The AI Regulator Act of 2026 Section-by-Section Summary is a ten-page document published on September 23, 2026 by the offices of Sen. Michael Bennet (D-Colo.) and Sen. Peter Welch (D-Vt.), alongside a release announcing the proposal and a one-pager. It summarizes, in 23 sections, a proposal to establish a Federal Digital Commission with authority over large digital platforms and developers of frontier AI models. It is a summary prepared by the sponsors, not bill text; the release calls the measure "a new proposal building off of" the sponsors' Digital Platform Commission Act of 2022 (reintroduced in 2023) and gives no bill number (Source: bennet.senate.gov). The proposal itself is described at AI Regulator Act (Bennet–Welch proposal, 2026).
Summary
The document describes an agency whose remit combines two strands. The first, carried over from the Digital Platform Commission Act, is regulation of digital platforms: designation of systemically important platforms, rules on data portability, algorithmic auditing and content-moderation transparency, an age-appropriate design code, review of mergers, and consumer complaints. The second, new in 2026, concerns frontier AI: a compute-based definition of frontier models, a catastrophic-risk standard, mandatory submission of models for testing before public distribution, a power to pause distribution, incident reporting, and an interagency committee for international coordination on catastrophic AI risk.
Definitions (section 2)
Section 2 excerpts the proposal's definitions:
- Algorithmic process — a computational process, including one derived from AI, that processes personal information or other data to make a decision, generate content, or determine how information is presented to users of a digital platform, including commercial content, social media posts and search results.
- Artificial intelligence — a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments, using machine- and human-based inputs to perceive environments, abstract those perceptions into models through automated analysis, and use model inference to formulate options for information or action.
- Catastrophic risk — a foreseeable and material risk that a developer's development, storage, use or deployment of a frontier model will materially contribute to the death or serious injury of more than 50 people, or more than $1,000,000,000 in damage to or loss of property, arising from a single incident involving a frontier model. Example incidents named are bio, chemical or cyber weapons, and "engaging in conduct with no meaningful human oversight or intervention."
- Critical safety incident — unauthorized access to, modification of, or exfiltration of frontier model weights that results in death or bodily injury; harm resulting from the materialization of a catastrophic risk; loss of control of a frontier model causing death or bodily injury; or a frontier model using deceptive techniques against its own developer to subvert the developer's controls or monitoring in a manner demonstrating materially increased catastrophic risk.
- Digital platform — an online service acting as an intermediary facilitating interactions between consumers, or between consumers and entities offering goods and services, including content primarily generated by algorithmic processes; it includes online services directly offering goods and services unless they are a de minimis part of the platform's business, and excludes news organizations' websites and small businesses, with a "small digital platform business" to be defined by the Small Business Administration within six months.
- Foundation model — an AI model made available or used in interstate or foreign commerce that is trained on a broad dataset, designed for generality of output, and adaptable to a wide range of distinctive tasks.
- Frontier model — a foundation model trained using more than 10^26 integer or floating-point operations, counting the original training run and any subsequent fine-tuning, reinforcement learning or other substantial modification the developer applies to a preceding foundation model. The text layer prints the figure as "10x26".
- Online service — a consumer-facing website, back-end online support system, or other facilitator of online activities.
- Systemically important digital platform and systemically important developer — special designations allowing additional regulation and oversight by the commission.
The commission and its jurisdiction (sections 3–6, 8)
Section 3 establishes the Federal Digital Commission to regulate digital platforms in the public interest and lists eleven purposes: protection from catastrophic risks posed by frontier AI models; innovation that promotes the public good, distributed economic growth, secure job markets and national security; security from overseas frontier AI models that threaten U.S. prosperity, national security or public health and safety; prevention of foreign manipulation of U.S. discourse, including use of platforms and AI models to spread false information that could harm economic interests, national security or elections; global agreement on frontier AI safety standards "that prioritize long-term well-being and freedom"; competition, including combating AI-based online scams; prevention of "harmful levels of concentration of private power over critical digital infrastructure"; a competitive marketplace of ideas; consumer protection from deceptive, unfair, unjust, unreasonable or abusive practices; fair, transparent and safe algorithmic processes; and access to platforms for civic engagement and economic and educational opportunity.
Section 4 gives the commission "plenary jurisdiction" over digital platforms or AI developers that provide services in the United States and affect interstate or foreign commerce, and a forbearance power to expand or restrict its jurisdiction over a platform or class of platforms. Section 5 provides for five commissioners, including a chair, appointed by the President and confirmed by the Senate for five-year terms, no more than three from one party, with qualifications, compensation and conflict-of-interest protections; the commission sits in Washington, D.C., may hold special sessions elsewhere, and must enter memoranda of understanding with the Federal Communications Commission, the Federal Trade Commission and the Department of Justice on overlapping jurisdiction. Section 6 sets the chair's duties, permits functional bureaus, directs the commission to prioritize hiring staff with backgrounds in computer science, data science, application development and technology policy, and lets any person aggrieved by a commission order, decision or action apply for review. Section 8 grants rulemaking authority and directs that rules be tailored to the size, market dominance or other attributes of particular platforms or developers.
Frontier AI provisions (section 4)
For systemically important developers, section 4 directs the commission to issue rules on:
- risk thresholds for assessing whether a frontier model has a capability that could pose a catastrophic risk;
- reasonable safeguards a developer may implement to mitigate catastrophic risk;
- reporting on governance policies and practices, risk-monitoring methodologies, risk-mitigation procedures, cybersecurity policies and practices, and testing, evaluation and training frameworks;
- a mandatory process for submitting models to the commission for testing and evaluation, after which the commission must approve or disapprove public distribution within 45 days, with an extension of no more than 30 days;
- a pause on public distribution of a disapproved model for no more than six months, or until the developer demonstrates material changes that satisfy safety requirements;
- reporting of critical safety incidents to the commission within 15 days of a known incident.
Platform provisions (sections 4, 9, 13–16, 21–22)
For systemically important digital platforms, the commission must decide within six months whether to issue rules on data portability and interoperability; fair, transparent algorithmic systems without harmful or anticompetitive bias; algorithmic auditing, accountability and explainability; transparency of terms of service and content-moderation policies; regular public risk assessments of harmful content, including from algorithmic processes; general transparency to enable commission oversight, third-party audits and trusted third-party research; and accessibility for populations such as blind users. It is also directed to develop an age-appropriate design code and age-verification standards (section 4).
Section 9 sets the designation criteria. A platform is designated systemically important if it is open to the public on one side with significant user engagement, conducts business primarily at the interstate or international level, and has operations with significant nationwide economic, social or political impacts, which the commission defines by public rulemaking considering such attributes as the ability to shape news dissemination, the ability to cause significant harm by excluding a person, market power, unique daily users, and business users' dependence on the platform. A developer is designated systemically important if it is engaged in developing a frontier AI model or meets other thresholds for AI-related annual expenditures. The commission may require annual reports from systemically important platforms.
Section 13 establishes a Research Office that conducts research, collaborates with outside experts and awards grants, and a pilot program giving vetted, financially disinterested experts access to platform data collected by the commission. Section 14 grants investigative authority over platforms' management and business practices. Section 15 routes Hart-Scott-Rodino filings for mergers involving a systemically important platform to the commission, which may request more information and recommend to the DOJ and FTC whether the merger should be consummated; the agencies must give the recommendation "substantial weight" and may reject a merger on its basis. Section 21 requires systemically important platforms or developers to maintain a U.S. presence and parent companies with significant foreign ownership to safeguard U.S. user data; section 22 requires them to identify bots and label the country of origin of posts.
Advisory, interagency and international bodies (sections 7, 10–12)
Section 7 creates an 18-member Technology Advisory Board — six representatives of digital platforms, six from nonprofit public-interest groups, academia and other financially disinterested experts, and six technical experts — which develops voluntary or enforceable codes, technical standards and policies through a "code process" initiated by the commission or "the Council." The process runs no less than six months and no more than a year, with public feedback and a public factual record; the commission must then approve or reject the recommendation no earlier than 45 and no later than 90 days after receiving it. Recommendations are not binding without commission approval, and the commission retains its ordinary rulemaking route.
Section 10 requires the commission to support other agencies on request, and requires federal agencies, including the FTC and DOJ, engaged in oversight of platforms or AI developers to consult the commission, and the commission to consult them. Section 11 establishes the Committee on Global Artificial Intelligence Coordination, chaired by the commission's chair, with the Secretaries of the Treasury, Homeland Security, Commerce, Defense, State, Energy and Labor, the Attorney General, the chairs of the FCC and FTC, and the Director of National Intelligence (nonvoting, ex officio), to engage foreign governments, the United Nations and the European Union on "international frameworks and governing standards to collectively regulate catastrophic AI risks." Section 12 allows any platform to petition for forbearance from the commission's rules; a petition not granted within 18 months is dismissed without prejudice, and decisions are subject to the Administrative Procedure Act.
Enforcement, oversight and funding (sections 16–20, 23)
Under section 16, anyone damaged by a digital platform or AI developer may complain to the commission or bring a civil action to petition for enforcement; a platform or developer that satisfies the complaint is relieved of liability for that violation, and if a platform does not, the commission investigates, concludes within six months, and may order payment to the complainant. Section 17 authorizes cease-and-desist and restitution orders, establishes a civil penalty for a digital platform that knowingly violates the Act at an amount the commission deems necessary to deter, and caps total annual penalties on a digital platform at 15 percent of its total global revenue in the preceding year. Section 18 allows appeals to the U.S. District Court for the District of Columbia or the district where the person resides. Section 19 creates a whistleblower protection program with penalty sharing and anti-retaliation protections. Section 20 requires the President, after five years, to appoint an independent panel to report to Congress on the commission's effectiveness, propose reforms, and recommend whether it should continue. Section 23 authorizes $200 million for fiscal year 2027, ramping to $500 million by fiscal year 2031.
Discrepancies noted
- Section 2 says the systemically important designations rest on "attributes outlined in Section 10," but the designation criteria appear in section 9; section 10 concerns inter-agency support.
- Section 7 says the code process may be initiated by the commission "or the Council," a body the summary does not otherwise define or establish.
- Sections 16 and 17 name "a digital platform" where the adjacent text names both platforms and AI developers: the commission's investigation on an unsatisfied complaint, the knowing-violation penalty, and the 15 percent cap are each stated for digital platforms. The sponsors' release describes the penalty as up to 15 percent "of a firm's prior-year global revenue" (Source: bennet.senate.gov). Whether the cap applies to AI developers is not stated in the summary.
- Section 22 uses "systematically important" where the rest of the document uses "systemically important."
Relation to other instruments
The frontier-model compute threshold (10^26 operations) and the catastrophic-risk harm levels (more than 50 deaths or serious injuries, or more than $1 billion in property damage from a single incident) match the figures in California's SB 53; the compute figure also appears in New York's RAISE Act and in the Artificial Intelligence Risk Evaluation Act (see Compute Thresholds). Unlike those instruments, the summary couples the definitions with an approval requirement: a model cannot be publicly distributed until the commission approves it or the review period lapses, and a disapproved model may be paused. The critical safety incident definition's four limbs (weight compromise causing harm, materialized catastrophic risk, loss of control, and deception of the developer) are compared with other incident-reporting triggers at AI Incident Reporting. The next day Sen. Ed Markey introduced the Cybersecurity and AI Board of Investigations Act (S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)), which creates an investigative body with no licensing or penalty power.
Provenance
Captured September 25, 2026 from the canonical Senate office host and extracted with pdftotext; page boundaries are marked in the raw file. The one-pager linked from the release was not captured. No bill text or bill number had been published at capture.
Relationships
- instance-of: Frontier AI Governance
- supports: AI Regulator Act (Bennet–Welch proposal, 2026) — primary document for the proposal
- related: AI Pre-Release Vetting — mandatory testing, 45-day approval window, six-month pause
- related: AI Incident Reporting — 15-day critical safety incident reporting
- related: Compute Thresholds — 10^26-operation frontier-model definition
- related: Coordinated slowdown proposals compared
- related: S. 5541 — Cybersecurity and AI Board of Investigations Act of 2026 (bill text)
- related: Michael Bennet, Peter Welch