Claude’s Invisible Ink: How Anthropic’s SynthID Watermark Works
European law is quietly reshaping how AI generates text worldwide. Here is what Anthropic’s watermark actually does, what it cannot do, and why it matters.
How the Watermark Works
When Anthropic announced that Claude would begin watermarking its output, most people pictured a hidden character, an invisible tag, or a code attached to the words — something you could strip out with the right tool. The reality is far more subtle. The watermark is not on the text. The watermark is the text.[1]
Claude generates language one token at a time. At each step, it faces a choice: given “The weather today was cold and ___,” it could pick overcast or grey. Under most circumstances, it does not matter much to the reader which word the model selects — the meaning is largely the same either way. Normally, that choice is settled by a random number generator.[1]
Anthropic’s watermarking system — built on Google DeepMind’s SynthID-Text technology — replaces that source of randomness with a secret pattern. Instead of using an arbitrary random number generator to pick the next word, watermarking uses a cryptographic key and a few words that come before to settle what word the model should pick. The individual words look completely normal. No reader would ever suspect anything. But across thousands of word choices, a statistical signature emerges.
As Anthropic explains it, the pattern is undetectable to the reader but detectable to anyone who has the key that encodes it. The words Claude picks are still random, but one can check the sequence and see if it is consistent with the choices Claude would make if using the key. If it is, one can assign a probability that the text was generated by Claude.[1]
What It Can and Cannot Detect
Anthropic is developing a watermark detection API that will allow users and third parties to check for the pattern. But the result is not a binary “Claude wrote this.” It is a probability. The detector examines thousands of tiny word choices and determines how likely it is that Claude was involved in producing the text.
Key limitations:
Length matters. The watermark does not work well on short texts. Two coin flips tell you nothing; a thousand tell you something. Anthropic notes that detecting a watermark “doesn’t work well on small samples, where there are fewer word choices and thus less information to go on.”[1]
Factual content resists watermarking. “Isaac Newton’s most famous work was called Principia ___” has to be “Mathematica.” The watermark only acts where there is a genuine choice between equally good words. Code and mathematics have generally less watermarking than prose.[1]
It identifies Claude, not you. The watermark carries no identifying information. It cannot be traced to a specific person, organization, or chat. As Anthropic states, “There’s nothing in the watermark, or its key, that would allow anyone to recover any information about the user, their organization, or their chats with Claude.”[1]
It does not determine authorship. Anthropic explicitly says the watermark “can only determine that Claude was likely involved with the content at some point. It cannot distinguish ‘Claude wrote this’ from ‘Claude heavily edited this.’”[1]
The Proofreading Problem
If a writer hands Claude a 2,000-word manuscript and asks it to fix six commas and three misspellings, the document is not meaningfully watermarked. The watermark only applies to words Claude chooses. When Claude proofreads text written by a person, what it gives back has generally only been lightly edited; because nearly all the words are the person’s, there is very little for the watermark to attach to.[1]
But if a writer asks Claude to rewrite an entire chapter, Claude is choosing essentially every word — and the pattern will be embedded throughout. This creates a gray zone. An author who drafts an original manuscript and uses Claude for light proofreading is in a fundamentally different position than one who has Claude generate the text wholesale. But the watermark detector cannot always tell the difference.
Can the Watermark Be Removed?
To some extent, yes. Anthropic acknowledges that “light editing probably won’t remove the watermark completely; a complete rewrite where every word is replaced will.” In the latter case, Anthropic notes, “it’s arguable whether the text can any longer be described as AI-generated.”[1]
The practical problem for authors is that there is no consumer-facing watermark checker available. You cannot test your own text to see if the pattern is still detectable. You are editing blind. Copying text into Word, changing fonts, or retyping it does nothing — because the words themselves are the watermark, not metadata attached to them.
The Brussels Effect: Why EU Law Is Shaping Your AI
Anthropic is implementing watermarking to comply with the European Union’s AI Act, specifically Article 50, which took effect on August 2, 2026. The law requires providers of generative AI systems to mark outputs in a machine-readable format and ensure they are detectable as artificially generated or manipulated.[2]
Anthropic, along with approximately 190 total signatories including Google, Meta, Microsoft, OpenAI, and others, signed the EU Code of Practice on Transparency of AI-Generated Content in July 2026.[1][3]
Here is the part that affects everyone: you do not have to live in the EU for this law to shape your AI tools. Anthropic says it does “not yet have a durable way to scope it by region,” so they are applying watermarking globally. In other words, European law is now shaping the version of Claude that the entire world uses.[1]
This phenomenon is called the Brussels Effect — when a global technology company changes its product to comply with a major market’s regulations, and everyone using that product lives with the result. Penalties for Article 50 breaches run up to 15 million euros or 3% of total annual worldwide turnover, whichever is higher.[2]
The Broader Landscape
Claude is not alone. Google developed SynthID-Text in the first place, and text generated through the Gemini app and web experience is already watermarked. OpenAI has not yet deployed text watermarking for ChatGPT, but it has already adopted SynthID technology for images and is a signatory to the EU Code of Practice.[3]
Platforms across the AI ecosystem are rushing to watermark AI-generated content. AI music platform Suno announced it will mark tracks created on its platform after a spate of legal challenges. Newsletter service Substack teamed up with Pangram to flag AI-generated content, with CEO Chris Best calling out “Claudefishing” — a term for people using AI to generate content without disclosure.[3]
This is part of a larger move toward AI provenance — the ability to trace AI-generated content back to its source. The EU is leading, but the technology and policy frameworks being built now will shape how AI content is identified and regulated worldwide.
The Danger: Moral Classes of Authors
The most important concern about watermarking is not technical. It is about how the publishing industry — and authors themselves — might use this technology. A watermark makes it incredibly easy to formalize a division between “pure” authors (no AI detected) and “suspect” authors (AI watermark present).[4]
But a watermark tells you almost nothing about the quality of a book, the originality of its ideas, or the work the author put in. It does not tell you whether the book is thoughtful, useful, or worth reading. The absence of AI does not tell you anything either — you can write extraordinary prose without AI, extraordinary prose with AI as one of your tools, or absolute garbage entirely by yourself.[4]
The watermark is a technical tool for statistical detection. What matters is what we do with it — and whether we let it become the next excuse to divide authors into the worthy and the unworthy. As author Laura Haelen observed, “We spent the last 15 years proving that a gatekeeper does not get to decide who is an author. And I have absolutely no interest in handing that power back now, just because the gatekeeper has a new test.”[4]
The 315-Use-Case Reckoning: How Federal AI Expansion, a Preemption Executive Order, and a Private Surveillance Network Are Reshaping Criminal Justice Policy
Federal AI deployment in criminal justice jumped 31% in a single year, while a December 2025 executive order tries to block state guardrails and a private nonprofit quietly builds America’s first live facial recognition network.
The Federal AI Expansion at Scale
The Department of Justice’s 2025 AI Use Case Inventory logged 315 entries, a 30.7% increase from the 2024 inventory and a staggering leap from just four reported use cases two years prior.[1] Of those 315, 114 were classified as “high-impact,” a designation covering AI systems whose decisions or actions carry a “legal, material, binding or significant effect on rights and safety.”[2] The inventory spans every corner of the agency, from the Federal Bureau of Prisons using predictive AI to classify inmate misconduct risk, to the ATF analyzing airline travel data for “atypical” passenger routes, to the U.S. Marshals Service considering Amazon Rekognition for facial recognition.[2]
What makes this expansion notable is not just its volume but the range of consequential decisions it touches. The Federal Bureau of Prisons’ “BRAVO Classification” system uses statistical techniques to predict the potential for misconduct among newly admitted inmates and assign security levels accordingly.[2] The DOJ’s Office of Justice also deployed an agentic AI tool called PATTERN, designed to predict recidivism risk for incarcerated adults and ostensibly reduce re-engagement with the justice system.[2] PATTERN is one of only four agentic AI use cases in the entire 2025 inventory, meaning it can take autonomous actions rather than merely generating outputs for human review, a category that did not exist in the prior year’s inventory at all.[2]
Privacy and civil liberties advocates have raised sharp concerns about the trajectory. Jay Stanley, a senior policy analyst with the ACLU’s Speech, Privacy, and Technology Project, said the DOJ’s 2025 inventory provides a “snapshot” of how the federal government “is aggressively seeking to test and exploit a wide variety of AI algorithms and sifting through data on ordinary people.”[2] Jacob Hoffman-Andrews, a senior staff technologist at the Electronic Frontier Foundation, noted that tools predicting criminality or recidivism inevitably make inferences from correlated variables: “Even if you tell us specifically not to, AI is very good at learning correlated variables and figuring out how to be biased.”[2]
The DOJ’s Own Roadmap for Guardrails
Before the current administration’s policy shifts, the Department of Justice produced a 77-page final report on AI in criminal justice, submitted in December 2024 pursuant to the Biden-era Executive Order 14110.[3] The report examines four main applications: identification and surveillance, forensic analysis, predictive policing, and risk assessment, and offers detailed recommendations for each.[4] While the policy environment has evolved since its publication, the technical and ethical considerations it identifies remain the most comprehensive federal framework for responsible AI deployment in criminal justice.
Identification and Surveillance
The report covers facial recognition technology, automated license plate readers, and other biometric systems. Its key recommendations include mandatory training for all system users, clear policies on authorized use cases, quality control procedures, regular auditing and monitoring, and transparent public reporting.[4] The core logic is that these tools offer genuine efficiency gains but require comprehensive policies and oversight to prevent the documented harms of false matches and biased performance across demographic groups.
Predictive Policing
The report distinguishes between place-based tools, which forecast crime hotspots, and person-based tools, which attempt to identify individuals likely to commit or be victimized by crime. Its recommendations stress community engagement before adoption, careful data selection to avoid feedback loops, independent third-party auditing of models and source code, and periodic reevaluation of models based on policing patterns and crime rates.[3] The underlying principle is that predictive models may carry a “veneer of neutrality” reinforced by automation bias, making it essential that humans retain a central role in choosing data, evaluating outputs, and determining interventions.[3]
Risk Assessment
Risk assessment instruments, which use quantitative analysis to estimate recidivism likelihood or failure to appear for trial, are currently used in some form by every state.[3] The report’s recommendations address tool design, implementation, and validation, emphasizing that risk assessment output should inform, not replace, judicial discretion. The Council on Criminal Justice’s summary of the report notes that successful implementation requires agencies to establish robust organizational structures, ensure public oversight and transparency, and develop appropriately trained workforces, both before and after deployment.[4]
The December 2025 Executive Order and the Preemption Push
On December 11, 2025, President Trump signed Executive Order 14365, titled “Ensuring a National Policy Framework for Artificial Intelligence.”[5] The order frames state-level AI regulation as an obstacle to national competitiveness, arguing that “State-by-State regulation by definition creates a patchwork of 50 different regulatory regimes that makes compliance more challenging, particularly for start-ups.”[5] It specifically calls out a Colorado law banning “algorithmic discrimination,” suggesting it may “force AI models to produce false results in order to avoid a ‘differential treatment or impact’ on protected groups.”[5]
The order establishes several mechanisms to challenge state AI laws:
AI Litigation Task Force: The Attorney General must establish a task force within 30 days to challenge state AI laws on grounds including unconstitutional interstate commerce regulation and federal preemption.[5]
Funding restrictions: The Secretary of Commerce must issue a policy notice specifying that states with “onerous” AI laws are ineligible for non-deployment funds under the BEAD broadband program, and other agencies must assess whether they can condition discretionary grants on states not enforcing such laws.[5]
FTC and FCC proceedings: The FTC must issue a policy statement on when state laws requiring alterations to AI model outputs are preempted by the Federal Trade Commission Act, and the FCC must initiate a proceeding to determine whether to adopt a federal reporting standard that preempts conflicting state laws.[5]
However, the order contains a critical carve-out. Section 8(b) specifies that the legislative recommendation for a federal preemption framework should not propose preempting state AI laws relating to “child safety protections,” “AI compute and data center infrastructure,” and crucially, “State government procurement and use of AI.”[5] This means states retain explicit authority to regulate how their own law enforcement and criminal justice agencies procure and use AI systems.
Why the Preemption Threat Rests on Legal Quicksand
A April 2026 analysis by Chiraag Bains, Alex Chohlas-Wood, and Katie Kinsey, published by the Brookings Institution, argues that the executive order’s threats to state regulation are built on dubious legal ground.[6] The authors make several key points about why states should not be deterred:
No inherent preemptive power: An executive order is not a law. The president cannot unilaterally override state statutes. The order merely directs agencies to take actions that might eventually create pathways for preemption, and each pathway faces serious legal obstacles.[6]
Dormant commerce clause weakness: The administration’s main theory is that state AI laws violate the dormant commerce clause, which primarily bars economic protectionism. But most state AI laws apply equally to in-state and out-of-state developers, and AI companies can tailor software to different markets, making it difficult to show excessive burden.[6]
Spending clause limits: The president cannot attach new policy conditions to BEAD funding by executive fiat. The BEAD statute does not mention AI, and AI deregulation does not advance Congress’s goal of expanding high-speed internet access, so the conditions would likely fail the constitutional requirement that funding conditions relate to grant program purposes.[6]
FTC and FCC authority limits: The FTC cannot preempt state regulation by mere assertion, and nothing in the FCC’s statutory authority, which is limited to the communications industry, authorizes it to regulate AI generally or preempt state AI laws.[6]
The authors conclude that the administration’s threats amount to intimidation designed to “discourage state policymakers from even trying to protect their constituents,” calling it “a direct assault on state sovereignty and violates basic principles of federalism.”[6] Both Republican and Democratic governors have signaled they will continue regulating AI regardless. California Governor Gavin Newsom cited the “failure of the federal government to enact comprehensive, sensible AI policy” when signing AI legislation. Utah Governor Spencer Cox argued his state should be allowed to regulate AI, noting that “the Supreme Court is going to back me up on that.” Florida Governor Ron DeSantis declared “we have a right” to pass AI regulation just weeks after Trump signed the order.[6]
New Orleans: America’s First Live Facial Recognition Network
While the federal-state regulatory battle plays out, a private nonprofit in New Orleans has quietly built what appears to be America’s first live facial recognition surveillance network. Project NOLA, founded in 2009 by former police officer Bryan Lagarde, operates a network of over 5,000 cameras mounted on private property by “volunteers” who pay annual connection fees.[7] About 200 of the network’s most advanced cameras have live facial recognition capability, constantly scanning faces in the French Quarter and comparing them against approximately 250 people on “hot lists.”[7]
The system exists in a legal gray zone. New Orleans Police Superintendent Anne Kirkpatrick determined in April 2025 that cooperation with Project NOLA might violate a city ordinance that allowed facial recognition but not live facial recognition from a third-party entity, and she paused the department’s use of real-time alerts.[7] Yet Project NOLA continues feeding information to police, sending “unsolicited” tips when its cameras spot individuals the department is looking for.[7] Federal law does not specifically address how law enforcement may use live facial recognition, though the Supreme Court ruled in United States v. Jones that police must have probable cause and a warrant to use technology to follow someone continuously.[7]
University of Washington law professor Ryan Calo described the arrangement as a potential “shell game” for legal responsibility, warning that “if the surveillance is done ‘by the community,’ by people that are not official actors, that it will circumvent those protections.”[7] The case illustrates how private actors can deploy AI surveillance technologies that police departments are legally or politically constrained from operating themselves, creating a workaround for constitutional protections.
State and Local Guardrails Taking Shape
Despite the federal preemption push, states and municipalities continue building guardrails for AI in policing. According to a Rutgers policy analysis, the pattern across jurisdictions reveals three consistent concerns driving restrictions: whether technologies actually improve outcomes, whether they perpetuate bias, and whether they maintain public legitimacy.[8]
Facial recognition bans and moratoria: San Francisco became the first major U.S. city to ban government use of facial recognition in 2019, followed by Boston and Portland. California imposed a temporary moratorium on facial recognition in police body cameras, while Washington and Colorado implemented requirements for accountability reporting, testing standards, and oversight before deployment.[8]
Predictive policing rollbacks: The LAPD discontinued its PredPol and LASER programs following audits and public criticism. Chicago ended its “Strategic Subject List” after independent evaluations found limited evidence of crime reduction. Santa Cruz banned predictive policing entirely, citing concerns that algorithms trained on historical data reinforce existing enforcement patterns rather than producing unbiased forecasts.[8]
Probable cause requirements: Several states now prohibit law enforcement from relying solely on facial recognition matches as the basis for arrests or probable cause, requiring independent evidence instead.[8] This approach addresses the root cause of documented wrongful arrests, primarily of Black individuals, by ensuring that an AI output triggers further investigation rather than direct action.
The Judicial Perspective: Guardrails for the Courtroom
As AI tools enter criminal courts, judges face questions about admissibility, transparency, and due process. A 2026 article published by the American Bar Association’s Criminal Justice Section and summarized by the National Association for Presiding Judges and Court Executive Officers outlines a framework for balancing innovation and constitutional protections.[9]
The article identifies five guardrails that courts and agencies should adopt:
Transparency and validation: AI systems must be independently validated before use in criminal proceedings, and vendors must disclose training data sources, bias mitigation strategies, and technical materials for defense review.[9] Idaho’s 2019 transparency law, which prohibits vendors from invoking trade-secret defenses when their risk assessment tools are used in liberty decisions, is cited as the leading model.[9]
Human oversight and accountability: AI must be advisory, not determinative. Judges must retain interpretive authority and apply tools to facts rather than allowing tools to supplant judicial reasoning. Prosecutors and defenders must contextualize AI evidence and avoid automation bias.[9]
Continuous review: AI systems evolve over time, and accuracy may degrade as models are updated or datasets shift. Courts must demand periodic validation reports and require vendors to notify them of substantive model changes. If a system begins performing poorly or exacerbating disparities, it must be corrected, suspended, or withdrawn entirely.[9]
The article emphasizes that the fundamental question is not whether AI should be used, as practice has already answered that, but how to use it responsibly. The authors warn that “embracing AI without scrutiny risks automating injustice” and conclude that “people, not algorithms, must remain the ones who make the decisions that determine liberty and justice.”[9]
The Path Forward
The current landscape presents a paradox. Federal AI deployment in criminal justice is accelerating rapidly, with 315 active use cases touching everything from inmate classification to predictive policing. At the same time, the federal government is actively trying to prevent states from building guardrails for the private sector’s development of these same technologies, even as state and local governments continue passing restrictions on their own agencies’ use.
The executive order’s carve-out for state procurement and use of AI is significant because it preserves the most direct lever for criminal justice oversight. States can still regulate how their own law enforcement agencies purchase, deploy, and monitor AI systems. The question is whether they will use that authority aggressively enough to keep pace with the technology’s spread, or whether the chilling effect of federal threats will slow the guardrail-building process.
The DOJ’s own December 2024 report provides a roadmap, with recommendations for community engagement before deployment, independent validation, data quality controls, and ongoing monitoring. Whether those recommendations survive the current administration’s deregulatory posture, and whether states continue filling the governance gap, will determine whether AI in criminal justice enhances fairness and efficiency or amplifies existing disparities under a veneer of algorithmic objectivity.
References
[1][PRIMARY] U.S. Department of Justice. “AI Use Case Inventory 2025.” justice.gov/ai/ai-inventory. 2025. https://www.justice.gov/ai/ai-inventory
[2][SECONDARY] FedScoop. “DOJ ramps up AI for legal work, crime predictions, surveillance, inventory shows.” fedscoop.com. 2025. https://fedscoop.com/justice-department-artificial-intelligence-ai-surveillance-inventory-predictive-technology-algorithm-bias/
[3][PRIMARY] U.S. Department of Justice, Office of Legal Policy. “Artificial Intelligence and Criminal Justice, Final Report.” December 3, 2024. https://www.justice.gov/olp/media/1381796/dl
[4][SECONDARY] Council on Criminal Justice. “DOJ Report on AI in Criminal Justice: Key Takeaways.” counciloncj.org. April 2025. https://counciloncj.org/doj-report-on-ai-in-criminal-justice-key-takeaways/
[5][PRIMARY] The White House. “Ensuring a National Policy Framework for Artificial Intelligence, Executive Order 14365.” December 11, 2025. https://www.whitehouse.gov/presidential-actions/2025/12/eliminating-state-law-obstruction-of-national-artificial-intelligence-policy/
[6][SECONDARY] Bains, Chiraag; Chohlas-Wood, Alex; and Kinsey, Katie. “States can—and should—regulate AI in criminal justice.” Brookings Institution. April 16, 2026. https://www.brookings.edu/articles/states-can-and-should-regulate-ai-in-criminal-justice/
[7][SECONDARY] Kaste, Martin. “Live cameras are tracking faces in New Orleans. Who should control them?” NPR. December 16, 2025. https://www.npr.org/2025/12/16/nx-s1-5616681/new-orleans-live-facial-recognition-surveillance
[8][SECONDARY] Gao, Canyu. “Slowing Down Artificial Intelligence in Policing.” New Jersey State Policy Lab, Rutgers University. April 27, 2026. https://policylab.rutgers.edu/publication/slowing-down-artificial-intelligence-in-policing/
[9][SECONDARY] MacKenzie, Hon. Brian W. and Wallace, David, Esq. “AI in the Criminal Courts: Balancing Innovation and Justice.” American Bar Association Criminal Justice Section. Spring 2026. https://napco4courtleaders.org/2026/06/ai-in-the-criminal-courts-balancing-innovation-and-justice/
AI automated and generated content. Human reviewed.
The Great American AI Act of 2026: A Bipartisan Bet on Federal Preemption
On June 2, 2026, President Donald J. Trump signed Executive Order 14409, “Promoting Advanced Artificial Intelligence Innovation and Security,” establishing a voluntary framework for federal review of frontier AI models. Two days later, Representatives Jay Obernolte (R-Calif.) and Lori Trahan (D-Mass.) released a 269-page bipartisan discussion draft called the Great American Artificial Intelligence Act of 2026, the first proposed comprehensive federal AI governance regime in the United States. The bill would mandate independent audits of the largest frontier models, modernize the Worker Adjustment and Retraining Notification (WARN) Act to account for AI-driven layoffs, and preempt state laws regulating AI model development for three years. The proposal has drawn support from major technology trade associations and fierce opposition from labor unions and consumer advocacy groups, exposing a deep fault line over whether national AI policy should favor speed of innovation or state-level consumer protection.
A Voluntary Federal Floor, Then a Mandatory One
Executive Order 14409 frames the administration’s approach to AI as one of promoting innovation over regulation. “The United States continues to lead the world in Artificial Intelligence (AI) because of the enormous talent and innovation of our AI industry, and because we refuse to stifle this innovation with overly burdensome regulation,” the order states.1 The executive order directs several agencies to act within 30 to 60 days: the Committee on National Security Systems to prioritize cyber defense of national security systems, the Secretary of the Treasury to form an AI cybersecurity clearinghouse in voluntary collaboration with industry, and the Director of the Office of Personnel Management to expand the United States Tech Force Information Cybersecurity Specialist hiring pathways.1
The order’s most consequential provision addresses what it calls “covered frontier models.” Within 60 days, the Secretary of the Treasury, the Secretary of War (via the NSA Director), and the Secretary of Homeland Security (via the CISA Director) are to develop a classified benchmarking process to assess the advanced cyber capabilities of AI models and to determine the threshold for designating a model as a “covered frontier model.”1 Developers would be able to engage the federal government to determine whether their models meet this designation, provide the government access to covered frontier models for up to 30 days before release to other trusted partners, and collaborate on selecting trusted partners for early access.1
Notably, the order explicitly disclaims any mandatory licensing regime: “Nothing in this section shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models.”1 The order also directs the Attorney General to prioritize enforcement of existing federal criminal statutes, including identity fraud (18 U.S.C. 1028), computer fraud and abuse (18 U.S.C. 1030), and wire fraud (18 U.S.C. 1343), against anyone who uses AI to illegally access or damage a computer.1
The Legislative Complement: Four Titles, One Frontier Threshold
The Great American Artificial Intelligence Act of 2026, released as a discussion draft on June 4, would go substantially further than the executive order by establishing binding federal obligations for the largest AI developers. The bill is organized into four titles: frontier AI governance, workforce, cybersecurity, and research, development, and international cooperation.2 Its core regulatory obligations apply only to “large frontier developers,” defined as companies with $500 million or more in annual gross revenue that have trained a frontier model, a threshold that captures OpenAI, Anthropic, Google, Meta, and xAI while intentionally excluding two-person startups, mid-size firms, open-weight developers, and academic researchers.3
Under Title I, large frontier developers must publish a public frontier AI framework that discloses whether a model could pose a “catastrophic risk,” defined as a foreseeable and material risk of death or injury to more than 50 people or more than $1 billion in property damage.4 Developers must retain an Independent Verification Organization (IVO), licensed through NIST’s Center for AI Standards and Innovation (CAISI), to perform semi-annual verification of compliance and assess the adequacy of the developer’s risk monitoring and mitigation.4 Violations of safety requirements or material misrepresentations to auditors carry penalties of up to $1 million per day.4 The bill requires developers to report AI critical safety incidents within 15 days of discovery and imminent risks within 24 hours, with state attorneys general permitted to opt in to receive safety reports.4
The bill also formally establishes CAISI in statute, repurposing what was the AI Safety Institute under the Biden administration, and authorizes $100 million per year for fiscal years 2027 through 2029, in addition to IVO licensing fees.4 The bill’s authorities terminate three years after enactment.4 Representative Obernolte framed the draft as the beginning of a deliberative process: “This discussion draft is an important step toward building a clear federal framework that promotes innovation, protects Americans from emerging risks, and ensures the United States continues to lead the world in AI.”4
The Preemption Fight
The bill’s most contentious provision would preempt state laws “specifically regulating the development” of AI models on a three-year sunset that expires in December 2029, while explicitly preserving state laws of general applicability, including those governing privacy, consumer protection, anti-discrimination, and civil rights.3 According to analysis of the draft, the preemption would apply to California’s AB 2013 (training data disclosure) and portions of SB 942 (content watermarking), and would “federalize” frontier safety laws in California, New York, and Illinois, though state attorneys general would retain enforcement authority over the federalized standards.4
Representative Trahan has sought to frame the preemption as conditional on the strength of the federal standard: “I will preempt, but only if we’re setting the strongest possible federal standard. A preemption provision in a bill with no meaningful federal standard is a moratorium, and I won’t sign onto that.”3 She also defended the sunset clause as a discipline against congressional inertia: “The sunset forces us back to the table… prevents Congress from doing what we did with Section 230 — implementing a policy and then never being required to seriously evaluate whether it’s working.”3
Critics reject the distinction between a moratorium and preemption with a strong federal standard. Brad Carson of Americans for Responsible Innovation (an advocacy organization that opposes preemption of state AI laws) argued that “this bill takes the current floor on state AI legislation and turns it into a federal ceiling, preventing state lawmakers from addressing emerging AI harms in an era of fast-moving technology.”4 Brendan Steinhauser, CEO of the Alliance for Secure AI, struck a similar note: “A national AI standard should protect at least as much as it preempts, and states across the country have already enacted laws that do far more for Americans on issues like child safety and consumer protections than this proposal.”4 Americans for Responsible Innovation launched an ad campaign in Massachusetts urging Representative Trahan to oppose the state law ban, calling preemption a “generational mistake.”4
Public Citizen (a consumer advocacy organization that opposes preemption of state AI laws), in a statement for the record before the House Subcommittee on Commerce, Manufacturing, and Trade, characterized preemption as creating a “decade-long deregulatory blackout” by eliminating existing state-level consumer protections without any federal replacement. “Gutting existing state protections and preventing the creation of new ones regardless of how severe the damage is simply reckless,” the group wrote.5 Public Citizen noted that Congress has failed for years to enact even basic AI protections and that states have filled the void with bipartisan, pragmatic laws, including deepfake pornography bans in two-thirds of U.S. states, deepfake election disinformation laws in half of states, Colorado’s comprehensive AI Act, Tennessee’s ELVIS Act protecting against voice cloning, and North Dakota’s requirement that healthcare decisions be made by doctors rather than AI.5
The group also pushed back against the argument that state regulation stifles innovation, citing the surging valuations of AI companies under existing state laws: OpenAI at $300 billion, Palantir at $281 billion, and Anthropic at $61.5 billion. “America is leading the world in the AI industry. If state regulations were truly unmanageable, the industry would not be surging,” Public Citizen wrote.5
Labor’s “Hard No”
The bill’s workforce provisions, which might have been expected to attract labor support, instead drew immediate opposition. Title II would establish an AI Workforce Research Hub within the Department of Labor within 90 days of enactment, update the WARN Act to require that employers specify when AI is a “substantial factor” in a mass layoff, and modernize federal data collection on AI adoption and workforce impacts.3 Under the amended WARN Act, when AI is a substantial factor in a qualifying mass layoff, the 60-day notice must state that AI was a substantial factor, the type and usage of the AI involved, a good-faith estimate of the percentage of job losses attributable to AI, and any steps taken to upskill or retrain affected workers.6
Nevertheless, the AFL-CIO, the American Federation of Teachers, and the Association of Flight Attendants-CWA issued a joint statement rejecting the bill on the day of its release (the AFT and AFA-CWA statements echoed a separate AFL-CIO statement). “Our response to the Great American AI Act is simple: Hard no. This bill is a giveaway to the AI industry and a handful of trillion-dollar companies, at the expense of American workers,” AFT President Randi Weingarten and AFA-CWA President Sara Nelson said.7 The union leaders cited polling showing that overwhelming majorities of Americans consider AI regulation an urgent issue and worry about harm AI is already causing in workplaces, schools, hospitals, and democracy.7 “Instead of acting to address Americans’ concerns, the Obernolte-Trahan bill blocks responsible state laws and gives free rein to a handful of billionaires who have already shown they cannot be trusted to put public interest ahead of profit,” they wrote.7
The House Democratic AI commission co-chairs, including Representative Ted Lieu (D-Calif.), said the draft “cannot serve as the basis for productive dialogue,” according to reporting on the bill’s reception.3 Fisher Phillips, an employment law firm, noted that Speaker Mike Johnson (R-La.) backs the bill but that it is opposed by labor advocates, civil society groups, and a House Democratic AI commission, and predicted that the bill is unlikely to advance before the August recess, though it jumpstarts the AI regulation conversation for the 2027 Congress.6
Tech Industry and the Innovation Argument
The Information Technology Industry Council, which represents major technology firms, praised the bill and called for a national standard. “Congress must set a national standard that enables responsible AI development, deployment, and adoption across the American economy,” the group said in a statement.8 The Business Software Alliance also supported the draft.3 The White House had urged lawmakers in March 2026 to enact legislation preempting state AI rules, and in December 2025 Trump threatened to withhold federal broadband funding from states whose AI laws were judged to hold back American AI dominance.8
Public Citizen disputed the competitiveness rationale as well, writing that “the suggestion that, unless we deregulate AI, the U.S. will ‘fall behind China’ is both false and offensive.”5 The group recommended that Congress reject any bill preempting state AI regulation and instead pursue a comprehensive federal law creating a nationwide floor of consumer rights based on state best practices, including mandatory labeling of AI-generated content, civil rights protections against algorithmic discrimination, and worker protections including transparency on AI use and the right to collectively bargain over automation.5
What the Bill Leaves to the States
Legal analysis of the bill emphasizes that its preemption is narrower than critics suggest. Fisher Phillips noted that the three-year preemption applies only to state laws regulating AI development, that is, the building and training of frontier models, and does not apply to how AI systems are used or deployed.6 Colorado’s algorithmic discrimination law, California’s automated decision-making technology regulations, Illinois’s AI Video Interview Act, Connecticut’s AI law, and New York City’s automated employment decision tool audit requirements would all remain intact.6
The bill also preserves whistleblower protections that extend beyond frontier developers to any employer. Protected activity includes lawful disclosures to regulatory officials, the attorney general, law enforcement, or Congress, and remedies for prevailing workers include reinstatement, double back pay with interest, compensatory damages, litigation costs, and attorneys’ fees. An anti-waiver provision provides that these rights cannot be waived by contract, policy, or arbitration agreement.6
Public Opinion and the Path Forward
The draft arrives amid significant public demand for AI regulation. According to an Annenberg Public Policy Center survey cited in coverage of the bill, 65 percent of Americans say the government has done too little to regulate AI, including 77 percent of Democrats and 53 percent of Republicans.3 Representative Trahan framed the bill as a response to that demand: “This bipartisan framework is designed to meet the challenges posed by this rapidly advancing technology without smothering American innovation. It protects workers, establishes real accountability for the most powerful frontier systems, and positions the United States to set the global standard on AI.”4
Whether the bill can bridge the gap between tech industry support and labor and consumer group opposition remains uncertain. The discussion draft is open for stakeholder and public feedback before formal introduction. Representative Obernolte said, “We are releasing this draft to hear from stakeholders, experts, and the public so we can strengthen the legislation before it is formally introduced.”8 Additional co-sponsors include Representatives Suhas Subramanyam (D-Va.), Scott Franklin (R-Fla.), Scott Peters (D-Calif.), and Erin Houchin (R-Ind.).4
The central question the bill poses is whether a federal standard focused narrowly on the catastrophic risks of frontier model development is a sufficient trade for displacing the broader patchwork of state laws that have emerged in Congress’s absence. For the technology industry and the Trump administration, the answer is yes: a single national framework reduces compliance burden and accelerates innovation. For labor unions and consumer advocates, the answer is no: a federal ceiling that preempts state protections, even with a sunset, risks creating the same regulatory vacuum that followed Congress’s inaction on social media. The next several months of stakeholder feedback and mark-up negotiations will determine whether the Great American AI Act becomes the foundation of U.S. AI governance or a cautionary tale of preemption without consensus.
Notes
[PRIMARY]Confidence: VERIFIED (2+ sources). Donald J. Trump, “Executive Order 14409: Promoting Advanced Artificial Intelligence Innovation and Security,” The White House, Presidential Actions, June 2, 2026, https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/. Cross-referenced against Reuters reporting on the order’s provisions.
[PRIMARY]Confidence: VERIFIED (2+ sources). Jay Obernolte and Lori Trahan, “The Great American Artificial Intelligence Act of 2026,” Discussion Draft, 269 pages, June 4, 2026, https://trahan.house.gov/uploadedfiles/the_great_american_ai_act_discussion_draft.pdf. Cross-referenced against Roll Call and TechPolicy.Press analyses confirming bill structure, thresholds, and provisions.
[SECONDARY]Confidence: VERIFIED (2+ sources). Justin Hendrix, “Unpacking the Great American Artificial Intelligence Act of 2026,” TechPolicy.Press, podcast transcript, June 14, 2026, https://techpolicy.press/unpacking-the-great-american-artificial-intelligence-act-of-2026/. Cross-referenced against the bill PDF, Roll Call, and Reuters. Trahan quotes verified against the podcast transcript.
[SECONDARY]Confidence: VERIFIED (2+ sources). Allison Mollenkamp, “Bipartisan AI Draft Proposes Three-Year Preemption of State Laws,” Roll Call, June 4, 2026, https://rollcall.com/2026/06/04/bipartisan-ai-draft-proposes-three-year-preemption-of-state-laws/. Cross-referenced against primary sources from Americans for Responsible Innovation (https://ari.us/reps-trahan-obernolte-propose-bill-to-preempt-state-ai-laws/) and the Alliance for Secure AI (https://secureainow.org/press-release/great-american-ai-act-falls-short-of-framework-necessary-to-protect-the-american-people/). Quotes attributed to Rep. Jay Obernolte, Rep. Lori Trahan, Brad Carson (Americans for Responsible Innovation), and Brendan Steinhauser (Alliance for Secure AI). Note: original draft contained a duplicate footnote (former fn5) pointing to the same Roll Call URL; merged into this single reference.
[ADVOCACY]Confidence: VERIFIED (2+ sources). J. B. Branch, “Federal Preemption of State AI Laws Is Dangerous and Reckless,” Public Citizen, statement for the record before the House Subcommittee on Commerce, Manufacturing, and Trade, 2026, https://www.citizen.org/article/federal-preemption-of-state-ai-laws-is-dangerous-and-reckless/. Bias disclosure: Public Citizen is a consumer advocacy organization that opposes preemption of state AI regulation. Cross-referenced against Reuters reporting.
[SECONDARY]Confidence: VERIFIED (2+ sources). Fisher Phillips LLP, “Congress Proposes First Comprehensive Federal AI Framework: Here’s What It Means for Employers,” Fisher Phillips Insights, June 9, 2026, https://www.fisherphillips.com/en/insights/insights/congress-proposes-first-comprehensive-federal-ai-framework. Cross-referenced against Roll Call and the bill PDF. WARN Act disclosure requirements and whistleblower protections verified against primary bill text.
[ADVOCACY]Confidence: VERIFIED (2+ sources). Randi Weingarten and Sara Nelson, “Union Leaders Urge Congress to Reject the Great American AI Act,” American Federation of Teachers, press release, June 4, 2026, https://www.aft.org/press-release/union-leaders-urge-congress-reject-great-american-ai-act. Bias disclosure: AFT and AFA-CWA are labor unions that oppose the bill. The AFT/AFA-CWA statement echoed a separate AFL-CIO statement (https://aflcio.org/press/releases/afl-cio-statement-trahan-obernolte-ai-discussion-draft). Cross-referenced against the AFL-CIO statement.
[SECONDARY]Confidence: VERIFIED (2+ sources). Courtney Rozen and David Shepardson, “US House Lawmakers Release Draft Bill to Prohibit State AI Rules,” Reuters, June 4, 2026, https://www.reuters.com/business/us-house-lawmakers-release-draft-bill-regulate-ai-2026-06-04/. Cross-referenced against the White House executive order and Roll Call. Quote attributed to the Information Technology Industry Council verified against Reuters reporting.