For almost four years Stephen Normandin delivered packages for Amazon in Phoenix. He enjoyed the work and was good at it. A 63-year-old Army veteran, he ran his routes out of a 2002 Toyota Corolla, earning excellent ratings and learning the local routes well enough that he was asked to train others. In August 2020, his managers at Amazon changed his taskload, now handing him predawn stops at locked apartment complexes and delivery lockers that would not open. As he troubleshot the snags for 30 minutes with Amazon’s internal support hotline, his ratings sank.
Weeks later he was fired via a letter which cited his falling delivery efficiency scores. His manager had not been involved; the software decided on its own. “There are a lot of things the algorithms don’t take into account,” Normandin told Bloomberg, “and the right hand doesn’t know what the left hand is doing.”
What happened to Normandin wasn’t an isolated instance. It was an example of how modern workforce management is designed to operate. Your boss used to keep tabs on you, take responsibility for your growth in your role, and decide your long-term status. Now every phase of employment — hiring, monitoring, and firing — is increasingly handled by automated systems of surveillance and evaluation. The trend of digital surveillance escalated during the remote work explosion of Covid, but far from receding with the pandemic, it has since only increased. You may work as much for an algorithm (or several) as for any human being.
In a 2025 survey of 1,500 American employers by the security firm ExpressVPN, 61 percent said they use AI to measure productivity, and 67 percent said they collect biometric data to track behavior and attendance. Researchers have a name for the shift: algorithmic management, meaning the use of software rather than supervisors to assign work, rate it, and decide who is falling behind. A human manager can watch a handful of people. Software can — in at least some senses — watch everyone, all the time, and forget nothing. It’s like a basketball game without referees where the outcome is decided by instant replay.
Americans do not approve of this approach. In a 2023 Pew Research poll, 71 percent opposed letting AI make the final decision in hiring, and by a margin of 55 to 14 percent of those surveyed rejected the idea of a machine deciding who should be fired. The lack of support is across the board; Americans disapprove of surveillance, evaluation, and dismissal by AI.
The law is only slowly catching up. Europe’s AI Act treats software that manages and evaluates workers as high risk, its most tightly regulated category (aside from “unacceptable risk,” which is banned entirely). In March of 2026 the EU’s Internal Market and Civil Liberties committee proposed to delay those rules, and its Parliament backed the postponement in August 2026. In the United States, the Consumer Financial Protection Bureau warned in 2024 that the secret algorithmic scores used to hire, promote, and fire people can count as consumer reports, which give workers the right to see their file and dispute what is in it. A few short months later, the bureau withdrew the guidance. The debate is stalled at the federal level.
That brings the fight to the states. State senator Jerry McNerney’s No Robo Bosses Act would have put a human being back in the loop, barring employers from relying solely on automated systems to discipline or fire workers. Newsom vetoed it in October 2025, calling the restrictions overly broad and noting that privacy regulations then in the works already covered part of the ground. They now do: California employers using automated decision-making technology for employment decisions must run risk assessments and give notice before use, with obligations starting April 1, 2026. Workers got a version of what the vetoed bill promised, written by privacy regulators instead of labor law. McNerney also came back with a revised bill that drops the advance notice Newsom objected to and requires notice after the fact instead; this version has passed the state senate.
Other states have moved further. Colorado will require notice to anyone hurt by an algorithm-influenced decision starting in 2027. Connecticut has made an automated tool no defense to a discrimination complaint as of October 1, 2026. And New York is weighing the Bossware and Oppressive Technologies Act, which would limit electronic monitoring and require notice before a machine renders a verdict on someone’s livelihood. It sits in committee in both chambers.
The legality of worker surveillance has nothing to do with how powerful the machines are. The question is whether a worker has a right not to be judged by code as opposed to a colleague who can be shown the evidence to discuss before rendering judgement. Normandin was denied that right. The program ran, made a judgment, and sent an email before locking him out of the system.
When software decides you are finished, who is left to appeal to? For a growing number of Americans, the answer is a name on an email sent by the same bot that surveilled them.
Note (8/14/26): This piece has been updated and slightly revised from its original version.
Reuben Steiger is a writer and entrepreneur based in Princeton, NJ. Over a 25-year career he has helped start companies including Second Life and has led global innovation for companies including Interpublic and Omnicom. His current focus is the scaling and adoption of AI technologies. He collects books about the future.
Firing by AI? Where's the safety net? Where's the recourse?
Not every problem or issue requires a high-tech solution. This article is a good example of that. With that being said, however, I'm glad I'm retired.