SANDBOX Act (S. 2750)

The Strengthening Artificial intelligence Normalization and Diffusion By Oversight and eXperimentation Act, or SANDBOX Act (S. 2750, 119th Congress), is a bill introduced by Sen. Ted Cruz (R-TX), chairman of the Senate Committee on Commerce, Science, and Transportation, on September 10, 2025. It would require the director of the White House Office of Science and Technology Policy (OSTP) to establish a federal "artificial intelligence regulatory sandbox program" under which applicants, and the director, could seek temporary waivers or modifications of federal rules in order to test, experiment with, or temporarily provide AI products, services or development methods to consumers (Source: congress.gov). Cruz presented it as the first bill under his five-pillar AI policy framework (Source: commerce.senate.gov).

Status and legislative history

The bill was read twice and referred to the Commerce Committee on the day of introduction, and congress.gov lists a single version and no further action (Source: congress.gov). It was released alongside Cruz's "Legislative Framework for American Leadership in Artificial Intelligence," which the committee described as a light-touch regulatory strategy (Source: commerce.senate.gov). Organizations backing the framework and bill at release included the Abundance Institute (Abundance Institute), the U.S. Chamber of Commerce and the Information Technology Industry Council (Source: commerce.senate.gov). A September 2026 Center for American Progress report analyzing the text describes it as a discussion draft that "has not been revised or advanced since" (The SANDBOX Act Uses AI Policy as a Trojan Horse for Unfettered Deregulation (Center for American Progress, September 2026)).

Scope and definitions

Section 701 defines the program's terms (Source: congress.gov):

• Artificial intelligence takes its meaning from section 5002 of the National Artificial Intelligence Initiative Act of 2020. An AI development method is "a business model or production method that, in whole or in part, uses one or more artificial intelligence systems," and an AI product or service is one that "uses or contains, in whole or in part," one or more AI systems.
• Covered provision has "the meaning given the term 'rule' in section 804(3) of title 5, United States Code" (the Congressional Review Act definition), "including any associated guidance, frequently asked questions publications, bulletins, or associated, derivative material and any rule the adoption of which is expressly required by statute."
• Health and safety risk is a risk likely to cause bodily harm to or loss of human life, or a substantial adverse effect on human health, in each case "including life before birth" or an unborn human.
• Risk of economic damage means "likely to cause tangible, physical harm to the property or assets of a consumer."
• Unfair or deceptive trade practice refers to acts unlawful under section 5 of the FTC Act or as defined in the FTC's 1983 Policy Statement on Deception and 1980 Policy Statement on Unfairness.

The stated purpose is to incentivize AI products, services and development methods, economic opportunity, job creation and AI innovation in the United States (§ 702(b)).

Key provisions

Establishment and applications

The director must establish the program within one year of enactment, including a standardized application, a process for director-submitted applications, and a published review process with criteria subject to public comment (§ 702(a)). A private application must confirm US jurisdiction and a US-incorporated or US-based business; disclose criminal convictions of the applicant or senior management; and, for each product or method, identify each covered provision to be waived or modified and why, describe how it would benefit consumers, improve operational efficiency, expand economic opportunity, create jobs or further AI innovation, explain how benefits outweigh reasonably foreseeable health and safety, economic-damage and unfair-practice risks, describe mitigation, state the requested term, confirm future compliance, and list agencies with jurisdiction (§ 702(c)(1)). The director may help applicants identify likely eligible provisions and relevant agencies and share anonymized information on other applications (§ 702(c)(2)). The director may also submit an application to waive or modify "any covered provision" if, "in the opinion of the Director," doing so "will advance the development, deployment, or use of artificial intelligence in the United States" (§ 702(c)(3)) (Source: congress.gov).

Agency review and decision

The director transmits completed applications to each applicable agency within 14 days. Agency heads review the application, soliciting input from the private sector and technical experts, on whether the plan would deliver the listed benefits and whether those benefits outweigh the three defined risk categories, and may use advisory boards or working groups. Each agency must approve or deny within 90 days, with one 30-day extension available. The record of decision must list the provisions and foreseeable risks; for an approval, describe mitigation and consumer protection "if applicable"; for a denial, explain how the waiver could cause the risks, their likelihood, and why partial approval or reformulation would not suffice; and, where the agency would deny unless risks are mitigated, recommend mitigation, after which the applicant has 60 days to revise for an expedited review of up to 60 days. If an agency submits no record by the deadline, "the Director shall presume" it does not object and may proceed. Where several agencies are involved, the director grants the application in full if all approve or in part if some deny (§ 702(d)) (Source: congress.gov).

Written agreement, appeals and judicial review

A granted waiver takes effect only after the applicant signs a written agreement with the director and the agency setting out the waived provisions and mitigation terms, including notice to the director and agencies of any incident causing consumer health and safety harm, economic damage or an unfair or deceptive practice within 72 hours (§ 702(e)). A denied applicant may appeal to the director, and for a director-submitted application the director prepares the response; within 60 days the director determines whether the appeal addresses the agency's concerns and approves or denies, filing a record of decision (§ 702(g)). Records of decision, renewals, failures to provide a written agreement, and revocations are final agency actions for judicial review (§ 702(h)). When a director-submitted application is granted, the director publishes the waiver in the Federal Register and any person may apply to use it (§ 702(f)) (Source: congress.gov).

Term, renewal, revocation and sunset

Waivers run for two years and may be renewed for up to four additional two-year periods; the director must grant renewal unless circumstances have materially changed or the participant is out of compliance and cannot cure. On noncompliance, the director must allow 30 days to correct, with further 30-day periods at the director's discretion, and may then revoke. The program terminates 12 years after it is established (§ 702(i), (j), (q)) (Source: congress.gov).

Participant terms and consumer protection

While a waiver is in effect and the participant complies with its agreement, the participant is not subject to "the criminal or civil enforcement of a covered provision specifically identified in the waiver or modification," and an agency may not pursue punitive action, including civil penalties, fines, or license suspension or revocation, for violations of waived provisions. Consumers' existing rights of action for damages or equitable relief cannot be waived, participants retain no immunity for criminal offenses not expressly identified in the waiver, and the federal government bears no liability for business losses on revocation (§ 702(k)). Before and during deployment, participants must publicly disclose their participation, any lack of licensing, that the product is being tested and may expose consumers to identified risks, that they are not immune from civil liability or from prosecution for unwaived provisions, that the product is a temporary demonstration, and the contact details of the National Artificial Intelligence Initiative Office for complaints (§ 702(l)). Participants must keep records and file reports 40 days after commencement, at the midpoint and before expiry of each term, covering consumer numbers, risks and mitigation, unanticipated risks, adverse incidents and benefits (§ 702(m)–(n)) (Source: congress.gov).

Reporting and congressional review of covered provisions

The director reports annually to Congress on applications, participants, public benefits and harms, provisions waived and consumers affected (§ 702(n)(2)), and coordinates with comparable state programs, including joint applications (§ 702(o)). Section 703 requires the director, after May 1 each year, to submit a special message detailing each covered provision the director "recommends should be amended or repealed as a result of persons being able to operate safely without those covered provisions," with counts of waivers and denials, rationales and recommended text. Any member may introduce a joint resolution of approval that repeals or amends listed provisions; committees are discharged after 10 legislative days, floor debate in the House is limited to two hours, the Senate motion to proceed is non-debatable "notwithstanding Rule XXII," and Congress must complete action within 60 legislative days of the message. Section 703(b)(1) refers to the sandbox program as established "under section 5107(b)" (Source: congress.gov).

Reactions

Organizations supporting the bill at its release included the Abundance Institute, the U.S. Chamber of Commerce and the Information Technology Industry Council (Source: commerce.senate.gov). TechPolicy.Press described it as a program to waive federal regulations for AI developers, with OSTP coordinating evaluation across agencies (Source: techpolicy.press).

The AFL-CIO wrote on October 16, 2025 in opposition to legislation that would exempt companies from future AI regulations, a letter cited by later critics of the bill (The SANDBOX Act Uses AI Policy as a Trojan Horse for Unfettered Deregulation (Center for American Progress, September 2026)).

In September 2026, Nicole Alvarez and Adam Conner of the Center for American Progress argued that the bill "uses the goal of promoting AI innovation to cloak a sweeping deregulatory agenda." Their report contends that it sets no threshold for how central AI must be to an activity, that the covered-provision definition reaches guidance documents, statutorily mandated rules and federal procurement regulations, that agency review is weighted toward approval, that the director's powers to initiate applications, presume consent and decide appeals concentrate authority in one White House official, that the Section 703 process could turn temporary waivers into permanent repeal, and that allowing the executive to suspend rules Congress required raises nondelegation and separation-of-powers concerns. It illustrates the argument with Clean Air Act permitting, Medicare Advantage prior-authorization rules and Fair Labor Standards Act overtime exemptions, and concludes that Congress "should reject the SANDBOX Act and any similar proposal" (The SANDBOX Act Uses AI Policy as a Trojan Horse for Unfettered Deregulation (Center for American Progress, September 2026)). Inside AI Policy reported the position on September 25, 2026, noting that it runs against Rep. Josh Gottheimer, who backs AI sandbox legislation for financial services (Source: insideaipolicy.com).

Comparison with other sandbox regimes

The bill differs from sector sandboxes such as the UK Financial Conduct Authority program and the Texas TRAIGA sandbox in its administrator (a White House office rather than a sector regulator), its reach (any covered provision across agencies rather than a single regulator's rules), and its link to a congressional repeal procedure; TRAIGA, like the SANDBOX Act, suspends enforcement of waived laws during testing but lists non-waivable prohibitions. See AI Regulatory Sandbox. Financial-sector proposals with narrower waiver authority include the Unleashing AI Innovation in Financial Services Act (H.R. 4801). America's AI Action Plan also calls for regulatory sandboxes.

Relationships

• instance-of: AI Regulatory Sandbox
• related: Ted Cruz — sponsor; Office of Science and Technology Policy (OSTP) — administering office
• contradicts: The SANDBOX Act Uses AI Policy as a Trojan Horse for Unfettered Deregulation (Center for American Progress, September 2026) — CAP report opposing the bill
• related: Unleashing AI Innovation in Financial Services Act (H.R. 4801) — financial-sector sandbox bill
• related: Congressional Review Act — source of the covered-provision definition and model for the Section 703 procedure
• related: America's AI Action Plan, Carolina Principles for Emerging Technologies — administration sandbox policy
• related: Nondelegation Doctrine, AI Federalism