Why AI Should Not Watch Screens or Automate Dismissals

In the past year, a major technology company scaled back an internal tool that recorded employees' keystrokes and clicks after staff objected. Another faces a federal lawsuit alleging that activity data helped decide who would be laid off. California now requires a human to verify any AI-assisted dismissal.

Each case points to the same lesson: when AI is used to watch people and decide their fate, it erodes the performance it was meant to protect. This article sets out what the evidence shows, where companies have reversed course, and what Cairo measures instead.

The gap leaders are trying to close

The pressure behind workplace surveillance is real. In Microsoft's Work Trend Index, 85% of leaders said hybrid work makes it hard to be confident their people are productive. In the same survey, 87% of employees said they were.^[1]

That gap is costly. Leaders who cannot see the work tend to reach for the signals that are easiest to capture: time online, keystrokes, clicks, messages sent. Monitoring software promises to turn those signals into certainty, and AI promises to act on them at scale.

The problem is that those signals describe activity. They say little about judgment, quality or the problem solved before it reached a customer. An organization that manages by them ends up optimizing for what can be counted and losing what cannot.

What surveillance actually produces

The most complete review to date combined 94 samples and more than 23,000 workers. It found no evidence that electronic monitoring improves performance. It did find higher stress, and the broader the monitoring, the more counterproductive behavior it was linked to.^[2] The mechanisms are well documented.

Stress reduces the capacity to think

Even mild stress that a person cannot control impairs the prefrontal cortex, the region that handles reasoning, planning and judgment.^[3] A team that knows every click and pause is recorded works under that condition all day. It makes more errors and solves fewer problems.

Loss of control reduces initiative

When an algorithm can issue warnings or recommend dismissals, employees lose any sense of control over their own stability. Job insecurity is consistently associated with lower performance and with less of the voluntary effort that goes beyond a job description.^[4]

People produce the signal the system rewards

Software measures what is easy to measure. Employees respond by generating those signals. In 2024, a major U.S. bank dismissed more than a dozen employees for simulating keyboard activity.^[5] Deloitte has described the wider pattern as productivity theater: mouse jigglers, empty emails, visits scheduled to satisfy a metric.^[6]

Monitoring invites the behavior it was meant to stop

In research published in Harvard Business Review, monitored employees were more likely to take unapproved breaks, disregard instructions and work slowly on purpose.^[7] When individual numbers decide a person's future, helping a colleague becomes a cost to one's own metrics, and cooperation declines.

Trust is the variable underneath all four. In Deloitte's research, 80% of employees with high trust in their employer feel motivated to work, compared with fewer than 30% of those with low trust.^[6] Surveillance spends that trust on data that has not been shown to improve performance.

Where companies reversed course

The clearest evidence comes from companies that tried these approaches and then retreated, or now defend them in court.

The pattern is consistent. Each system measured activity because activity is easy to capture. Each produced legal exposure, reputational cost or worse service. Where companies corrected course, the correction brought human judgment back into the decision.

Regulators are drawing the same line

Law is following the evidence. California's governor signed three workplace AI bills on September 30, 2026:^[14][15]

• SB 947, the No Robo Bosses Act. Employers may not rely solely on an automated decision system to discipline or dismiss a worker. A human must corroborate the decision with other evidence, and affected workers must be told when such a system was used. It takes effect July 1, 2027.
• AB 1883. Prohibits AI-driven workplace surveillance tools from recognizing or inferring a worker's emotional state, and from collecting neural data. It takes effect January 1, 2027.
• AB 1331. Limits surveillance in workplace bathrooms and lets employees leave monitoring devices behind during breaks.

The European Union went further earlier. Since February 2025, the EU AI Act has prohibited emotion recognition in the workplace, with narrow exceptions for medical and safety reasons.^[16]

For employers, the direction is clear. Tools that watch people and decide for them now carry regulatory risk on top of their operational cost.

What Cairo measures, and why

Cairo starts from a simple premise: organizations improve when their people improve. Measurement serves that purpose. It identifies what each person contributes, where they can grow and how to develop them in the work they actually do.^[22]

Each profile draws on four groups of indicators:

The data comes from activities people take part in: practice scenarios built on real workplace dilemmas, short assessments within learning modules, progress through courses, and performance information the organization already documents.

From profile to contribution

Cairo does not return a list of scores. It synthesizes them into a working archetype and a contribution thesis: a short statement of the person's principal strength and their priority area for growth. One profile might read: strongest at surfacing the critical fact that unlocks a decision; next step, delegating more strategically.

That emphasis on strengths is deliberate, and the business case for it is well established. When employees strongly agree that their manager focuses on their strengths, 67% are engaged. When they strongly disagree, 2% are.^[18] Across 1.2 million employees, teams that received strengths-based development showed 14% to 29% higher profit, 10% to 19% higher sales and markedly lower turnover.^[17]

Knowing where someone adds the most value changes everyday decisions: who leads which project, how work is divided within a team, which responsibilities stretch a person and which waste their talent.

From contribution to development

Each growth area connects to a learning path. People take short courses, around ten minutes each, and rehearse difficult conversations, cross-team conflicts, negotiations and feedback in AI-guided scenarios before facing them at work.

This design responds to the central weakness of corporate training. People apply an estimated 10% to 30% of what they learn, and transfer depends on three conditions: content that is useful for the role, an early opportunity to use it and practice that resembles real situations.^[19][20] A path built from a person's own profile and work meets all three. Cairo then tracks progress over time, so the investment in training can be tied to visible change in how people work and collaborate.

From people to teams

Cairo Index aggregates profiles at team level. Leaders see how well a team collaborates, its collective confidence and resilience, its engagement, and early signs that workload is outpacing the team's capacity. That view lets them adjust pace and priorities before performance or retention suffers.

Cairo Rosetta, a behavior model in development, is designed to help leaders anticipate how teams may respond to a change or a message, so they can prepare it well.

Where human judgment stays

AI is well suited to work that benefits from scale: mapping skills across a large organization, spotting where a team needs development, preparing a leader before a difficult conversation. It is poorly suited to decisions that turn on context, intent and fairness.

A dismissal, a promotion, a disciplinary measure or a restructuring belongs in the second group. These decisions affect a person's livelihood and a team's trust in its leadership. They require someone who can weigh circumstances a dataset does not hold, explain the reasoning and answer for the outcome.

Uninformed decisions carry their own cost. In Gallup's research, 42% of employees who left voluntarily said their organization could have prevented it, and 45% said no leader had discussed their work or future with them in the three months before.^[21] Leaders who decide without understanding their teams create the conflict and mistrust that drive those departures.

The role of AI is to close that gap: to give leaders a clearer picture of their people, so that the decisions they make are informed, explainable and their own.

The bottom line

Productivity and culture show up in the same results. The most valuable use of AI with people is helping each one do more of what they do best, and grow where it matters for the work. Organizations that understand how their people think, develop them in the work they actually do and keep decisions about them in human hands improve both.

Want to see what that looks like for your team? Create your Cairo workspace.

Sources

1. Microsoft, Work Trend Index: Hybrid work and productivity paranoia, 2022.
2. Ravid, D. M., White, J. C., Tomczak, D. L., Miles, A. F., & Behrend, T. S., A meta-analysis of the effects of electronic performance monitoring on work outcomes, Personnel Psychology, 2023.
3. Arnsten, A. F. T., Stress signalling pathways that impair prefrontal cortex structure and function, Nature Reviews Neuroscience, 2009.
4. Jiang, L., & Lavaysse, L. M., Cognitive and affective job insecurity: A meta-analysis, Journal of Management, 2018.
5. Bloomberg Law, Wells fires over a dozen for “simulation of keyboard activity”, 2024.
6. Deloitte Insights, Workplace monitoring and the lack of trust in the workplace, 2023.
7. Thiel, C., Bonner, J. M., Bush, J., Welsh, D., & Garud, N., Monitoring employees makes them more likely to break rules, Harvard Business Review, 2022.
8. CBS News, 26 Meta workers sue over alleged AI-aided layoffs, July 2026.
9. HR Grapevine, Meta scales back AI employee monitoring tool after staff backlash, June 2026.
10. Legal 500, AI on the layoff list: Meta lawsuit and Workday rulings, 2026.
11. Tech.co, Klarna reverses AI customer service overhaul, May 2025.
12. EPIC, HireVue halts use of facial analysis, 2021.
13. Silicon Republic, Barclays scraps tech that warned employees to “avoid breaks”, 2020.
14. HR Dive, California revamps AI protections for workers in flurry of bill signings, October 2026.
15. Recording Law, California enacts workplace surveillance laws AB 1331, AB 1883, 2026.
16. Lewis Silkin, Understanding the EU AI Act's prohibited practices, 2025.
17. Gallup, Strengths-based employee development: The business results, 2016.
18. Gallup, Q12 meta-analysis, 11th edition, 2024.
19. Grossman, R., & Salas, E., The transfer of training: What really matters, International Journal of Training and Development, 2011.
20. Weinbauer-Heidel, I., & Ibeschitz-Manderbach, M., What Makes Training Really Work, 2018.
21. Gallup, 42% of employee turnover is preventable but often ignored, 2024.
22. Cairo, How Cairo works: The method behind people intelligence.