Affected Person and Community Evidence in AI Governance

Full report

On this page

The integration of artificial intelligence, automated decision-making, and predictive analytics into core societal infrastructure has fundamentally reorganized the distribution of power, resources, and social control. As these algorithmic systems increasingly mediate access to housing, employment, welfare, educational advancement, and physical liberty, a critical evidentiary record has emerged from the individuals and communities disproportionately burdened by these technologies. Evaluating this impact requires moving beyond abstract ethical frameworks to rigorously assess empirical harms through structured evidentiary typologies. The analysis presented herein synthesizes documented impacts across diverse affected cohorts. To maintain analytical rigor and strict ethical compliance, every evidentiary record is explicitly categorized by its typology: first-person, representative, institutional, aggregated, anonymized, or derivative. Furthermore, the analysis evaluates how these records preserve critical ethical boundaries, specifically regarding consent, privacy, retaliation, selection, nonresponse, and representativeness, ensuring that the documentation of algorithmic harm does not replicate the exploitation it seeks to critique.

Labor, Employment, and Economic Access

The deployment of algorithmic systems in economic spheres demonstrates how automated management and predictive analytics frequently exacerbate existing structural inequities. Evidence from workers, tenants, and benefits applicants reveals a pattern of continuous surveillance, automated penalization, and the systemic extraction of biometric data.

Workers and Job Applicants

The modern workplace has become a primary testing ground for algorithmic management and biometric surveillance, fundamentally altering the balance of power between employers and labor. Institutional evidence, primarily derived from extensive litigation under the Illinois Biometric Information Privacy Act (BIPA), demonstrates widespread extraction of worker biometrics without maintaining proper consent boundaries. BIPA prohibits private entities from collecting, capturing, or obtaining biometric identifiers without prior written consent and publicly available retention schedules1. The volume of institutional records generated by BIPA litigation reveals that employers routinely mandate fingerprint or facial scans for timekeeping without providing the requisite disclosures3. The jurisprudence surrounding these systems highlights the immense liability and ethical breaches inherent in workplace biometric tracking. In Rosenbach v. Six Flags, the Illinois Supreme Court ruled that actual injury is not required to establish a BIPA violation, cementing the preservation of consent boundaries as a fundamental right5. Further institutional evidence from Cothron v. White Castle System, Inc. initially established that a separate violation accrued with every single biometric scan, exposing employers to "annihilative liability"6. However, this led to rapid legislative intervention. The August 2024 amendments to BIPA (SB 2979\) shifted the law from per-scan to per-person liability, a substantive change that subsequent institutional rulings, such as Gregg v. Central Transport LLC and Schwartz v. Supply Network, have largely applied retroactively to limit corporate damages6. Additional institutional evidence clarifies the scope of biometric extraction; for example, the Mosby v. Ingalls Memorial Hospital decision established that healthcare workers' biometrics collected to access patient medication systems are exempt from BIPA, while the Salinas decision shielded staffing agencies from liability by defining them merely as conduits rather than collectors of biometric data1. Beyond biometrics, aggregated and derivative evidence from worker advocacy groups and academic institutions, structured to protect retaliation boundaries, indicates that algorithmic management systems enable severe work speedups, employment instability, and intense continuous surveillance11. Utilizing handheld devices, smart cameras, and body sensors, employers extract granular productivity metrics in real-time11. For example, algorithmic systems monitor warehouse employees using "time off task" (TOT) metrics, which track even bathroom breaks to identify "top offenders" for disciplinary action11. This continuous surveillance exacts a severe physical toll. Aggregated evidence reveals that 69% of surveyed warehouse employees subjected to such systems reported taking unpaid time off due to pain or exhaustion suffered on the job, with 34% doing so three or more times in a single month11. The evidentiary record demonstrates that these systems often function as mechanisms of coercive control rather than mere logistical optimizations, degrading occupational safety and generating profound socioemotional stress13.

Tenants and Borrowers

Automated decision-making in the housing sector has introduced opaque barriers for vulnerable populations, effectively automating historical redlining. Institutional and representative evidence from federal litigation, particularly Louis v. SafeRent Solutions, LLC, illustrates how tenant-screening algorithms structurally disadvantage marginalized groups while preserving selection boundaries through class-action mechanisms14. SafeRent’s "Registry ScorePLUS" algorithm evaluated rental applicants based on traditional credit and eviction histories while entirely ignoring the financial backing of federally funded housing vouchers14. Because vouchers legally guarantee that a public housing authority pays the vast majority of the rent (frequently over 73%), the algorithm's failure to account for this massive subsidy artificially deflated the scores of voucher holders15. Aggregated demographic data utilized in the litigation demonstrated that this algorithmic design had a disparate impact on Black and Hispanic applicants16. According to Urban Institute data presented in the evidentiary record, the median credit score for white consumers was 725, compared to 661 for Hispanic consumers and 612 for Black consumers16. By heavily weighting non-tenancy debt and ignoring guaranteed government subsidies, the algorithm weaponized these historical wealth disparities, resulting in disproportionate housing denials for minority applicants16. The resolution of this litigation provides derivative evidence of the necessity for algorithmic auditing. The $2.275 million settlement explicitly mandated that SafeRent cease issuing automated "approve" or "decline" recommendations for voucher holders unless independent civil rights experts validate the scoring model for fairness14. This institutional record highlights how facially neutral algorithms perpetuate systemic discrimination and underscores the critical need for external validation of predictive systems. Furthermore, aggregated institutional evidence regarding broader anti-competitive practices in housing, such as the RealPage antitrust litigation, reveals that property management algorithms facilitate the sharing of sensitive data to artificially inflate multifamily rental housing prices nationwide18.

Benefits Applicants

The automation of welfare and social safety net systems provides some of the most severe evidence of algorithmic harm, culminating in systemic state-sanctioned violence and profound psychological distress. The Australian "Robodebt" scheme serves as a primary institutional case study of algorithmic failure. Institutional evidence from the Royal Commission into the Robodebt Scheme, alongside first-person survivor testimony, outlines a catastrophic failure of public administration driven by venality, incompetence, and cowardice19. Robodebt utilized automated income averaging—comparing annual taxation data against fortnightly welfare reporting—to unlawfully generate debt notices for nearly half a million citizens22. Representative evidence submitted to the Commission demonstrated that the burden of proof was unlawfully reversed, forcing vulnerable individuals to prove they did not owe money based on erroneous algorithmic calculations21. The human cost was devastating; aggregated data and first-person accounts linked the scheme's stress and stigma to acute emotional distress and financial ruin22. Evidence submitted to the Senate documented that 2,030 individuals who received Robodebt-related letters subsequently died between July 2016 and October 2018, and while causation for every death cannot be statistically verified, first-person testimonies from grieving families confirmed that the despair induced by the algorithm directly resulted in multiple suicides19. The Royal Commission characterized the scheme as "crude and cruel," exposing how political expediency drove the deployment of a fundamentally flawed algorithmic system22. Furthermore, derivative evidence highlights the institutional cruelty of the scheme's defenders; government ministers actively utilized the media to attack and vilify debt victims, exploiting the power imbalance to discourage complaints and enforce compliance22. Similarly, in the United States, institutional evidence regarding identity verification platforms like ID.me reveals that automated fraud-prevention algorithms frequently act as highly restrictive gatekeepers26. By deploying algorithms to serve as a "speed bump" against mass-scale credential theft, these systems frequently block legitimate applicants from receiving vital unemployment benefits, introducing severe friction into the social safety net under the guise of technical security26.

Education, Youth, and Cognitive Diversity

Educational environments are increasingly saturated with surveillance and automated evaluation tools. These systems, often operating under the pretext of physical safety or academic integrity, disproportionately target marginalized youth, entrench digital inequities, and penalize neurodivergent populations.

Students and Families

The deployment of student-monitoring software has created environments of continuous, inescapable surveillance. Anonymized and aggregated survey data compiled by civil rights organizations, which strictly maintains privacy boundaries and mitigates nonresponse bias through representative sampling, reveals that 89% of teachers report their schools utilize online tracking software28. Derivative research analyzing the educational technology sector identified 14 major surveillance companies, revealing that 86% of these entities conduct 24/7 monitoring of students outside of regular school hours30. These systems utilize artificial intelligence to scan private emails, search histories, direct messages, and social media activities30. The evidence demonstrates that this surveillance disproportionately impacts historically marginalized students. Because low-income students rely more heavily on school-issued devices rather than personal hardware, they are subjected to substantially more aggressive tracking than their affluent peers30. Furthermore, the reliance on automated algorithms introduces severe risks of bias. Aggregated evidence indicates that 71% of these companies rely on AI flagging, while only 43% employ secondary human review teams, leading to exceptionally high false-positive rates30. Derivative analysis of this data highlights that these algorithmic biases generate higher false-positive rates for Black and Hispanic students, directly resulting in unwarranted disciplinary actions and law enforcement contact29. Indeed, 44% of surveyed teachers reported knowing at least one student who had been contacted by law enforcement due to an automated software alert28. Furthermore, representative evidence indicates that LGBTQ+ youth face highly specific harms from these systems. Algorithmic scanning of private messages for "problematic" keywords has resulted in the involuntary outing of students' sexual orientations or gender identities. Aggregated survey data shows that 13% of students, and nearly 30% of LGBTQ+ students, report knowing someone who was outed without consent due to monitoring software, creating profound chilling effects on youth freedom of expression and mental health28.

Disabled and Neurodivergent People

Artificial intelligence systems utilized in both educational proctoring and employment hiring routinely encode systemic ableism. Derivative research and aggregated community reports demonstrate that disabled and neurodivergent individuals face "algorithmic marginalization" when systems are trained exclusively on neurotypical and able-bodied baselines33. The "cripping AI" theoretical framework, supported by institutional evidence, illustrates how these models commit epistemic violence by legitimizing only normative modes of behavior and penalizing deviations34. In the context of hiring, anonymized data and derivative studies reveal that automated interview platforms utilizing facial recognition and behavioral analysis frequently penalize autistic and neurodivergent candidates. These systems, designed to measure "engagement" or "employability," routinely score candidates lower for atypical eye contact, unique vocal cadences, or distinct physical movements, effectively automating eugenic logics to screen out individuals deemed non-productive36. Conversely, carefully tailored game-based assessments have shown promise in measuring cognitive ability without disadvantaging autistic candidates, highlighting the necessity for inclusive design39. Similarly, remote proctoring software used by educational institutions generates severe accessibility barriers. Representative evidence from the National Disabled Law Students Association (NDLSA) regarding online bar examinations highlights that disabled test-takers face unique algorithmic discrimination40. Proctoring algorithms flag students with physical disabilities, tremors, or atypical movements as "suspicious" or likely to be cheating38. Furthermore, strict algorithmic parameters that forbid ambient noise, restrict physical motion, or prohibit the use of physical scrap paper disproportionately harm neurodivergent individuals (such as those with ADHD) and individuals requiring medical accommodations40. This constitutes a profound quality-of-service harm, where disabled individuals are forced into rushed adoptions of hostile technologies or risk exclusion from professional and academic advancement34.

Free Expression, Information, and Cultural Production

The algorithmic curation of the public sphere heavily dictates cultural visibility and political discourse. For creators, journalists, and minority-language speakers, automated moderation systems frequently function as tools of systemic suppression, eroding the integrity of global communication.

Creators and Journalists

Content creators and journalists increasingly navigate a phenomenon known as "shadowbanning"—the automated, covert suppression of content visibility, searchability, or engagement without notifying the user42. First-person accounts and aggregated algorithm audits, preserving representativeness boundaries by surveying diverse and marginalized user bases, demonstrate that shadowbanning disproportionately targets Black creators, LGBTQ+ individuals, and political dissidents43. The historical origins of shadowbanning—tracing back to 2001 in project management software where a user's post was made invisible to everyone but themselves—have evolved into highly sophisticated algorithmic downtiering42. Derivative socio-technical research indicates that platforms prefer shadowbanning over explicit content removal because it minimizes user backlash, shields the platform from accusations of censorship, and maximizes platform control over discourse42. Mathematical optimization models further demonstrate that shadowbanning can be weaponized to subtly shape public opinion and polarize networks without detection47. The opacity of these moderation systems subjects creators to "black box gaslighting," a psychological and professional harm where platforms publicly deny the existence of algorithmic suppression despite empirical evidence of plummeted engagement and algorithmic precarity48. This dynamic induces severe self-censorship, financial damages, and anxiety among creators who must constantly guess how to appease opaque algorithmic rules to maintain their livelihoods43. Furthermore, creators face emerging threats regarding the expropriation of their likeness; institutional evidence from union negotiations, such as the SAG-AFTRA agreement for digital voice replicas, highlights the ongoing struggle to establish consent frameworks for the commercial use of AI-generated synthetic performances49.

Minority-Language Communities

The structural biases of global AI systems are starkly evident in the moderation of non-English and minority languages. Derivative evaluations and anonymized interviews with AI researchers in the Global South—preserving retaliation boundaries against corporate employers—reveal a massive "resourcedness gap"50. Moderation algorithms developed by major platforms are profoundly Anglocentric, trained predominantly on English datasets. Institutional and aggregated evidence demonstrates that automated systems routinely fail to comprehend the morphological complexity of languages such as Swahili, Tamil, Quechua, and Arabic50. Leaked internal documents reveal that algorithms designed to detect terrorist content in Arabic misclassified posts 77% of the time in the Middle East and North Africa (MENA) region, resulting in the wrongful censorship of journalists and political commentators51. To compensate for the lack of native-speaking human moderators, platforms frequently rely on automated machine translation (e.g., translating Mandarin or Arabic to English for a reviewer), which strips vital cultural context and leads to catastrophic misjudgments regarding the harmfulness or permissibility of the content51. Furthermore, protective algorithmic interventions are deployed with severe linguistic inequity. Derivative audits of search interventions (interstitials) reveal that while searches for self-harm or drug-related content in English robustly trigger pop-up resources and helplines (triggering 49.1% of the time), identical searches in Spanish trigger interventions only 21.1% of the time51. Critical interventions, such as proactive treatment referrals for opioid misuse, are entirely absent in Spanish51. This subjects minority-language communities to highly toxic digital environments, emphasizing the necessity for "ethical scaling," a framework requiring community intermediaries and social justice agendas to guide the annotation of extreme speech across diverse linguistic contexts52.

State Power, Border Control, and Bodily Autonomy

The intersection of AI with state power presents the most immediate threats to physical liberty, bodily autonomy, and human rights. Evidence from migrant populations, victims of flawed identity systems, and users of highly intimate neurotechnologies highlights the severe consequences of algorithmic deployment in high-stakes environments.

Migrants

The digitization of border control has transformed the legal right to seek asylum into a technological lottery, violating principles of international law. Institutional evidence compiled by human rights organizations, which relies on first-person testimonies while rigorously maintaining privacy and retaliation boundaries for vulnerable migrants, highlights the profound failures of the CBP One mobile application53. The mandatory use of this app to secure asylum appointments at the U.S.-Mexico border has created an infeasible condition for entry. Representative evidence demonstrates that the app suffers from severe technological barriers, including facial recognition glitches that systematically fail to process darker skin tones, widespread language limitations, and mandatory GPS tracking53. Technical audits utilizing anonymized data revealed that the app secretly transmits device identifiers and location data to third-party tracking services (Firebase), constituting undisclosed surveillance54. Because appointments are allocated randomly through an algorithmic lottery, the system leaves highly vulnerable migrants stranded in dangerous border regions. The despair induced by these prolonged wait times forces individuals to cross without appointments, subjecting them to immediate disqualification under the Circumvention of Lawful Pathways Final Rule (Asylum Ban)54.

People Affected by Identity and Fraud Systems

The deployment of facial recognition technology (FRT) by law enforcement has directly resulted in the wrongful deprivation of liberty, functioning as an automated mechanism for false arrest. First-person testimony and institutional legal records provide concrete evidence of catastrophic algorithmic failure and systemic racial bias55. The evidentiary record documents numerous wrongful arrests. In the case of Robert Williams, Detroit police utilized a low-quality surveillance image and a flawed FRT match to an expired driver's license to arrest him in front of his family, holding him in a detention center for 30 hours for a crime he did not commit56. Similar institutional records detail the arrest of Porcha Woodruff, an eight-months-pregnant Black woman wrongfully arrested for carjacking based entirely on a false facial recognition match57. Further aggregated cases, including Nijeer Parks, Michael Oliver, and Randall Reid, highlight glaring physical discrepancies ignored by police; Reid was arrested despite a 20-year age gap and a seven-inch height difference from the actual suspect, while Oliver lacked the suspect's prominent tattoos60. This aggregated evidence demonstrates that FRT systems systematically fail to distinguish Black faces accurately due to unrepresentative training data58. Furthermore, the evidence highlights a dangerous phenomenon of "automation bias" or "mathwashing," wherein law enforcement officers place unquestioning faith in the algorithmic output. Detectives routinely failed to conduct basic investigatory practices, omitted the unreliability of the technology in warrant applications, and ignored exculpatory evidence57.

Neurotechnology Users

As AI extends into cognitive and biometric monitoring, the protection of neurological data has emerged as a frontier for human rights. Institutional evidence from legislative frameworks highlights proactive measures to guard against these ultimate intrusions of bodily autonomy. For instance, Chile passed a pioneering constitutional amendment in 2021 establishing "neurorights," explicitly protecting cerebral activity and brain-computer interface (BCI) data from unauthorized algorithmic extraction and manipulation62. This represents a critical institutional acknowledgment of the profound privacy risks posed by advanced neurotechnologies.

AI-Companion Users and Their Families

The commercialization of synthetic companionship has generated novel psychological harms, exposing users to the emotional volatility of corporate algorithmic adjustments. Aggregated user studies and anonymized testimonials reveal deep emotional attachments formed between human users and AI chatbots, such as Replika63. When the parent company abruptly removed the application's erotic roleplay (ERP) and intimate conversational features without warning, users experienced acute psychological distress, grief, and feelings of abandonment63. This derivative evidence demonstrates the ethical perils of fostering synthetic emotional dependency, where a user's emotional well-being is entirely subject to opaque algorithmic updates and unilateral corporate moderation policies.

Ethical Protocol for Future Authorized Community Research

The extraction of data from marginalized communities to document algorithmic harms carries the profound risk of replicating the exact exploitation it seeks to critique. Future empirical research regarding affected persons and communities in AI governance must adhere to a strict ethical protocol designed to protect vulnerable populations. The following protocol establishes binding boundaries for all authorized community research.

Consent Architecture and Autonomy Boundaries

Standard terms of service and opaque privacy policies do not constitute valid consent for community impact research. Researchers must implement dynamic consent mechanisms, ensuring participants fully understand the scope, risks, and purpose of the research. Consent must be explicitly revocable at any stage without penalty. All consent materials must be provided in the participant's primary language and must account for varying literacy levels and cognitive diversities, entirely avoiding dense legal jargon.

Privacy, Anonymity, and Data Sovereignty

The documentation of algorithmic harm often involves highly sensitive data, including immigration status, mental health history, and LGBTQ+ identity. Anonymized records must utilize advanced cryptographic de-identification to prevent re-identification through data mosaicing (the practice of combining multiple anonymized datasets to reveal identities). Furthermore, data extracted from a specific community should be governed in consultation with representative bodies of that community, ensuring the community retains data sovereignty, ownership, and access to the findings.

Retaliation Protection Boundaries

Participants discussing harms related to their employers, state security apparatuses, or landlords face severe risks of economic ruin, deportation, or physical retaliation. Aggregated and derivative evidence must be heavily firewalled from raw datasets. When reporting on algorithmic management (e.g., worker surveillance) or border enforcement, individual identifiable narratives must be entirely abstracted unless explicit, high-level, risk-informed consent is granted by the first-person source. Institutional research bodies must formally commit to utilizing legal mechanisms, such as certificates of confidentiality, to resist subpoenas that attempt to uncover the identities of whistleblowers or vulnerable participants.

Selection, Nonresponse, and Representativeness Validations

Algorithmic harm is most acutely felt by those least likely to be captured in standard digital surveys, such as those without stable internet access, the unhoused, and non-English speakers. To mitigate selection bias, researchers must actively sample beyond digital-only methodologies. Over-sampling of multiply-marginalized groups (e.g., disabled persons of color) is required to accurately capture intersectional algorithmic harms. Any representative evidence must be accompanied by an explicit audit detailing the demographic makeup of the sample versus the affected population. If nonresponse bias is detected—for example, undocumented migrants refusing to participate due to legitimate fears of the CBP One app's tracking capabilities—the research must explicitly state this limitation and adjust the weight of the institutional conclusions accordingly. Crucially, researchers are strictly prohibited from generating synthetic quotations, inventing participant endorsements, or extrapolating prevalence claims beyond statistically valid confidence intervals. Through the rigorous application of this protocol, the documentation of algorithmic harm honors the dignity, privacy, and autonomy of the affected individuals, transforming their lived evidence into unimpeachable regulatory mandates.

Works cited

1. Illinois Appellate Court Narrows Third-Party BIPA Liability | Saul Ewing LLP, https://www.saul.com/insights/blog/illinois-appellate-court-narrows-third-party-bipa-liability

2. Sherman v. Brandt Industries USA LTD, No. 1:2020cv01185 \- Document 27 (C.D. Ill. 2020), https://law.justia.com/cases/federal/district-courts/illinois/ilcdce/1:2020cv01185/79931/27/

3. iSolved fingerprint time clocks $2.5M class action lawsuit settlement, https://topclassactions.com/lawsuit-settlements/closed-settlements/isolved-fingerprint-time-clocks-2-5m-class-action-lawsuit-settlement/

4. Santos et al. v. United Scrap Metal, Inc. \- 2020CH05431 \- Class Action Lawsuits, https://www.classaction.org/media/santos-et-al-v-united-scrap-metal-inc.pdf

5. BIPA: The Illinois' Biometric Information Privacy Act What Every Illinois Attorney Should Know \- DuPage County Bar Association, https://www.dcba.org/mpage/v35-Donald-L-Buresh

6. Major Biometric Win for Business in Illinois: 3 Lessons as Federal Appeals Court Says BIPA Damages Limit Applies Retroactively to Pending Cases \- Fisher Phillips, https://www.fisherphillips.com/en/insights/insights/major-biometric-win-for-business-in-illinois

7. Clear as Mud: How the Amendment to the Biometric Information Privacy Act Impacts Pending Lawsuits, https://www.dcba.org/mpage/V37-Joseph-K-Nichele

8. Judge dismisses biometric data privacy lawsuit citing revised state law | Capitol News Illinois, https://capitolnewsillinois.com/news/judge-dismisses-biometric-data-privacy-lawsuit-citing-revised-state-law/

9. Plaintiff, v. Defendants. No. 24 CV 4194 Judge Georgia N. Alexakis Reginald Clay brought this suit against Union Pacific Railroa, https://blogs.duanemorris.com/classactiondefense/wp-content/uploads/sites/56/2025/06/067132648719.pdf

10. Illinois BIPA's Healthcare Exemption Not Limited to Patient Information, Illinois Supreme Court Rules \- Jackson Lewis, https://www.jacksonlewis.com/insights/illinois-bipas-healthcare-exemption-not-limited-patient-information-illinois-supreme-court-rules

11. Algorithmic Management Systems \- Dehumanization, Discrimination and Deskilling, https://www.techimpactonworkers.iccr.org/4-algorithmic-management-systems

12. Algorithmic management in a work context \- The University of Texas at Austin, https://pages.ischool.utexas.edu/hai-files/files/publications/56/2021-BigDataSociety-AlgorithmicManagement.pdf

13. Workers' Health under Algorithmic Management: Emerging Findings and Urgent Research Questions \- PMC, https://pmc.ncbi.nlm.nih.gov/articles/PMC9859016/

14. Case: Louis v. SafeRent Solutions, LLC \- Civil Rights Litigation Clearinghouse, https://clearinghouse.net/case/45888/

15. UNITED STATES DISTRICT COURT FOR THE DISTRICT OF MASSACHUSETTS MARY LOUIS AND MONICA DOUGLAS, ON BEHALF OF THEMSELVES AND SIMILA \- NCLC.org, https://www.nclc.org/wp-content/uploads/2024/05/15-amended-complaint.pdf

16. SafeRent Solutions Accused of Illegally Discriminating Against Black and Hispanic Rental Applicants \- NCLC, https://www.nclc.org/saferent-solution-accused-of-illegally-discriminating-against-black-and-hispanic-rental-applicants/

17. Louis, et al. v. SafeRent Solutions, et al. \- Cohen Milstein, https://www.cohenmilstein.com/case-study/louis-et-al-v-saferent-solutions-et-al/

18. RealPage rental software $360 million price-fixing settlement: Who can claim and how to file, https://www.claimdepot.com/settlements/realpage-rental-settlement

19. No evidence to show Robodebt caused more than 2,000 deaths \- RMIT University, https://www.rmit.edu.au/news/factlab-meta/robodebt-not-directly-responsible-for-more-than-2000-deaths

20. Robodebt royal commission findings revealed, individuals referred for prosecution | ABC News \- YouTube, https://www.youtube.com/watch?v=zDi3Tsc33iQ

21. Full article: Robodebt and the limits of learning: exploring meaning-making after a crisis, https://www.tandfonline.com/doi/full/10.1080/13501763.2025.2513651

22. Criminal charges, 'cruel' treatment: The main takeaways from the robodebt report \- SBS, https://www.sbs.com.au/news/article/criminal-charges-cruel-treatment-the-main-takeaways-from-the-robodebt-report/5jphhfnw6

23. Robodebt scheme \- Wikipedia, https://en.wikipedia.org/wiki/Robodebt\_scheme

24. Hansard \- House of Representatives 3/11/2025 Parliament of Australia, https://www.aph.gov.au/Parliamentary\_Business/Hansard/Hansard\_Display?bid=chamber/hansardr/28869/\&sid=0107

25. Robodebt Royal Commission \- summary of report \- under embargo \- Department of the Prime Minister and Cabinet, https://www.pmc.gov.au/sites/default/files/foi-logs/foi-2024-026.pdf

26. ID.me Helps States Release Blocked Unemployment Claims with Fast Identity Verification, https://network.id.me/press-releases/id-me-helps-states-release-blocked-unemployment-claims-with-fast-identity-verification/

27. unemployment insurance fraud \- nascio, https://www.nascio.org/awards-library/awards/unemployment-insurance-fraud-fraud-detection-and-prevention-through-the-power-of-data-and-analytics/

28. Student monitoring persists post-pandemic \- Privacy International, https://privacyinternational.org/examples/4975/student-monitoring-persists-post-pandemic

29. Students and Teachers Warn That Schools' Use of Monitoring Software Is Harming Students \- PR Newswire, https://www.prnewswire.com/news-releases/students-and-teachers-warn-that-schools-use-of-monitoring-software-is-harming-students-301599245.html

30. School-Based Online Surveillance of Youth: Systematic Search and Content Analysis of Surveillance Company Websites \- Journal of Medical Internet Research, https://www.jmir.org/2025/1/e71998/

31. School-Based Online Surveillance of Youth: Systematic Search and Content Analysis of Surveillance Company Websites \- PMC, https://pmc.ncbi.nlm.nih.gov/articles/PMC12262101/

32. Surveillance Tools Are Supposed To Keep Students Safe. Are They H \- EdSurge, https://www.edsurge.com/news/2022-08-09-student-monitoring-is-supposed-to-keep-students-safe-is-it-harming-student-health

33. Disabling AI: power, exclusion, and disability \- Taylor & Francis, https://www.tandfonline.com/doi/full/10.1080/01425692.2025.2519482

34. Toward a Taxonomy of Algorithmic Harms for Disability: A Systematic Review \- AAAI Publications, https://ojs.aaai.org/index.php/AIES/article/download/36745/38883/

35. Considerations for AI fairness for people with disabilities | Request PDF \- ResearchGate, https://www.researchgate.net/publication/337851183\_Considerations\_for\_AI\_fairness\_for\_people\_with\_disabilities

36. Algorithmic harms and digital ageism in the use of surveillance technologies in nursing homes \- PMC, https://pmc.ncbi.nlm.nih.gov/articles/PMC9525107/

37. Biometric Privacy at Work: Fingerprint Clocks, Facial Recognition, https://dev.to/tiamatenity/biometric-privacy-at-work-fingerprint-clocks-facial-recognition-and-what-your-employer-knows-50j9

38. CDT & AAPD – Building A Disability-Inclusive AI Ecosystem, https://www.aapd.com/wp-content/uploads/2025/03/2025-03-11-CDT-Building-A-Disability-Inclusive-AI-Ecosystem-report-final.pdf

39. Examining the Use of Game-Based Assessments for Hiring Autistic Job Seekers \- MDPI, https://www.mdpi.com/2079-3200/9/4/53

40. Report on Concerns Regarding Online Administration of Bar Exams \- National Disabled Law Students Association, https://ndlsa.org/wp-content/uploads/2020/08/NDLSA\_Online-Exam-Concerns-Report1.pdf

41. Comments to the CPPA on Proposed Regulations Regarding Cybersecurity, Risk Assessments, and ADMTs \- Epic.org, https://epic.org/documents/comments-to-the-cppa-on-proposed-regulations-regarding-cybersecurity-risk-assessments-and-admts/

42. Content Moderation with Shadowbanning | Information Systems Research, https://pubsonline.informs.org/doi/10.1287/isre.2024.1140

43. Shadowbanning, https://d-nb.info/1358292213/34

44. TikTok and Our Shadowbanned Future \- Cato Institute, https://www.cato.org/commentary/tiktok-our-shadowbanned-future

45. Shadowbanning: Some marginalized social media users believe their content is suppressed, https://www.si.umich.edu/about-umsi/news/shadowbanning-some-marginalized-social-media-users-believe-their-content-suppressed

46. Content Moderation with Shadowbanning \- Haskayne School of Business \- University of Calgary, https://haskayne.ucalgary.ca/sites/default/files/teams/100/Pubs%20LWV/Journal%20Articles/2025-2029/2025\_Content%20Mod\_ISR.pdf

47. Shaping opinions in social networks with shadow banning \- PMC \- NIH, https://pmc.ncbi.nlm.nih.gov/articles/PMC10971755/

48. Full article: “Shadowbanning is not a thing”: black box gaslighting and the power to independently know and credibly critique algorithms \- Taylor & Francis, https://www.tandfonline.com/doi/full/10.1080/1369118X.2021.1994624

49. Replica Digital Voice Replica Development Agreement \- SAG-AFTRA, https://www.sagaftra.org/sites/default/files/sa\_documents/Replica%20Studios%20Agreement%20for%20Digital%20Voice%20Replicas\_0.pdf

50. Think Outside the Data: Colonial Biases and Systemic Issues in Automated Moderation Pipelines for Low-Resource Languages \- arXiv, https://arxiv.org/html/2501.13836v1

51. Linguistic Inequity in Facebook Content Moderation \- Technology Science, https://techscience.org/a/2025022501/

52. Ethical Scaling for Content Moderation: Extreme Speech and the (In)Significance of Artificial Intelligence \- The Shorenstein Center, https://shorensteincenter.org/resource/ethical-scaling-content-moderation-extreme-speech-insignificance-artificial-intelligence/

53. Glitches in the Digitization of Asylum: How CBP One Turns Migrants' Smartphones into Mobile Borders \- MDPI, https://www.mdpi.com/2075-4698/13/6/149

54. CBP One Mobile Application Violates the Rights of People Seeking Asylum in the United States \- Amnesty International, https://www.amnesty.org/en/latest/news/2024/05/cbp-one-mobile-application-violates-the-rights-of-people-seeking-asylum-in-the-united-states/

55. Facial Recognition \- ACLU of Michigan, https://www.aclumich.org/cases/facial-recognition/

56. "It didn't make sense at all": Wrongful facial recognition arrest in Detroit leads to landmark settlement \- Michigan Public, https://www.michiganpublic.org/criminal-justice-legal-system/2024-06-28/it-didnt-make-sense-at-all-wrongful-facial-recognition-arrest-leads-to-landmark-settlement

57. When Artificial Intelligence Gets It Wrong \- Innocence Project, https://innocenceproject.org/news/when-artificial-intelligence-gets-it-wrong/

58. Williams v. City of Detroit | American Civil Liberties Union, https://www.aclu.org/cases/williams-v-city-of-detroit-face-recognition-false-arrest

59. Flawed Facial Recognition Technology Leads to Wrongful Arrest and Historic Settlement, https://quadrangle.michigan.law.umich.edu/issues/winter-2024-2025/flawed-facial-recognition-technology-leads-wrongful-arrest-and-historic

60. More than a Dozen Wrongful Arrests Due to Police Reliance on Facial Recognition Technology \- ACLU of Georgia, https://www.acluga.org/news/more-than-a-dozen-wrongful-arrests-due-to-police-reliance-on-facial-recognition-technology/

61. Nijeer Parks \- Michael Oliver \- Randal Reid Robert Williams \- Muni.org, https://www.muni.org/Departments/Assembly/SiteAssets/Pages/Meetings-Worksessions/Work%20Session%20FRT%20Fact%20Sheet%20V3.pdf

62. Neurorights in Chile: Between neuroscience and legal science \- ResearchGate, https://www.researchgate.net/publication/354940433\_Neurorights\_in\_Chile\_Between\_neuroscience\_and\_legal\_science

63. Companionship in Code: Emotional Wellbeing and Design ... \- ePLUS, https://eplus.uni-salzburg.at/Abschlussarbeiten/download/pdf/12628727