In case you’re listening to about it for the primary time on this put up, there’s a proposed 10-year federal secure harbor moratorium on state and native AI regulation that’s presently embedded in H.R. 1 aka the One Huge Lovely Invoice Act (OBBBA). Huge Tech has been training on secure harbors for many years now to the nice detriment of the remainder of us and this one is so broad and imprecise that it may actually make main AI platforms unstoppable. And I say “main AI platforms” as a result of there actually is not any different type given the capital prices concerned—even earlier than they attempt to lay off a lot of these prices onto the taxpayer in a manner that will make even the Indianapolis Colts’ possession blush.
IAPP’s Cobun Zweifel-Keegan has a great rendering of the secure harbor in an intensive weblog put up:
The language takes some effort to parse, particularly because it consists of double-negative nested exceptions to exceptions, however the finish consequence seems to be as follows. If a state or native regulation imposes a “substantive design, efficiency, data-handling, documentation, civil legal responsibility, taxation, charge, or different requirement” on coated AI programs, it will be topic to the 10-year enforcement moratorium, until the requirement (A) is imposed beneath federal regulation or (b) is imposed by a typically relevant regulation that treats non-covered fashions and programs “that present comparable features” the identical as coated AI fashions or programs.
In fact, it’s type of a stretch to seek advice from “comparable features” in technology-neutral regulation in relation to AI–there isn’t actually something on the market with “comparable features.” So the nested-exceptions language is, a minimum of to me, meaningless.
Let’s name a spade–this new secure harbor give-away to the Greatest of Huge Tech represents a harmful enlargement of presidency regulation. And you understand what they are saying about essentially the most terrifying phrases within the English language.
The secure harbor moratorium is a regulatory free lunch for AI builders, one which echoes and compounds the financial distortions created by earlier secure harbors in Part 230 of the Communications Decency Act, the Digital Millennium Copyright Act (DMCA), and Title I of the Music Modernization Act (MMA). Every of those statutes created uneven authorized immunities that transferred enforcement prices and authorized publicity from highly effective tech intermediaries onto creators, shoppers, and the general public.
At its core, the AI secure harbor moratorium acts as a blanket legal responsibility protect—prohibiting states from imposing current legal guidelines or enacting future legal guidelines addressing algorithmic bias, information provenance, exploitative labor displacement, deepfakes, or unauthorized scraping and coaching. This isn’t deregulatory neutrality; authorities is placing its thumb on the dimensions and deciding that in relation to states defending their residents from Huge Tech, residents lose.
If handed, the OBBBA secure harbor creates an energetic preemption that eliminates each current safeguards and any future efforts to determine guidelines that Huge Tech doesn’t like, with out requiring AI companies to internalize the price of the harms they trigger.
- Customers bear the price of fraud, discrimination, and content material manipulation.
- Artists and authors undergo unremunerated appropriation of their property rights.
- Labor markets are distorted by unchecked deployment of generative programs skilled on uncompensated labor.
This socialization of the draw back danger is matched by the privatization of the upside from a bunch of earnings streams from AI and speculative premiums in capital markets from the inevitable IPOs.
This moratorium continues a troubling sample seen in:
Part 230: Immunized platforms from legal responsibility for user-generated content material, fostering unmoderated harassment, piracy, and disinformation.
DMCA: Created the fake de facto “DMCA license” and offloaded enforcement onto rights holders through sclerotic notice-and-takedown programs, whereas platforms profited from huge content material archives that creators couldn’t afford to take down, a lot much less litigate.
Title I of the MMA: Created a retroactive blanket license for digital music makes use of, administered by the MLC—a government-endorsed non-public monopoly, overlaying up huge infringement and permitting Spotify’s IPO to go ahead which was the actual situation for which songwriters have but to be compensated.
These statutes promised steadiness however delivered structural entrenchment of platform energy, undermining collective enforcement and due course of rights. Everybody of them was abused.
Secure harbors personify “lease in search of” economics, when companies achieve wealth not by creating worth, however by manipulating the political or authorized setting to extract financial features — normally by way of particular remedy, subsidies, or regulation. These free riders profit from crushing shopper and artist opposition to Huge Tech’s dangerous conduct and snarf up non-public property with out paying for it in any respect, not bearing even the price of the identical harms that every other firm must bear from launching a foreseeably faulty product.
Like its predecessors, the AI moratorium is framed as vital to resolve a nonexistent downside. The U.S. Chamber of Commerce will inform you that the moratorium will “promote innovation” by avoiding the dreaded “patchwork of state legal guidelines” that will impose the price of damaging externalities created by their members like Google again on these members. However in fact, it entrenches a federal regulatory void, denying states the power to:
- Implement shopper safety legal guidelines towards misleading or harmful AI merchandise.
- Regulate employment harms or misclassification arising from AI-generated labor substitution.
- Shield creators from the unauthorized ingestion of their expressive works of all types by generative programs—from child photos on Fb or social media to Google Books which was at all times meant for use by an AI.
This “pretextual innovation” framework assumes that AI companies deserve each retrospective and potential immunity—however requires no corresponding transparency, licensing, or restitution. It isn’t innovation coverage; it’s a free-rider enterprise subsidy by way of authorized insulation.
The true monetary worth of the moratorium lies in what it allows:
- Investor confidence in ‘zero-liability’ deployment.
- AI mannequin scaling with out buying licensed coaching information.
- Content material ingestion at industrial scale from state-protected domains.
- Avoidance of union or employee protections by way of preemption of state labor legal guidelines.
- Leaves academic establishments weak to unregulated AI purposes, probably compromising scholar privateness, fairness, and security.
On this mild, the moratorium isn’t merely a procedural provision—it’s a profit-maximizing mechanism, changing authorized immunity into financial benefit.
If generative AI is as transformative as its proponents declare, then its deployment should be matched with symmetrical authorized obligations to these relevant to different merchandise put into commerce. There could be no justification for granting a decade-long moratorium on accountability—particularly when the prices are borne by the very communities and creators whose work has constructed the datasets and economies that AI now monetizes.
The moratorium isn’t impartial. It’s a coverage of structured asymmetry: privatizing upside, socializing draw back, and foreclosing democratic regulation. It’s, in impact, a second DMCA—however for every little thing else.
It is usually a second Part 230—however for everybody.
It should go.
