Center for American Progress

Fact Sheet: Recommendations for the Department of Labor To Take Further Action on AI
Fact Sheet

Fact Sheet: Recommendations for the Department of Labor To Take Further Action on AI

This fact sheet offers recommendations for how the Department of Labor can utilize its authorities to address artificial intelligence (AI).

A line of workers holding picket signs marches outside an Amazon warehouse with two delivery trucks parked at it.
Amazon delivery drivers and dispatchers strike to protest unfair labor practices at the company's Palmdale, California, warehouse and delivery center on July 25, 2023. (Getty/Robyn Beck)
Read the full report

This fact sheet collects the recommendations from Chapter 2: “The Department of Labor” of the joint report from Governing for Impact (GFI) and the Center for American Progress, “Taking Further Agency Action on AI: How Agencies Can Deploy Existing Statutory Authorities To Regulate Artificial Intelligence.” The chapter notes how the U.S. Department of Labor (DOL) oversees numerous statutes, from the Fair Labor Standards Act (FLSA) to the Family and Medical Leave Act (FMLA), that can potentially help address the challenges and opportunities of artificial intelligence (AI) as it affects workers. These recommendations stem from DOL-enforced statutes identified in the chapter that could be used to address AI through regulations, subregulatory guidance, and enforcement practices. Among other authorities, the DOL could use these statutes to ameliorate known harms by updating wage and hour regulations, guarding workers’ safety and health against the negative impacts of automated management, and ensuring that automated benefits administration is transparent and fair. The goal of these recommendations is to provoke a generative discussion about the following proposals, rather than outline a definitive executive action agenda. This menu of potential recommendations demonstrates that there are more options for agencies to explore beyond their current work, and that agencies should immediately utilize existing authorities to address AI.

Read the full chapter

Fair Labor Standards Act: Recordkeeping and reporting

Based on this authority, the DOL could consider the following actions:

  • Issue new recordkeeping and reporting rules, pursuant to 29 U.S.C. § 211(c), to require employer records to ensure legibility and transparency of wage determinations made by automated systems and to require periodic reports to the Wage and Hour Division (WHD) of those records from employers using AI-driven wage and scheduling technology. Such regulations would help combat black-box wage determination and discrimination1 that can make workers’ wages unpredictable and irregular,2 as well as ensure that such wage determinations satisfy the minimum wage and overtime requirements of the FLSA. As documented by Veena Dubal, professor of law at the University of California, Irvine, many workers are subject to algorithmic management and wage setting that withholds or reduces compensation for work when doing so benefits the company.3 This can make it difficult for workers to appreciate the connection between time spent working and amount of income generated, or to understand and correct errors in their compensation, and can also result in opaque wage setting that violates minimum wage or overtime laws.4 The DOL contemplated a similar rulemaking in the early 2010s that would have required recordkeeping and disclosure to workers about their status as employees or independent contractors and detailed information about how their pay is computed, but a regulation was never proposed.5
  • Launch investigations, pursuant to its administrative subpoena power in 29 U.S.C. § 211(a),6 of employers to ensure compliance with minimum wage and overtime provisions. The WHD could prioritize investigation of employers that are noncompliant with the reporting rules mentioned, are in industries with large numbers of employee complaints, or are in industries with high penetration of automated wage and scheduling technologies. These investigations could produce valuable information about the characteristics of automated systems that make minimum wage and overtime violations more likely to occur and encourage employers’ compliance with their legal obligations under the FLSA.

Fair Labor Standards Act: Minimum wage and overtime

Based on the above-cited authority, the DOL could consider the following action:

  • Issue updated interpretive regulations at 29 C.F.R. Part 785, pursuant to 29 U.S.C. § 211(c), that allow only employers who track time manually through analog methods to engage in timesheet rounding7 and establish a presumption against application of the de minimis rule in cases where employers use highly precise timekeeping technology.8 These changes would eliminate an outdated regulatory regime that allows companies to use sophisticated timekeeping technology to facilitate wage theft by exploiting rules meant to minimize the burden of pen-and-paper wage and hour calculations. Given the ubiquity and ease of digital timekeeping, there is no longer a compelling justification for allowing practices such as rounding employees’ hours to the nearest quarter-hour or failing to treat short periods of working time as compensable for minimum wage and overtime compliance.9

Unemployment compensation

Based on the above-cited authority, the DOL could consider the following actions:

  • Update quality control program regulations at 20 CFR § 602.21, pursuant to 42 U.S.C. §§ 503(a)(1) and 1302, to require states to undertake audits and submit their results to the DOL for any automated or AI-driven benefits determination system. This could help ensure that states provide unemployment compensation to individuals consistent with federal law, provide for human in-the-loop review of any algorithmic denial of benefits, and ensure fair human adjudication for appeals of those denials. The current quality control program regulations were promulgated based on this same statutory authority.10 These regulations would guard against states’ use of automated systems to deny coverage to eligible individuals (or worse, wrongfully accuse them of fraud),11 a use case cited by the Office of Management and Budget (OMB) as presumptively rights-impacting, and therefore it should be subject to heightened scrutiny.12 This proposal is closely related to the actions directed in Section 7.2(b) of the president’s 2023 executive order on AI, which aims to ensure the equitable distribution of public benefits. For example, the executive order directs the U.S. Department of Agriculture to issue guidance to state, local, and Tribal governments that address the use of AI systems in benefits distribution. It requires such guidance to ensure that such systems, among other things, maximize program access; require governments to notify the Department of Agriculture of AI use; create opt-out opportunities for benefit denial appeal; and enable auditing to ensure equitable outcomes.13
  • Issue a new unemployment insurance program letter (UIPL) to guide states specifically on where and how AI can and should be implemented for unemployment insurance administration. This new UIPL should incorporate the minimum risk management practices for the presumed rights-impacting use of AI from theOMB M-24-10 AI memo14 and any subsequent guidance. For example, utilizing AI to flag potential fraud must be accompanied by the minimum risk practices from the OMB M-24-10 AI memo, such as carrying out AI impact assessments, testing the systems in the real world before widespread deployment, and ongoing monitoring to ensure equity.15 The DOL should clarify that these requirements extend to any vendor a state unemployment insurance system contracts with to provide services.

Occupational Safety and Health Act

Based on the above-cited authority, the DOL could consider the following actions:

  • Begin the standard-setting process, pursuant to 29 U.S.C. § 655(b), to regulate the use of electronic surveillance and automated management (ESAM) in the workplace to the extent that it creates hazards to workers’ physical and mental safety and health. Such regulation could mitigate the increasingly unsustainable pace of work enforced by these systems, which leads to ergonomic injury and increased risk of accidents. For example, the Washington State Department of Labor and Industries has fined Amazon repeatedly for forcing its warehouse workers to work at punishing speeds that exacerbate the risk of injury.16 The state’s citations specifically reference the “direct connection” between Amazon’s ESAM and workplace musculoskeletal disorders.17 A standard on ESAM would also reduce the harmful effects that these systems can have on workers’ mental health. As early as 1987, the now-defunct U.S. Office of Technology Assessment recognized that ESAM increases employee stress, heightening job strain risk.18

Of course, the Occupational Safety and Health Administration’s (OSHA) standard-setting process is uniquely slow and resource intensive for the agency,19 and the process would need to be informed by additional research to design an effective policy. So, in the meantime, the following recommendations should be considered:

  • Issue new subregulatory guidance and bring general duty clause enforcement actions related to companies’ use of ESAM in ways that harm worker safety and health. As GFI has urged in past advocacy efforts, OSHA should follow the lead of Washington state by more directly tying ESAM use to physical and mental health hazards.20 Enforcement actions based on unsafe ESAM use could be taken because of the already ongoing DOL investigation into high injury rates at Amazon warehouses.21
  • Update existing subregulatory guidance about sector-specific ergonomic risks to include a discussion of how ESAM can increase musculoskeletal injury risk. As described in a GFI report in 2023, OSHA could update the ergonomics guidance documents for poultry processing and grocery warehousing and create a new ESAM-conscious ergonomic risks guidance document for the warehousing industry.22 The guidance could describe best practices to prevent ergonomic injuries—such as quota transparency, worker involvement in quota setting, and rest breaks—and how ESAM systems should be adjusted to accommodate those best practices.
  • Update injury reporting regulations at 29 C.F.R. Part 1904, pursuant to 29 U.S.C. § 657, revising OSHA’s log of work-related injuries and illnesses (Form 300) to collect information about automated systems used in the tasks, job roles, or workplaces in which the worker was working at the time of injury or illness. Additionally, OSHA could update Form 300 to include a column identifying when injuries are musculoskeletal.23 This would allow OSHA to develop a better understanding of the precise causal mechanisms between ESAM and these injuries and inform the substantive policymaking described above.
  • Request research from the National Institute for Occupational Safety and Health, pursuant to 29 U.S.C. § 671(d), to fund and conduct further research to study ESAM’s effect on job strain and physical injury.24

Employee Retirement Income Security Act: Adverse benefits determination and disclosure

Based on the above-cited authority, the DOL could consider the following actions:

  • Update regulations at 29 C.F.R. § 2560.503-1, which implement the denial-of-claims disclosure and appeal requirements at 29 U.S.C. § 1133. The current regulations state, for example, that in the case of an adverse benefit determination by a group health plan, a participant is entitled to request a copy of any “internal rule, guideline, protocol, or other similar criterion” that was relied on in making the adverse determination.25 An updated regulation could require affirmative disclosure of a plain-language description of any algorithmic determination involved in a benefits determination, as well as the results of an equity audit conducted in a manner similar to that recommended in the OMB M-24-10 AI memo.26 Additionally, the updated regulations could clarify that the appeal process authorized by 29 U.S.C. § 1133(2) and outlined at 29 C.F.R. § 2560.503-1(h) requires that appeals of benefits denials be heard by a human. This update could come as part of the DOL’s announced review of the Employee Retirement Income Security Act (ERISA) disclosures pursuant to the Setting Every Community Up for Retirement Enhancement (SECURE) Act 2.0.27
  • Update regulations at 29 C.F.R. § 2520.102-3(l) to amend the summary of plan description to include a plain language description of any automated and algorithmic systems that the plan uses to make determinations that could “result in disqualification, ineligibility, or denial or loss of benefits,”28 as well as whether the system has been externally audited or the administrator has instituted safeguards such as opt-out mechanisms for participants who would prefer human-made determinations. This would provide some transparency to workers and advocates about the decisions that plan administrators make with the help of AI-driven systems. This update could also come as part of the DOL’s announced review of ERISA disclosures pursuant to the SECURE Act 2.0.29

Employee Retirement Income Security Act: Investment advice

Based on the above-cited authority, the DOL could consider the following actions:

  • Update regulations at 29 C.F.R. § 2550.404a-1(c), pursuant to 29 U.S.C. § 1104, to revise the investment duty of loyalty in light of the risks that AI-driven investment allocation technologies can create and potential conflicts of interest. The updated regulation could be similar to the U.S. Securities and Exchange Commission’s rulemaking proceedings that seek to prevent investment advisers from using algorithms that create conflicts of interest between the adviser and the investor’s retirement goals.30 Importantly, plan fiduciaries should be required to ensure that AI-driven investment advice or allocations are not improperly weighted toward decisions that maximize fees and commissions at the expense of retirement savers. Such regulations could also require an audit of any AI-driven or otherwise automated investment allocation technologies for the potential for conflicts of interest.
  • Issue new regulations, pursuant to 29 U.S.C. § 1104(c)(5), requiring algorithmic transparency and legibility to plan participants and beneficiaries for default asset allocations.31
  • Update the statutory transactions exemption at 29 C.F.R. § 2550.408g-1(b)(4), “Arrangements that use computer models,” to strengthen the existing auditing requirements and institute other AI-specific requirements, taking into account the DOL’s approach in the proposed revisions to the Prohibited Transaction Exemption 2020-02.32 Alternatively, or in addition to updating the exemption, the DOL could issue guidance that more fully describes the term “computer model” and identifies AI applications to which this exemption may apply.

Labor Management Reporting and Disclosure Act

Based on the above-cited authority, the DOL could consider the following action:

  • Issue a regulation or subregulatory guidance, in the form of independent guidance documents or in the LM-10 form instructions, that explains how forms of ESAM can chill workers’ exercise of their Section 7 rights under the National Labor Relations Act and when they must be reported in employers’ LM-10 forms. The use of worker surveillance to thwart organizing activities is well-documented.33 The regulation or guidance could explain how that might require employers to report their expenditures on such technologies. They could reference the memo issued by the National Labor Relations Board’s general counsel on the subject,34 as well as prior guidance from the DOL on surveillance reporting.35 Additional guidance may empower workers, unions, and labor watchdogs to report employer noncompliance to the DOL.

Worker Adjustment and Retraining Notification Act

Based on the above-cited authority, the DOL could consider the following action:

  • Update regulations at 20 C.F.R. § 639.3(i), pursuant to 29 U.S.C. § 2107(a), to explain that, in the case of a completely or primarily remote workforce, the term “single site of employment” applies to the employer’s entire workforce. In the case of algorithmic management, the DOL should clarify that all workers subject to the same or similar algorithm are considered one single site of employment. Updated regulations could also ensure that workers subject to intermittent deplatforming caused by algorithmic optimization have maximal protections possible under the Worker Adjustment and Retraining Notification (WARN) Act.

Family and Medical Leave Act

Based on the above-cited authority, the DOL could consider the following actions:

  • Update regulations at 29 C.F.R. Part 825, pursuant to 29 U.S.C. §§ 2615(a)(1) and 2654, to require legibility and transparency of automated systems36 that make any determinations bearing on the allocation or approval of FMLA leave, along with any other applicable minimum practices for rights-impacting AI from the OMB M-24-10 AI memo.37 This would implement the transparency protections recommended by the White House’s AI Bill of Rights and ensure that employers’ use of automated systems does not unlawfully restrain workers’ exercise of their rights under the FMLA. Because FMLA determination algorithms are likely bound up in other human resource management systems, this proposal could also provide transparency of those benefits processes as well. Specifically, these updated regulations should require:
    • At 29 C.F.R. § 825.301, legibility and transparency around use of automated systems to make FMLA designations
    • Legibility and transparency around use of automated systems to review, request, or otherwise process certifications under 29 U.S.C. § 2613
    • Legibility and transparency around use of automated systems to provide eligibility notices, at 29 C.F.R § 825.300(b); rights and responsibilities notices, at 29 C.F.R. § 825.300(c); and designation notices, at 29 C.F.R. § 825.300(d)
    • At 29 C.F.R. § 825.302, legibility and transparency around use of automated systems for employees to provide notice of the use of leave or to transmit information around scheduling of intermittent leave under 9 U.S.C. § 2612(b) and (e)
  • Update regulations by modifying 29 C.F.R. § 825.220, pursuant to 29 U.S.C. §§ 2615(a)(1) and 2654, to prohibit employers from using FMLA data as inputs to any automated management system that may make an employment decision based, in part, on an employee’s use or nonuse of FMLA leave. This would reduce employers’ ability to weaponize employees’ data against them to retaliate for using FMLA leave. Under these recommended updated regulations, the automated management system must strictly segregate and keep confidential any information provided for FMLA certification pursuant to 29 C.F.R. § 825.500(g).
  • Update subregulatory guidance under 29 C.F.R. § 825.301(a) prohibiting automated systems from using information other than that received from the employee or the employee’s authorized spokesperson in designating FMLA leave pursuant to 29 C.F.R. § 825.301(a). Existing regulation already prohibits the conduct for employers and would also apply to automated systems used by employers, but additional clarification is essential to restrict automated systems that would improperly combine data sources.


  1. Veena Dubal, “The House Always Wins: the Algorithmic Gamblification of Work,” LPE Project, January 23, 2023, available at
  2. For example, in one 2017 lawsuit against Uber, a class of drivers alleged that there was a discrepancy between their contracted rate (a fixed proportion of an often-inflated rider’s fare payment) and their actual rate (a backend mileage- and time-based rate), which resulted in systematic underpayment and breach of contract. See Dulberg v. Uber Technologies Inc. and Rasier, class action complaint, U.S. District Court for the Northern District of California, 3:17-cv-00850 (February 21, 2017), available at
  3. Many are misclassified as independent contractors and may be beyond the reach of the FLSA, though the DOL’s new rulemaking on independent contractor versus employee status will reduce the severity of misclassification.
  4. See Dubal, “The House Always Wins: The Algorithmic Gamblification of Work.” See also Zephyr Teachout, “Surveillance Wages: A Taxonomy,” LPE Project, November 6, 2023, available at
  5. Office of Information and Regulatory Affairs, “Right to Know Under the Fair Labor Standards Act,” available at (last accessed May 2024); Office of Information and Regulatory Affairs, “Right to Know Under the Fair Labor Standards Act,” available at (last accessed May 2024).
  6. See U.S. Department of Justice, “Report to Congress on the Use of Administrative Subpoena Authorities by Executive Branch Agencies and Entities” (Washington: 2002), available at for a thorough discussion of administrative subpoena powers held by executive agencies.
  7. Legal Information Institute, “29 C.F.R. § 785.48(b) – Use of time clocks,” available at (last accessed May 2024).
  8. Legal Information Institute, “29 C.F.R. § 785.47 – Where records show insubstantial or insignificant periods of time,” available at (last accessed May 2024).
  9. Elizabeth Chika Tippett, “How Employers Profit from Digital Wage Theft Under the FLSA,” American Business Law Journal 55 (2) (2018): 315–401, available at; Charlotte S. Alexander and Elizabeth Tippett, “The Hacking of Employment Law,” Missouri Law Review 82 (4) (2017): 973–1022, p. 990, available at
  10. See S. Department of Labor, “Final Rule, Federal-State Unemployment Compensation Program; Unemployment Insurance Quality Control Program,” Federal Register 52 (171) (1987): 33506–33522, available at
  11. See, for example, Robert N. Charette, “Michigan’s MiDAS Unemployment System: Algorithm Alchemy Created Lead, Not Gold,” IEEE Spectrum, January 24, 2018, available at
  12. Shalanda D. Young, “M-24-10 Memorandum for the Heads of Executive Departments and Agencies: Advancing Governance, Innovation, and Risk Management for Agency Use of Artificial Intelligence” (Washington: Office of Management and Budget, 2024), Appendix I, 2.l., p. 33, available at
  13. Executive Office of the President, “Executive Order 14110: Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence,” Federal Register 88 (210) (2023): 75191–75226, at Section 7.2(b), available at
  14. Young, “M-24-10 Memorandum for the Heads of Executive Departments and Agencies: Advancing Governance, Innovation, and Risk Management for Agency Use of Artificial Intelligence,” at 5.c., pp. 15–24.
  15. Ibid., at 5.c.iv.A.,5.c.iv.B and 5.c.i.v.C, pp. 17–23.
  16. Lauren Rosenblatt, “Fine with fines? Amazon isn’t making enough changes to protect warehouse workers, Washington state says,” Tech Xplore, March 29, 2022, available at
  17. Washington State Department of Labor and Industries, “Citation and Notice: Amazon Com Services,” May 4, 2021, available at
  18. Office of Technology Assessment, “The Electronic Supervisor: New Technology, New Tensions” (Washington: U.S. Government Printing Office, 1987), available at See, generally, Governing for Impact and Center for Democracy and Technology, “Memos to the White House and federal agencies,” April 3, 2023, available at
  19. David Michaels and Jordan Barab, “The Occupational Safety and Health Administration at 50: Protecting Workers in a Changing Economy,” American Journal of Public Health 110 (5) (2020): 631–635, available at; U.S. Government Accountability Office, “Report to Congressional Requesters: Workplace Health and Safety: Multiple Challenges Lengthen OSHA’s Standard Setting” (Washington: 2012), available at
  20. Governing for Impact and Center for Democracy and Technology, “Memos to the White House and federal agencies.”
  21. U.S. Department of Labor, “US Department of Labor finds Amazon exposed workers to unsafe conditions, ergonomic hazards at three more warehouses in Colorado, Idaho, New York,” Press release, February 1, 2023, available at While making up roughly one-third of the national warehouse workforce, Amazon workers account for 49 percent of all warehouse injuries in the country. See Mitchell Clark, “Amazon workers made up almost half of all warehouse injuries last year,” The Verge, April 12, 2022, available at
  22. Governing for Impact and Center for Democracy and Technology, “Memos to the White House and federal agencies.”
  23. See U.S. Department of Labor, “US Labor Department’s OSHA temporarily withdraws proposed column for work-related musculoskeletal disorders, reaches out to small businesses,” Press release, January 25, 2011, available at; Occupational Safety and Health Administration, “Occupational Injury and Illness Recording and Reporting Requirements” (Washington: U.S. Department of Labor, 2003), available at
  24. Governing for Impact and Center for Democracy and Technology, “Memos to the White House and federal agencies,” 02-1.
  25. Legal Information Institute, “29 C.F.R. § 2560.503-1(g)(1) – Claims procedure,” available at (last accessed May 2024).
  26. Young, “M-24-10 Memorandum For The Heads Of Executive Departments And Agencies: Advancing Governance, Innovation, and Risk Management for Agency Use of Artificial Intelligence,” at 5.c.v.A., p. 21.
  27. Office of Information and Regulatory Affairs, “Improving Participant Engagement and Effectiveness of ERISA Retirement Plan Disclosures,” available at (last accessed May 2024).
  28. Legal Information Institute, “29 U.S.C. § 1022(b) – Summary plan description,” available at (last accessed May 2024).
  29. Office of Information and Regulatory Affairs, “Improving Participant Engagement and Effectiveness of ERISA Retirement Plan Disclosures.”
  30. U.S. Securities and Exchange Commission, “Fact Sheet: Conflicts of Interest and Predictive Data Analytics” (Washington: 2023), available at
  31. Amy Caiazza, Rob Rosenblum, and Danielle Sartain, “Investment Advisers’ Fiduciary Duties: The Use of Artificial Intelligence,” Harvard Law School Forum on Corporate Governance, June 11, 2020, available at
  32. Employee Benefits Security Administration, “Proposed Amendment to Prohibited Transaction Exemption 2020-02,” Federal Register 88 (212) (2023): 75979–76003, available at; Fred Reish, “The New Fiduciary Rule (8): Special Issues—Robo Advice and Investment Education,” JD Supra, December 4, 2023, available at
  33. See, for example, Jo Constantz, “‘They Were Spying On Us’: Amazon, Walmart, Use Surveillance Technology to Bust Unions,” Newsweek, December 13, 2021, available at; Indigo Oliver, “McDonald’s spies on union activists – that’s how scared they are of workers’ rights,” The Guardian, March 2, 2021, available at,the%20Chicago%20and%20London%20offices%E2%80%9D.&text=This%20comes%20after%20years%20of,unionization%20of%20their%20o.
  34. National Labor Relations Board, “NLRB General Counsel Issues Memo on Unlawful Electronic Surveillance and Automated Management Practices,” Press release, October 31, 2022, available at
  35. U.S. Department of Labor, “OLMS Fact Sheet: Form LM-10 Employer Reporting Transparency Concerning Persuader, Surveillance, and Unfair Labor Practice Expenditures” (Washington: 2022), available at; Jeffrey Freund, “How We’re Ramping Up Our Enforcement of Surveillance Reporting,” U.S. Department of Labor Blog, September 15, 2022, available at
  36. See, for example, Ecotime by HBS, “Ensure Time-Savings and Compliance With FMLA Software,” available at (last accessed February 2024).
  37. Young, “M-24-10 Memorandum for the Heads of Executive Departments and Agencies: Advancing Governance, Innovation, and Risk Management for Agency Use of Artificial Intelligence.”

The positions of American Progress, and our policy experts, are independent, and the findings and conclusions presented are those of American Progress alone. A full list of supporters is available here. American Progress would like to acknowledge the many generous supporters who make our work possible.


Reed Shaw

Policy Counsel

Governing for Impact


Technology Policy

Our team envisions a better internet for all Americans, advancing ideas that protect consumers, defend their rights, and promote equitable growth.

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.