Should AI make workplaces clearer—or workers more visible?
Eleven US caregivers and employment-support professionals described hidden workplace expectations as an access barrier for autistic and IDD workers. Their technology ideas also exposed a surveillance risk: support data could be repurposed for monitoring or discipline.
The short version
3 key points
- 1.The study interviewed 11 US caregivers and employment-support professionals for 60 to 90 minutes; it did not interview workers directly.
- 2.Researchers developed five themes of ambiguity and grouped them into evaluative, procedural and relational dimensions.
- 3.Participants imagined real-time cues, feedback-processing tools and physiological guidance, but none was built or tested and each could increase employer surveillance.
The answer: clarify the workplace before instrumenting the worker
AI could help translate vague feedback or point a worker toward an authorised next step, but this study's strongest finding is organisational rather than technical. Autistic and intellectually and developmentally disabled workers can be judged against expectations that become explicit only after they have acted on a guess. Making those expectations clear before they carry consequences is an employer responsibility, not a defect-correction task for the worker.
The technology concepts proposed in the interviews included discreet real-time redirection, assistance interpreting feedback and prompts linked to physiological signals. None was built, deployed or evaluated. The same systems that make a workplace more legible to a worker could make the worker more legible to an employer—recording behaviour, stress or communication and inviting productivity scoring, discipline or pressure to conform.
The evidence comes from an unreviewed US preprint submitted on 15 October. Researchers at several US universities interviewed 11 caregivers and employment-support professionals. They explicitly state that these proxy accounts do not substitute for workers' own accounts. The paper's HTML version did not identify project funding or a competing-interest statement, limiting assessment of those interests.[1]
Who took part and how the study was conducted
Participants included parents, one grandparent caregiver, a job coach and other support professionals who currently or recently supported an autistic or IDD person in paid work, volunteering or work-like training. Ten reported between 13 and 30 years of experience. Recruitment used regional disability-service organisations, supported-employment providers and autism community groups. The sample was purposive and small; it was designed to explore recurring experiences, not estimate how common they are.
Each Zoom interview lasted 60 to 90 minutes. Participants described work histories, support needs and concrete episodes involving unclear tasks, social cues, evaluation or routes to help. At the end, they used a general-purpose model such as ChatGPT or Gemini to explore a possible support for one or two situations they had just described. The AI was a co-design probe, not the object of a performance test.
The university's institutional review board approved the study. Participants consented, could skip questions and received a $40 digital gift card. Zoom generated initial transcripts, which researchers corrected and pseudonymised. Two coders used reflexive thematic analysis. The process produced 153 initial codes, 13 subthemes and five primary ambiguity themes, plus 54 codes, ten subthemes and three technology-support themes. Reflexive thematic analysis interprets patterns; it does not seek inter-rater reliability as if themes were objective labels.[1]
What workplace ambiguity looked like
The five themes covered a hidden curriculum of social norms, difficulty evaluating performance expectations, misjudged social boundaries, opaque formal structures and institutional barriers to support. The authors regrouped them into three dimensions. Evaluative ambiguity concerns whether work is acceptable and whether feedback is coaching or a formal warning. Procedural ambiguity concerns what action is allowed, who has authority and how an accommodation can be requested. Relational ambiguity concerns how coworkers and supervisors interpret behaviour that formal job descriptions do not mention.
The common mechanism was timing. Information became clear after the worker had already been judged, upset a colleague or missed a route to assistance. In the participants' accounts, caregivers and job coaches then performed unpaid interpretive labour: reconstructing what a supervisor meant, identifying a policy owner or translating an indirect criticism into a concrete action.
This framing matters because it rejects a one-sided deficit model. A worker should not need an emotion detector to compensate for a supervisor who withholds direct feedback. Nor should a chatbot's polite paraphrase be mistaken for authoritative clarification. Only the organisation can confirm whether a message is a warning, what standard applies and which person can approve an adjustment.[1]
The technology ideas—and the surveillance boundary
Participants imagined ambient cues that might signal when a conversation had gone on too long, tools that turn unclear feedback into reviewable steps, and systems using physiological signals to prompt self-regulation. Grounding the exercise in a recent workplace episode kept the ideas concrete. It still did not show that the ideas are wanted by workers, technically accurate or safe in use.
Real-time behavioural redirection could become automated masking: a system coaching a person to appear neurotypical rather than changing an inaccessible environment. Feedback processing could confidently misread sarcasm, authority or urgency. Physiological guidance could infer stress inaccurately while creating a sensitive stream of health-like data. If an employer receives that stream, a support tool can become performance monitoring by another name.
The authors propose that the worker control when assistance is active, what it collects and who can access it. Caregivers or job coaches may help configure support when invited, but should not automatically receive data or override the worker. Employers should prohibit using support data for productivity scoring, discipline or emotion classification. Any AI interpretation should remain separate from the original workplace communication and display uncertainty.[1]
What employers and public services can do now
Managers do not need to wait for an AI product to improve access. They can state the applicable performance standard, the requested change and the consequence of the message. They can identify who owns an accommodation process, provide a private route for clarification and make changed expectations explicit. These are testable process changes with fewer privacy risks than continuous sensing.
Procurement teams should require worker-led design, local disability and employment-law review, data minimisation, deletion controls and a ban on secondary use. A pilot should measure whether workers understand expectations sooner, whether disputes and distress fall, and whether workers can refuse the tool without disadvantage. It should also check false prompts, unequal performance across communication styles and whether supervisors become less direct because they assume the tool will translate them.
The evidence would change most with direct, accessible research involving autistic and IDD workers across different support needs and job settings. A co-designed prototype could then be compared with a non-AI organisational intervention such as clearer supervision and written feedback. Outcomes should include worker autonomy, comprehension, stress, job retention and complaints—not just manager satisfaction or task speed.[1]
What this means for people
- Autistic and IDD workers could gain timely clarification, but should not have to accept surveillance or automated behavioural conformity to receive support.
- Managers can reduce ambiguity through direct feedback, named decision-makers and accessible procedures before buying technology.
- Caregivers and job coaches may support configuration when invited, but the worker should control data, visibility and withdrawal.
Global context
The interviews were conducted through US organisations and reflect US employment and support systems. Hidden expectations and surveillance risks travel across borders, but accommodation law, worker representation, benefits and supported-employment structures differ. Direct research with workers in each context is necessary.
What the evidence does not yet show
- Only 11 caregivers and support professionals were interviewed; the study does not estimate prevalence.
- Workers themselves were not participants, and proxy views may diverge most on relationships and self-determination.
- The technology concepts were generated in one interview session and were not built or evaluated.
- Supervisors, coworkers and human-resources staff were not interviewed.
- The manuscript is an unreviewed preprint and the public HTML did not identify project funding or competing interests.
What to watch next
- Worker-led, accessible interviews across diverse disability and communication profiles
- A comparison between AI support and clearer non-AI management practices
- Evidence on autonomy, stress, retention and complaints rather than productivity alone
- Enforceable limits on employer access and secondary use of support data
Living evidence record
Impact record IAI-1DUK33P
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
16 October 2026
Source trail
1 direct source across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Single-source reporting disclosure
This record analyses one direct source. It can establish what arXiv published or reported, but it is not independent corroboration of every performance claim or predicted outcome. The confidence label will change only when broader evidence is added.
Evidence trail
Sources used for this report
Links checked 16 October 2026
Research question, topics and regions
How caregivers and employment-support professionals understand unstated workplace expectations for autistic and IDD workers, and what technology supports they imagine
Topics: Disability inclusion · Workplace AI · Autism · Intellectual and developmental disabilities · Worker surveillance
Regions: United States · North America · Global
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Watch the background, then check the evidence
UNESCO
AI-generated illustration · not a video frame
YouTube loads only when you choose. Press play in the player to start. If playback is unavailable, use “Watch on YouTube” below.
AI, jobs & human choice · Background explainer
Published 2024-10-14 · 11 min 27 sec
UNESCO Talks: Interview with 2024 Nobel laureate Professor Daron Acemoglu
Economist Daron Acemoglu discusses how choices about AI affect workers, inequality and society. Pair the interview with our comparison of task exposure, hiring and skills evidence.
Evidence check: An expert interview from 2024, offering a perspective on policy and work rather than a measured forecast of job losses.
Source checked 14 October 2026. Check available captions on the publisher’s video page.
The Impact Brief · Free
Keep following the stories that matter to you.
Get our weekday briefing on what changed and why it matters. Choose your topics below.
Continue the story
There’s more to this story
Work & Skills
Who is gaining from AI at work?
A Gallup survey of 15,482 US employees finds that regular AI users report more speed, creativity and work quality, but frequent use is concentrated among graduates, managers and workers who already have better jobs. The results are self-reported associations, not a productivity trial.
8 min · 2 sources
Work & Skills
Does an AI label bias code review—or does job rank matter more?
In a randomized vignette experiment, 447 Microsoft engineers did not penalize an AI-use disclosure on identical code, but rated the same code and author more favourably when the fictional author was labelled Principal rather than Level 1. One AI-normalized US workplace cannot settle how other teams respond.
9 min · 2 sources
Work & Skills
Can AI agents configure complex contract workflows?
OpenAI reports that GPT-6 Astra scored 55.0% across 11 Ironclad tasks versus 41.6% for GPT-5.6 Sol, while simulated time fell from 37.0 to 19.2 minutes. The company-run benchmark shows progress—not autonomous legal work or measured customer productivity.
7 min · 1 source
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.