Automated Decision Safeguards
Legal rules that constrain how employers use AI to hire, monitor, discipline, and dismiss workers, mandating human oversight, transparency, and limits on the data that trains workplace AI.
What it is:
Automated decision-making safeguards are statutory rules governing how employers may use AI systems to make or inform decisions about workers. As algorithmic tools spread across hiring, performance management, and termination, these safeguards establish binding limits on that use. This can include requirements that a human review consequential decisions before they take effect, that workers be told when and how AI is being used to evaluate them, that AI systems be audited for bias, and that the personal data workers generate on the job not be repurposed to train the systems that may replace them. Unlike liability rules, which allocate financial responsibility after harm occurs, these are ex ante conduct rules that define what an employer may and may not do in the first place, regardless of whether a specific harm can later be proven.
These safeguards are the legislated counterpart to the protections that unions negotiate through collective bargaining (see Union & Bargaining Rights). Where bargaining secures advance notice, algorithmic transparency, and data protections for workers who have the leverage to negotiate them, statutory safeguards extend equivalent rights to the far larger share of the workforce that is not unionized. The two approaches are complementary: bargaining can be more responsive and go further in a given workplace, while legislation sets a floor that applies universally and does not depend on workers first organizing.
The case for these rules rests on the asymmetry between employer and worker in an AI-mediated workplace. A worker disciplined or dismissed by an automated system may never know that an algorithm drove the decision, what data it used, or whether it was accurate. Employers, meanwhile, can adopt these systems at scale to cut costs with little obligation to explain or justify the outcomes. Automated decision-making safeguards aim to preserve human accountability and due process at the point where AI meets a worker's livelihood, ensuring that consequential judgments about people remain explicable and contestable rather than opaque and automatic.
The challenge:
The central difficulty is drawing the line between AI use that warrants regulation and the vast range of ordinary software that now contains some automated component. Definitions of 'automated decision system' or 'workplace surveillance tool' that are too broad risk including routine productivity software; too narrow, and employers can evade the rules by keeping a nominal human "in the loop" who rubber-stamps algorithmic outputs. California's Governor Newsom vetoed an earlier version of the No Robo Bosses Act (SB 7) in October 2025 on these grounds. Enforcement is also challenging when the underlying systems are opaque; a worker who suspects an algorithm treated them unfairly typically cannot see the system that did it, which is why laws like California's SB 947 pair substantive limits with disclosure rights, a written post-use notice, and a private right of action allowing affected workers to sue for damages.
Recommended Reading:
Real-world precedents:
The EU's AI Act classifies AI systems used in employment — recruitment and selection, and for decisions affecting promotion, termination, and performance monitoring — as "high-risk" under Annex III. High-risk classification triggers obligations on employers deploying these systems, including human oversight, transparency to affected workers, and continuous monitoring. The full obligations for these employment systems are scheduled to apply from December 2027, with penalties for deployer non-compliance reaching up to €15 million or 3% of global turnover.
New York City's Local Law 144, effective January 2023, was the first US law to directly regulate AI in hiring. It requires employers using automated employment decision tools for hiring or promotion to commission an independent bias audit within the prior year, publicly post a summary of the results, and notify candidates that such a tool is being used. A December 2025 New York State Comptroller audit found the city's enforcement of the law had been ineffective.
California's AB 1883, awaiting Governor Newsom's signature or veto by the September 30 deadline, would prohibit employers from using an AI-powered workplace surveillance tool to recognize a worker's emotional state or to collect data generated by measuring a worker's nervous system.
SB 947, the "No Robo Bosses Act," if enacted, from July 1, 2027 would bar employers from relying on an automated decision system (ADS) to make disciplinary or termination decisions unless a human corroborates the output with supporting evidence, give affected workers the right to a description of their own data used and a written post-use notice, prohibit using an ADS to infer protected status, and provide anti-retaliation protection enforced through a $500-per-violation penalty and a private right of action.
SB 951, if enacted, would amend California's WARN Act to require that layoff notices caused in whole or substantial part by AI or automation disclose the job functions being automated and the type of system responsible, and would require the state to publish quarterly summaries of AI-driven displacement.