Data Privacy, Ethics & Society · Module 07 / 14 · Phase 3: Surveillance, power, asymmetry
Tonight's reading: NELP, When “Bossware” Manages Workers (July 2025)
6:00 to 9:50, three ways of working. Click any block to jump there.
Count the room. Tap + for each hand.
Mouse jiggler, scheduled emails, keeping a chat window “active.”
Consent to the tracking, or don't earn a living.
The easy answer
“Workplace surveillance is wrong.” It won't survive contact. Some monitoring catches a fatigued driver before a crash.
Tonight's job
Figure out which monitoring can be defended on the evidence, and where that defense collapses. Then notice who gets watched to the second.
What bossware is, how common it is, and what changes when watching turns into judging.
NELP uses “bossware” as an umbrella for two linked things. The first collects. The second decides.
1 · Digital surveillance
2 · Automated decision system
Decision
Source: Tung, Sonn, Pinto & Boxerman, NELP, “When ‘Bossware’ Manages Workers,” July 2025, pp. 4–5. Scope: managing workers after hire; hiring algorithms are out of scope.
Not just Amazon and Uber. NELP points to hundreds of smaller vendors embedded in staffing agencies, temp firms and sector platforms.
Data: Hertel-Fernandez, National Survey of Workplace Technology, summer 2024, nationally representative, ~1,300 workers (Washington Center for Equitable Growth). Category figures as reported in his March 2026 follow-up. The 2024 write-up reports “any form” as 68%.
Adapted from NELP Section 1A and Appendix 1, “Increased Employer Powers with Bossware.”
Remote worker, public comment to the White House OSTP, 2023
“Once the lengthy verification login process is done, I am on camera all day.… no further information is told to us about all violations the AI is looking for and what else is the AI tracking.”
Quoted in NELP, p. 5.
Click a tile to show NELP's example.
NELP Section 1, A–H, pp. 8–15.
Category 1
Knowledge workers under keystrokes, screenshots, webcams, activity scores. A condition of the job.
Signature response: gaming. Mouse jigglers, scheduled emails, productivity theater.
Category 2
Warehouses and gig platforms. The system scores, ranks and disciplines, often with no human in the loop.
Signature feature: judgment. And the sharpest class line.
10 minutes of setup, then 20 minutes of steelman lab in groups.
The realistic alternatives
The strongest case for monitoring is usually against the biased manager, not against nothing.
Notice the grade on each before you notice the size of the number.
Reduction in detected fatigue events, by who receives the alert.
This is the hardest fact of the night: the safety mechanism and the surveillance mechanism are the same mechanism.
Two cautions: these are the vendor's own figures, and they count fatigue events detected, not crashes.
Seeing Machines Ltd., submission to Australia's National Transport Commission (HVNL review), Nov 2020; also submission to NSW Parliament. Vendor-produced.
Groups of 3–4. Each group gets one practice. Build the strongest honest case for it.
Practices (count off 1–6)
Deliverable: one sentence on the board
“[Practice] benefits who by preventing harm , which matters because best evidence , and it beats the realistic alternative .”
Rule: no criticism yet. If you catch yourself arguing against it, write the objection on a sticky note and set it aside for Part 3.
The rubric's word choice is deliberate. Your paper has to tell the difference.
Higher rungs are more trustworthy, not automatically right. A meta-analysis can wash out a tool that works in one narrow setting.
measured it? Do they profit if it's big?
was measured? The outcome, or a proxy for it?
what? Before/after, or against a comparison group?
from the ledger? Whose costs weren't counted?
Try it: “Amazon's van cameras cut accidents 48%.”
Who: Amazon. What: accidents (good, a real outcome). Versus: unclear, before/after in a pilot. Missing: drivers told Motherboard the cameras flagged them for checking mirrors or being cut off, costing them pay.
The theft the system caught was tiny. The revenue jump was 120 times bigger. So what produced it?
The authors test explanations, including workers selling harder once they know they're watched and a general sense of more oversight. Notice: this monitored sales transactions, not bodies, faces or keystrokes.
Pierce, Snow & McAfee, “Cleaning House,” Management Science 61(10), 2015, pp. 2299–2319. Per-location weekly figures.
Dots are correlations (ρ); lines are 95% confidence intervals. Lines crossing 0 mean no clear effect.
Stress goes up, more so when monitoring is invasive. Transparency improves how workers feel about it.
Performance isn't plotted: the authors report no significant overall effect.
Ravid, White, Tomczak, Miles & Behrend, “A meta-analysis of the effects of electronic performance monitoring on work outcomes,” Personnel Psychology 76 (2023), 5–40.
Amazon's share of U.S. warehouse employment vs. its share of the industry's serious injuries, 2022.
Strategic Organizing Center (a union coalition) analysis of Amazon's OSHA filings: 2022 figures via AP, Apr 2023; 2024 figures from SOC, May 2025. Amazon disputes the “serious injury” (DART) measure. A Senate HELP Committee investigation (Dec 2024) also tied injuries to pace pressure.
Workers with more intensive productivity monitoring were substantially more likely to report
Biggest harms
came where workers said monitoring was used to discipline them or suppress their voice.
GAO (2024) found the same tools can raise workers' sense of safety in some settings and raise injury risk when they push pace.
Hertel-Fernandez, Washington Center for Equitable Growth, Oct 2024 (survey; correlational). U.S. GAO, Digital Surveillance of Workers, GAO-24-107639, Aug 2024.
“Measuring productivity with mouse movements is like using a sundial as a stopwatch.”
Jared Spataro, Microsoft, Fortune, Sept 2022
Goodhart in action: in May 2024 Wells Fargo fired more than a dozen wealth-management staff over “simulation of keyboard activity creating impression of active work.”
Microsoft Work Trend Index, Sept 2022: 20,000 people, 11 countries. Wells Fargo: FINRA disclosures reported by Bloomberg, June 13, 2024.
When we're back: evidence court. Put your steelman on trial.
Back in your groups. Put your own steelman on trial.
As each group reports, move its practice. The room can overrule.
A second lens, justice or virtue, and the problem of consent.
Illustrative, built from cases in NELP and the course page. Each tick is one moment of measurement; it is not survey data.
Share who are workers of color. NELP p. 12, on occupations where misclassification and bossware are common (construction, delivery, home care, agriculture, personal care, ridehail, janitorial).
“Every move you make is being tracked, so you feel uncomfortable, and you're not able to really communicate with your co-workers about the unsafe work environment.”
Wendy Taylor, Amazon worker, to The Guardian; quoted in NELP p. 14.
Virtue question for the paper: what kind of worker, and what kind of manager, does this practice train people to become?
Pick a worker, then argue about the dials
A discussion device, not a formula. The dials are the argument.
A driver-facing fatigue camera
The sensor never changed. Every switch is a design choice an employer makes.
When we're back: four corners, then your own writing.
NELP's proposed bright-line bans
Selected models in NELP's Figure 1
Solid: enacted. Dashed: proposed as of July 2025; status may have changed since.
20 minutes of quiet drafting. Leave with a thesis and your strongest objection.
Format
~1.5–2 pages, double-spaced, PDF or Word. Cite NELP (2025) plus at least one more source. Submit on Canvas.
The warning
You will be marked down for skipping straight to condemnation. Half the points are earned before you criticize anything.
Outline to copy
1 STEELMAN Practice + category: Who benefits / harm prevented: Best evidence (rung on the ladder): Compared with what: 2 EVIDENCE TEST (NELP + 1 more source) Who measured / what / versus what / what's missing: Verdict: delivers / depends on design / only claims to 3 SECOND LENS (justice OR virtue) Who is watched to the second, who isn't, and why: 4 CONSENT Alternatives? Information? Negotiable? Refusable? Coerced, or constrained? 5 LANDING Where I come down: Strongest objection: My answer:
Partner swap · 3 min
Exit ticket · one sentence
The practice I'll defend, and the exact point where its defense collapses.