# Why employees hide their AI use, and what leaders should do about it

Explainer · By Nadi Haque, Founder, Nolli · Published 2026-10-17 · Updated 2026-10-17 · Prices and features checked 2026-10-16

Canonical page: https://withnolli.com/blog/why-employees-hide-their-ai-use

**Short answer.** Employees hide AI use because the company has not made it safe or worth sharing. In the surveys I read, they cite fear of looking lazy or less competent, fear of being replaceable or handed more work, and unclear rules. Bans do not stop it. I would write down what is allowed, answer the first person who shares with a question, and give credit for uses others adopt.

## Key takeaways

1. In five surveys between 31% and 57% of AI users or employees hold something back, but each survey asked something different, so read the figures as a pattern and not as a rate.
2. The fear is not paranoid: in four experiments with 4,439 people, observers rated people helped by AI as lazier and less competent, so a clearer policy alone will not end the silence.
3. Start with a one-page rule on what is allowed, a good first reply from managers, and credit for people whose methods others adopt, and skip the crackdown.

People hide their AI use because their company gave them good reasons to. I do not think that is a character flaw, and I do not think it is mainly a compliance problem. It is a design failure. Nobody built the place where an employee can say what they did with AI and come out ahead.

On 2026-10-16 I read six survey reports, four research papers and one essay to see what the evidence says. A large minority of AI users hold something back. The reasons they give are mostly about being judged, with unclear rules second. The evidence for any fix is thin, and I will say where.

My position, so you can argue with it: write down what is allowed, answer the first person who shares with a question and not with more work, and give credit when someone’s method gets used by others. Skip the crackdown. The order I would do it in is near the end. For the wider topic, see our guide to shadow AI.

## How many hold back, and why you cannot average the surveys

Five surveys asked some version of the question. They asked different things of different people, so read the bars as a pattern.

**Share who hold back, by survey and by what each one asked**

- KPMG and University of Melbourne: admit non-transparent use, such as avoiding saying they used AI: 57%
- Microsoft and LinkedIn: reluctant to admit using AI for their most important tasks: 52%
- Slack: would be uncomfortable telling their manager they used AI for at least one of 11 tasks: 48%
- PagerDuty: cite avoiding scrutiny from managers or leadership as a reason they would hide AI use: 33%
- Deloitte UK: use generative AI for work without their employer’s knowledge: 31%

Source: Five different questions, so the bars compare a pattern and not a rate. Bases are in the table below; each survey is linked in Sources., #sources

What each survey asked, of whom, and who is behind it

| Survey | Who was asked, and when | What the percentage is of | Who is behind it |
| --- | --- | --- | --- |
| KPMG and University of Melbourne | 48,340 people in 47 countries, of whom 32,352 were employees. Published April 2025. | Employees. The report’s wording is “over half (57%)” admit non-transparent use, which includes presenting AI work as their own. | University researchers; KPMG co-funded the study |
| Microsoft and LinkedIn | 31,000 knowledge workers in 31 markets, 2024-02-15 to 2024-03-28, by Edelman Data & Intelligence. | People who use AI at work. | Microsoft, which sells Copilot |
| Slack | More than 17,000 desk workers worldwide. Published November 2024. | All desk workers asked. It is a “would be” question, not a record of what anyone did. | Slack, a workplace software company |
| PagerDuty | 1,250 office professionals outside IT at companies with $500 million or more in revenue, in Australia, Japan, the UK and the US, 2026-04-09 to 2026-04-20, by Wakefield Research. | People who have used AI for work. | PagerDuty, a software vendor |
| Deloitte UK | 25,000 UK workers aged 18 to 70, 2026-05-07 to 2026-06-10, by Ipsos UK. Published 2026-09-16. | The 16,749 respondents who use generative AI for work. | Deloitte UK’s own research, from its press office |

Every figure is self-reported. Four of the five come from companies that sell software or advice in this area, and KPMG helped fund the fifth. The questions also drift between hiding use, hesitating to admit it and using a tool without permission, which are three different behaviors. I trust the shape, which is that a third to a half hold back, and not any single percent.

The sources also disagree on small things. Bloomberg’s report on the Deloitte survey, as far as I could read it (the page is cut off), says 63% of respondents worried a manager might think AI could do their job. Deloitte’s own release says 64% of weekly users. I use Deloitte’s. The Melbourne authors give 61% for avoiding revealing AI use in their Conversation article and 57% for the broader measure in the report, and on the same base a result for the broader measure cannot be smaller than one of its parts. The report does not say why in the text I read.

## What the silence costs

USC Marshall researchers Eric Anicich and Jeslyn Brouwers surveyed 604 US employees who use AI at work daily or several times a day. About four in five said sharing their AI techniques would benefit their team. Nearly one in three had withheld AI-related knowledge, workflows or techniques from coworkers or employers anyway. Anicich’s example is plain: if your best analyst is fast because of a method nobody has seen, you cannot train anyone else on it, and you do not know what you are paying for (USC Marshall, 2026-08-28).

There is a second cost, on the leader’s side of the desk. Deloitte’s shadow AI page says the gap between policy and practice makes it hard for IT teams to measure GenAI use, so internal tools and expensive licenses often go unused (Deloitte UK). You cannot judge a rollout you cannot see.

A third cost may fall on the person hiding. An Academy of Management Proceedings paper, “When Hiding Hurts”, analyzed multi-wave data from 207 employees and reports that concealing AI use reduced reflective learning and raised impostor feelings, which lowered task performance (abstract). I read only the abstract, and 207 people is a small base, so I hold that one loosely.

## Why people say they do it

Slack asked the people who said they would be uncomfortable telling their manager. Among them, 47% said using AI feels like cheating, 46% feared being seen as less competent and 46% feared being seen as lazy. The least common reason was that company policy discourages or forbids AI, at 21% (Slack, 2024).

PagerDuty reports these reasons for AI users. 33% would hide use to avoid scrutiny from managers or leadership, 30% because policies are restrictive or to avoid judgment from coworkers, and 29% because they are unsure whether their use is allowed. Another 29% said they would not hide it at all (PagerDuty, 2026). Microsoft found 53% of AI users worry that using it on important tasks makes them look replaceable.

The USC study names three worries: a damaged reputation, speed that brings more work or a new job as the company’s AI spokesperson, and proof that the tool can do your job. Deloitte’s respondents give the second worry some ground. They say they save 70 minutes a week on average, and most of the time saved goes to doing more work for the same employer.

**How hiding happens, in the order people describe it**

1. **A person finds a use that works**: Deloitte UK’s users say they save 70 minutes a week on average. Self-reported.
2. **They look for the rule**: Only 34% of employees in the Melbourne survey say their employer has a policy on generative AI. 29% of PagerDuty’s AI users are unsure whether their own use is allowed.
3. **They picture the reaction**: Cheating, less competent or lazy, for 46% to 47% of the uncomfortable in Slack’s survey. Replaceable, for 53% of AI users in Microsoft’s.
4. **They see no credit and a cost**: USC: the worry that speed brings more work, or the AI spokesperson job. Nothing in the surveys says what they would gain.
5. **They keep it quiet**: The method stays with one person, and the company never sees it.

Then loop back: The next person watches what happens to the first one who shares

My drawing of the chain the sources describe, step by step. The surveys ask about each step separately, so it is a model of the path and not a measured sequence. The loop comes from USC: if the worries come true right after someone shares, the researchers say that knowledge will not be shared again.

Ethan Mollick named the same three reasons in 2023, in an essay about “secret cyborgs”: banned tools pushed people to their phones, AI output is judged differently once people know it is AI, and the person who automates 90% of a job has a reason not to tell the boss. He summed them up as people not wanting to get in trouble (One Useful Thing, 2023). His one number is a Twitter poll he calls unscientific. I cite him for the reasoning, not the data.

There is one more group, and fear may not explain it. Deloitte counts 22% of GenAI users as people who believe their employer would approve if it knew, and reads them as too relaxed about the rules. I read them as people with no quick way to ask. That is my reading, not Deloitte’s. A company with a fast, public way to say yes would, I suspect, lose part of its “hidden” use quickly.

## The fear is not paranoia

It is tempting to treat the fear as a misunderstanding that a better memo will clear up. The evidence says the fear has a basis.

- **4,439**: Participants in four preregistered online experiments. People expected to be judged as lazier, less competent and less diligent for using AI, and observers (1,215 in the second experiment) did rate AI-helped employees that way. (source: Reif, Larrick and Soll, PNAS, 2025, https://pmc.ncbi.nlm.nih.gov/articles/PMC12088386/)
- **48%**: Of the workers in PagerDuty’s survey who had used AI they thought was not allowed, the share who then faced formal consequences such as a warning or disciplinary action. 53% received informal guidance. Wakefield Research for PagerDuty, 1,250 office professionals, April 2026. (source: PagerDuty, 2026, https://www.pagerduty.com/resources/ai/learn/survey-office-professionals-used-ai-tools-at-work-despite-not-being-allowed/)
- **64%**: Of weekly generative AI users in Deloitte UK’s survey who are worried managers will think AI can do their jobs. Ipsos UK fieldwork, May to June 2026. (source: Deloitte UK press release, 2026-09-16, https://www.deloitte.com/uk/en/about/press-room/british-workers-spend-one-billion-pounds-of-their-own-money-on-gen-ai-for-work.html)

The PNAS experiments also found two exits. In the hiring experiment with 1,718 managers, the penalty came mostly from managers who rarely use AI themselves. Those who used it more often were less likely to see an AI user as lazy. And in the last experiment the penalty disappeared for a task where AI was described as useful. The limits matter: the participants were online raters reading short descriptions and making hypothetical hires, not coworkers in a real office. I read the exits as a hint for managers. A manager who uses AI every week, and says what for, may do more than a policy page.

The first reply matters more than any memo. Brouwers and Anicich say employees who shared their AI use could tell within the first thirty seconds whether the supervisor was supportive. If the worries came true right after sharing, the researchers say, that knowledge will not be shared again. They also point out that a company memo claiming a sharing culture is a very different thing from showing one.

## Is it the rules or the fear? The sources disagree

Trestlewire’s headline on a study of 635 marketing workers says it ties hiding AI use to fear and “not bad policy”. I read the paper, Mızrak and Karakaya in Behavioral Sciences, and the headline goes further than the numbers do. The authors surveyed 635 marketing employees who actively use AI, twice and six weeks apart (November 2025 and January 2026), recruited through professional networks. Trestlewire places them in Istanbul; the paper text I read does not name where they work. Here is Table 9, the correlation of each condition with what they call “AI disclosure silence”.

**How closely each condition tracked hiding AI use among 635 marketing employees who use AI**

- Perceived creativity threat (raises hiding): +0.58
- Fear of negative evaluation (raises hiding): +0.56
- Perceived job insecurity (raises hiding): +0.51
- AI anxiety (raises hiding): +0.49
- AI policy clarity (lowers hiding): −0.46
- Trust in management (lowers hiding): −0.45
- Psychological safety (lowers hiding): −0.43
- AI and digital self-efficacy (lowers hiding): −0.39

Source: Mızrak and Karakaya, Behavioral Sciences, June 2026, Table 9. Bars show the size of each correlation; the sign is in the label. Correlation, not cause., https://doi.org/10.3390/bs16060994

Fear of judgment and the threat to creativity rank first. But AI policy clarity sits at −0.46, level with trust in management and ahead of psychological safety. It is not a footnote. The paper’s abstract lists unclear AI policy among the routes to hiding, and says no single condition was necessary. That is a finding I like: hiding showed up through many combinations of fear, distrust and unclear rules, so there is no one lever.

The limits are the authors’ own. The data are self-reported on a sensitive topic, so some people probably under-reported. The sample is not probability-based and is all marketing. The paper says the results are associations and predictions, not causal effects. A second study by the same lead author, of 564 consulting and professional-service staff in Türkiye, points the same way: fear of a competence penalty and disclosure anxiety were the strongest predictors, and official tools that people found adequate went with less concealment (Journal of Intelligence, abstract). Because the lead author is the same, I count it as one team’s two samples and not as independent confirmation.

Put Slack beside it and the sources split. In Slack’s survey, policy was the least cited reason for discomfort at 21%. In PagerDuty’s, 30% cited restrictive policy or coworker judgment in one answer, and 29% were unsure whether use was allowed. Different lists of reasons, different results. My reading is that rules and fear are not rivals. A rule says what is allowed. Fear is about what happens to you for being the one who says. A clear rule inside a punishing culture gets silence, and a warm culture with no rule gets guesses.

Deloitte’s data point to the same thing from another side. In financial services, where in-house tools and company-paid licenses are more common, 19% of GenAI users still pay for a tool themselves, against 17% across the UK. Handing out approved tools did not end self-paid use. Among the 3,562 respondents who pay for their own tools or use ones that may not be approved, 21% said those tools outperform the company’s.

## What each lever can reach

A company has four obvious things it can do about hidden AI use. It can ban, allow with clear rules, reward sharing, or try to watch. I sorted the causes above against those four. The grid is my judgment from the sources, not a count.

**Which causes each lever speaks to, in my reading of the sources** (none, partly or fully)

| | Ban | Allow, with clear rules | Reward sharing | Watch and measure |
| --- | --- | --- | --- | --- |
| Unclear rules | partly | fully | none | none |
| Fear of looking lazy or less competent | none | partly | fully | none |
| Fear of replacement or more work | none | none | partly | none |
| No credit for good uses | none | none | fully | none |
| The company cannot see what people do | none | partly | fully | partly |

My judgment from the sources in this post. Full means the evidence I read supports it plainly, half means partly or by implication, empty means I found nothing. No source tested any of these as a fix. “Reward sharing” follows the USC researchers’ list of credit, protected time and a share of gains.

Ban. In the Melbourne and KPMG survey, uploading company information to public AI tools was most common among employees whose employer had banned generative AI, at 67%. It was 56% where a policy guided use, 33% where there was no policy and 38% where people were unsure. The authors read this as a sign that bans may be ineffective. It is an association across self-reports, not proof that bans cause the behavior, and it measures uploads, not hiding. Mollick saw the same pattern in 2023 anecdotally.

Allow, with clear rules. It covers the cause that is easiest to fix, and the marketing study’s table says that cause counts. It leaves the social fear mostly where it was. A policy page cannot tell an employee how their manager will react.

Reward sharing. The USC researchers name possible rewards: credit in performance reviews for methods others adopt, protected time to keep experimenting, and a share of the gains once a workflow spreads. Mollick goes bigger, with cash prizes worth a year’s salary and promotions. I would start with the USC list. A large prize on a company with no safety just puts a spotlight on whoever was hiding.

Watch and measure. Deloitte says controlling use is another matter from seeing it, because it is almost impossible to track which tools people use and where. Watching without telling people also risks the trust you are trying to build. Our shadow AI guide covers how to look for AI use without eroding trust: tell people what you will look at, start with the least intrusive sources, and report by tool and team.

## What I would do first

I have not seen a study that tested a fix and then measured hiding. So this order is my judgment, built from what the sources say about causes, and it is cheap enough to try without proof.

**The first four weeks, in the order I would run them**

- **Week 1, Write the one-page rule.** List the AI tools that are approved, in review and not allowed. Say what data may go into each. Name who answers requests for a new tool, and by when; pick a number of days and meet it. Say what happens to people who used an unapproved tool before the page existed. My answer is nothing. See AI acceptable use policy.
- **Week 2, Ask every team what they use AI for.** Run it as an amnesty for past use. Ask for the task, the tool and what it saves, not for a confession. Report by tool and team, not by person. This is also where you meet the people who believe their use would be approved if you knew, 22% of Deloitte’s GenAI users.
- **Week 3, Brief managers on the first reply.** When someone shows a method, the first reply is a question about how it works and what they need. It is not more work, not a comment about their hours, and not a request to become the company’s AI spokesperson. The USC researchers say people judge the manager in the first thirty seconds.
- **Week 4, Credit the first three and share them.** Pick three methods people showed in week 2 and credit the people in their reviews. Give them protected time, tell the company who found what, and turn the methods into skills others can use. See what AI skills are.

> **Do not promise what you cannot keep** Mollick suggests guarantees that no one will be laid off because of AI use. Make that promise only if you will keep it. If you will not, say what you will do instead, especially about the time AI saves. Deloitte’s respondents say most of it goes into more work for the same employer. A worker who hears no answer will assume that one.

**What I would skip in month one**

- [ ] Punishing past use of unapproved tools, because it teaches the next person to stay quiet.
- [ ] Buying detection software before you have asked anyone, because the people you ask will tell you more than the logs will.
- [ ] A disclosure form with no rule behind it, because people cannot tell what a disclosure will be used for.
- [ ] A big prize before the one-page rule and the first-reply habit exist.

## What I could not find

I found no study that gave a company a fix, such as an amnesty, a reward scheme or a manager briefing, and then measured whether hiding fell. The marketing study’s authors themselves say experiments and field interventions are the next step. Everything above is a cause-and-correlation picture plus my own judgment about what to try.

I also cannot say how much hiding there is. The surveys disagree because they ask different things, nearly all rely on self-report, and four of the five come from companies with something to sell. Three of the four research papers are small or narrow: 635 marketing employees, 564 consultants in Türkiye and 207 employees in an abstract. The PNAS work has the most people but they are online raters judging hypothetical workers. The USC study covers only heavy daily users, and I read a news release about it, not the paper.

There are four things I did not use. I opened Cornerstone OnDemand’s survey, Ivanti and Laserfiche through search results and quoted none of them, because I did not read them in full. Bloomberg’s Deloitte story was cut off after its first lines. TechRadar’s 2026-09-16 piece on whether to tell your boss I did not read.

> **What I read, and its limits.** On 2026-10-16 I read the pages in Sources as raw text. Deloitte UK, Microsoft and LinkedIn, Slack, PagerDuty and the Melbourne and KPMG report are surveys or reports from organizations with a stake in the topic; I label each where it appears. The Behavioral Sciences paper I read in full through Europe PMC, including its correlation table. It does not state where the respondents work, so the Istanbul detail comes from Trestlewire. The Journal of Intelligence paper and the “When Hiding Hurts” paper I read only as abstracts. TechRadar’s report of the PagerDuty survey agreed with the PagerDuty page, and I used the PagerDuty page. Percentages are quoted exactly as each source gives them, with the base the source states. Where a source did not state the base, I say so.

## Where Nolli fits, and where it does not

I run Nolli, so weigh this accordingly. Nolli does not fix the trust problem, and I would not claim it does. It learns how work gets done from the work itself: the apps your admin allows on company devices, the email and calendar your company connects, and the logs of AI agents such as Claude Code and Codex. So a company can see where AI is already part of the work without waiting for each person to confess it, and share what works as skills. Your company sets what Nolli learns from, and its findings are about the work, not a person. That cuts both ways. A tool that learns from the work raises the trust question this post is about, so tell people what it looks at and why before you switch it on. Nolli does not write your AI policy, brief your managers or run your credit scheme. There is no public price. Request a demo. See Nolli for AI enablement teams.

## Questions people ask

**Why do employees hide their AI use from their managers?**

In the surveys I read, the stated reasons are fear that AI use looks like cheating or looks lazy or less competent, fear of looking replaceable or being handed more work, and unclear rules. In Slack’s survey of desk workers, 46% of those uncomfortable feared looking less competent and 21% cited a policy against AI.

**How many employees hide their AI use?**

It depends on the question. Five surveys put it between 31% and 57%: Deloitte UK 31% use it without the employer’s knowledge, Slack 48% uncomfortable telling their manager, Microsoft 52% reluctant to admit it for important tasks, and Melbourne and KPMG 57% admit non-transparent use. Treat the range as a pattern and not as a rate.

**Is the fear of being judged for using AI justified?**

Partly. Four preregistered experiments with 4,439 online participants found that observers rated people helped by AI as lazier and less competent. The penalty came mostly from raters who rarely use AI themselves, and it disappeared for a task where AI was described as useful.

**Does banning AI stop employees from hiding it?**

The evidence I read says no. In the Melbourne and KPMG survey, uploading company information to public AI tools was most common where the employer banned generative AI, at 67%, against 33% where there was no policy. That is an association from self-reports, not proof that bans cause it.

**What should a leader do first about hidden AI use?**

Write a one-page rule on which tools and data are allowed, with a fast answer to requests. Then brief managers to answer the first person who shares with a question and not with more work, and give credit for methods others adopt. No study I found tested this order, so treat it as a judgment.

**Is hiding AI use the same as shadow AI?**

They overlap. Shadow AI means using AI tools without the company’s approval. Hiding AI use also covers approved tools whose use people do not mention, such as passing AI-assisted work off as their own. See our shadow AI guide.

## Sources

- "British workers spend nearly £1bn of their own money on GenAI for work, landmark Deloitte research finds." Deloitte UK, September 16, 2026. https://www.deloitte.com/uk/en/about/press-room/british-workers-spend-one-billion-pounds-of-their-own-money-on-gen-ai-for-work.html
- "What we do in the AI shadows." Deloitte UK, accessed October 16, 2026. https://www.deloitte.com/uk/en/issues/generative-ai/genai-workforce-survey-shadow-ai.html
- Jacob Reid. "UK Workers Hide AI Use From Their Bosses, Deloitte Survey Finds (first lines only)." Bloomberg Law, September 16, 2026. https://news.bloomberglaw.com/artificial-intelligence/uk-workers-hide-ai-use-from-their-bosses-deloitte-survey-finds
- Alexander Bernard. "Why Are Employees Reluctant To Disclose AI Use to Their Bosses?." USC Marshall School of Business, August 28, 2026. https://www.marshall.usc.edu/news/why-are-employees-reluctant-to-disclose-ai-use-to-their-bosses
- Filiz Mızrak and Turhan Karakaya. "Why Hide AI Use? Psychological Configurations and Explainable Machine Learning Evidence from Marketing Work." Behavioral Sciences (MDPI), full text via Europe PMC, June 15, 2026. https://doi.org/10.3390/bs16060994
- Sana Whitfield. "A Study of 635 Marketing Workers Ties Hiding AI Use to Fear, Not Bad Policy." Trestlewire, October 4, 2026. https://www.trestlewire.com/article/hiding-ai-use-at-work-fear-of-judgment-study
- Filiz Mızrak, Gözde Mert, Ahmet Erkasap, Ali Özcan and Mehmet Mert. "Hidden Generative AI Use in Consulting: Social-Cognitive and Organizational Factors in AI Disclosure and Concealment (abstract)." Journal of Intelligence (MDPI), September 1, 2026. https://doi.org/10.3390/jintelligence14090196
- Jessica A. Reif, Richard P. Larrick and Jack B. Soll. "Evidence of a social evaluation penalty for using AI." Proceedings of the National Academy of Sciences, via PubMed Central, 2025. https://pmc.ncbi.nlm.nih.gov/articles/PMC12088386/
- Peikai Li, Lin Ma, Niannian Dong, Lu Xing and Shuai Yuan. "When Hiding Hurts: How Stealth Use of Generative AI Resources Impairs Task Performance (abstract)." Academy of Management Proceedings, July 2025. https://doi.org/10.5465/amproc.2025.20611abstract
- Nicole Gillespie, Steve Lockey, Tabi Ward, Alexandria Macdade and Gerard Hassed. "Trust, attitudes and use of artificial intelligence: A global study 2025." The University of Melbourne and KPMG, April 2025. https://assets.kpmg.com/content/dam/kpmgsites/xx/pdf/2025/05/trust-attitudes-and-use-of-ai-global-report.pdf
- Nicole Gillespie and Steven Lockey. "Major survey finds most people use AI regularly at work, but almost half admit to doing so inappropriately." The Conversation, April 28, 2025. https://theconversation.com/major-survey-finds-most-people-use-ai-regularly-at-work-but-almost-half-admit-to-doing-so-inappropriately-255405
- "AI at Work Is Here. Now Comes the Hard Part (2024 Work Trend Index Annual Report)." Microsoft and LinkedIn, May 8, 2024. https://www.microsoft.com/en-us/worklab/work-trend-index/ai-at-work-is-here-now-comes-the-hard-part
- "The Fall 2024 Workforce Index Shows Executives and Employees Investing in AI, but Uncertainty Holding Back Adoption." Slack, November 12, 2024. https://slack.com/blog/news/the-fall-2024-workforce-index-shows-executives-and-employees-investing-in-ai-but-uncertainty-holding-back-adoption
- "PagerDuty Survey: 66% of Office Professionals Have Used AI Tools or Services at Work Despite Not Being Allowed Under Company Policies." PagerDuty, survey by Wakefield Research, June 2026. https://www.pagerduty.com/resources/ai/learn/survey-office-professionals-used-ai-tools-at-work-despite-not-being-allowed/
- Efosa Udinmwen. "Workers secretly use banned AI tools at work while believing company policies clearly forbid their growing everyday usage." TechRadar, June 18, 2026. https://www.techradar.com/pro/intentionally-hide-using-ai-66-of-office-workers-admit-to-secretly-using-banned-ai-tools
- Ethan Mollick. "Detecting the Secret Cyborgs." One Useful Thing, June 18, 2023. https://www.oneusefulthing.org/p/detecting-the-secret-cyborgs