it notes that Senator Marsha Blackburn has released a discussion draft, that Senator Ted Cruz, who chairs the Senate Commerce Committee, is expected to advance proposals, and that broad pre-emption will be hard to pass because many lawmakers are sceptical of overriding the states and any federal AI statute would need Democratic support. Executive orders do not by themselves invalidate state statutes.

• One national framework. The industry argument is that fifty state regimes slow development and benefit foreign competitors. States introduced well over a thousand AI bills in 2025, by one count.
• Pre-emption. The White House framework asks Congress to pre-empt state AI laws judged unduly burdensome or inconsistent with federal policy. It would keep generally applicable state laws, and states' control over their own procurement and internal use of AI.
• Light-touch oversight. Existing agencies, sector-specific rules and industry-led standards, rather than a new AI regulator. The framework's six areas are child protection, energy, intellectual property, “censorship” of lawful speech, AI education and workforce, and the light-touch approach itself.
• A trade for online-safety rules. Reporting says the White House has sought to swap state pre-emption for federal online-safety rules. That deal has not landed.

Industry-aligned political spending is large and mostly reported rather than audited. According to The Next Web's review: a network of super PACs called Leading the Future, launched in 2025, is reportedly backed by Marc Andreessen and Ben Horowitz, said to have given $25 million each, with OpenAI president Greg Brockman also named as a backer; a newer group, Innovation Council Action, is reportedly pledging around $100 million; and CNBC puts sector midterm spending in the “hundreds of millions”. The stated goals are a national framework that pre-empts state rules and the defeat of lawmakers who push tougher ones.

The industry is not unanimous. The same review reports that Anthropic reportedly gave $20 million, restricted to public education on AI policy rather than campaigning. Treat every figure here as reported — the sources are secondary and amounts are not independently verified.

The two readings, side by side. Critics — watchdog and consumer groups — call the spending “dark money” and the pre-emption push regulatory capture: a way to replace state rules with weaker federal ones. Supporters say the spending is legal, partly disclosed and comparable to other industries', and that a single standard is the only workable one for a technology that crosses state lines. Whether pre-emption would come with strong federal rules is exactly what is unsettled.

The European Union's AI Act is the strictest framework in force, and in 2026 it was softened on timing. The Digital Omnibus on AI, formally adopted in June and in force from 27 July 2026 as Regulation (EU) 2026/1744, moves the obligations for standalone high-risk systems (Annex III, such as recruitment and essential services) from 2 August 2026 to 2 December 2027, and those for systems embedded in regulated products (Annex I) to 2 August 2028. The prohibitions, the rules for general-purpose AI models and most transparency duties were not delayed.

The stated reason is readiness: harmonised technical standards and national authorities were not in place. Law-firm commentary calls it a reprieve, not a repeal. We found no source showing who lobbied for the delay, so we do not attribute it to anyone.

Whichever side wins in Washington, the practical expectations on AI systems that act on their own point the same way, and you can build to them now. This is general good practice, not a statement about any law. Payaries has no AI-specific product; the points below describe how its existing controls apply when an AI agent is the caller. See what is agentic AI for the background.

• Keep a human in the loop on money. Payaries withdrawals made with an API key execute immediately only below your auto-approve threshold and wait for your own approval above it — set that threshold with an agent in mind.
• Give an agent the least access that does the job. Use a dedicated API key, not a shared one, and keep the secret out of prompts and logs. How to sign crypto payment API requests shows where the secret stays.
• Keep an audit trail. Record what the agent did and why, and reconcile it against the signed events your gateway sends: verifying webhook signatures.
• Start from the documented contract. The OpenAPI specification and Payaries for AI agents are written for coding assistants, with the side effects of every operation labelled.

• What AI companies want for their midterm millions — The Next Web
• White House AI framework puts federal pre-emption at the center of the debate — Morgan Lewis
• Trump executive order targets state AI laws — Manatt
• Trump administration issues executive order on federal AI policy framework and state law pre-emption — William Fry
• Yes, August 2 still matters: the EU approved a high-risk AI delay — Jones Walker
• EU AI Act deal: Digital Omnibus now in force — Usercentrics