Digital Society Harms

Digital Society Harms is the catalogue of empirically documented and structurally anticipated negative impacts arising from the deployment of digital, algorithmic, and AI systems at population scale across information, psychological, economic, social, human-rights, child-safety, environmental, an…

Semantic Classification

Content

Compositional Relationships (Components)

SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:hasPart society:InformationHarms))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:hasPart society:PsychologicalHarms))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:hasPart society:EconomicHarms))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:hasPart society:SocialHarms))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:hasPart society:HumanRightsHarms))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:hasPart society:ChildSafetyHarms))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:hasPart society:EnvironmentalHarms))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:hasPart society:LabourHarms))

## Dependency Relationships
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:requires society:HarmTaxonomy))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:requires society:EvidenceBase))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:requires society:IncidentReporting))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:requires society:RegulatoryFramework))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:dependsOn society:PlatformTransparency))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:dependsOn society:ResearcherDataAccess))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:dependsOn society:WhistleblowerProtection))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:dependsOn society:IndependentAudit))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:dependsOn society:EmpiricalHarmMeasurement))

## Capability Relationships
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:enables society:OnlineSafetyRegulation))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:enables society:AlgorithmicAccountability))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:enables society:HarmMitigation))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:enables society:PlatformLiability))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:enables society:UserEmpowerment))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:supports society:DigitalRightsAdvocacy))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:supports society:ChildSafetyOnline))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:supports society:WorkerProtection))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:supports society:DemocraticResilience))

## Implementation Relationships
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:implements society:OfcomOnlineSafetyActCodes))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:implements society:AdaLovelaceHarmsFramework))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:implements society:MITAIRiskRepositoryTaxonomy))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:implements society:ICOChildrenCode))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:implements society:CouncilOfEuropeHUDERIA))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:uses society:AlgorithmicAudit))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:uses society:ImpactAssessment))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:uses society:TransparencyReporting))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:uses society:IndependentResearch))

## Reduction Relationships
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:reduces society:HumanWellbeing))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:reduces society:InformationIntegrity))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:reduces society:DemocraticDeliberation))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:reduces society:LabourDignity))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:reduces society:ChildSafety))

## Association Relationships
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:relatedTo society:Deepfakes))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:relatedTo society:AlgorithmicBias))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:relatedTo society:SurveillanceCapitalism))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:relatedTo society:AttentionEconomy))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:contrastsWith society:TechnoUtopianism))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:contrastsWith society:IndividualChoiceFraming))
SubClassOf(society:DigitalSocietyHarms
  ObjectSomeValuesFrom(society:contrastsWith society:MarketOnlySolutions))

## Data Properties (Characteristics)
DataPropertyAssertion(society:hasIdentifier society:DigitalSocietyHarms "SO-1001"^^xsd:string)
DataPropertyAssertion(society:authorityScore society:DigitalSocietyHarms "0.87"^^xsd:decimal)
DataPropertyAssertion(society:catalogueSize society:DigitalSocietyHarms "700"^^xsd:integer)
DataPropertyAssertion(society:harmCategoryCount society:DigitalSocietyHarms "8"^^xsd:integer)
DataPropertyAssertion(society:ofcomScheduleCount society:DigitalSocietyHarms "3"^^xsd:integer)
DataPropertyAssertion(society:csamPagesActioned2024 society:DigitalSocietyHarms "291273"^^xsd:integer)
DataPropertyAssertion(society:ewasteMTonnes2022 society:DigitalSocietyHarms "62"^^xsd:integer)

## Annotations
AnnotationAssertion(rdfs:label society:DigitalSocietyHarms "Digital Society Harms"@en)
AnnotationAssertion(rdfs:comment society:DigitalSocietyHarms "Catalogue of empirically documented and structurally anticipated negative impacts arising from deployment of digital and AI systems at population scale, formalised through Ofcom Online Safety Act Schedules 5/6/7, Ada Lovelace Institute Harms Framework, MIT AI Risk Repository 700+ risks, Stanford HAI, and Council of Europe HUDERIA, grouped into information, psychological, economic, social, human-rights, child-safety, environmental, and labour categories with documented incidents (Molly Russell 2022, Sewell Setzer 2024, Cambridge Analytica 2018, Frances Haugen 2021, Snowden 2013, IWF CSAM 291K pages 2024) mapped against UK enforcement landscape (Ofcom OSA Phases 1-3 2025, ICO Children's Code, DSIT, AI Security Institute) and civil-society/academic ecosystem (Ada Lovelace, ORG, OII, LCFI, Manchester/Leeds/Sheffield/Newcastle hubs); contrasts with techno-utopian, individual-choice, and market-only framings."@en)
AnnotationAssertion(dcterms:identifier society:DigitalSocietyHarms "SO-1001"^^xsd:string)
AnnotationAssertion(dcterms:subject society:DigitalSocietyHarms "Online Safety, Algorithmic Harm, Platform Governance, Digital Rights"@en)

)

Property Characteristics

AsymmetricObjectProperty(society:reduces) AsymmetricObjectProperty(society:enables) AsymmetricObjectProperty(society:implements) TransitiveObjectProperty(society:dependsOn) FunctionalDataProperty(society:catalogueSize) FunctionalDataProperty(society:harmCategoryCount)

About Digital Society Harms

• Digital Society Harms is the umbrella ontology category capturing the empirically documented and structurally anticipated negative impacts that arise when digital, algorithmic, and AI systems are deployed at population scale. Unlike treatments that frame harm in single-incident terms (a specific deepfake, a specific moderation failure, a specific data breach), this category insists on the systemic and cumulative character of digital harm — the architectural choices made by platforms, the optimisation targets of recommendation systems, the labour conditions of moderation, the regulatory gaps in data-protection regimes, and the political economy of attention markets collectively produce harm patterns that are predictable, reproducible across platforms, and resistant to interventions that address symptoms rather than structures.
• The category serves three operational purposes within the ontology. First, it provides a taxonomy scaffold linking to formal harm classifications (Ofcom Online Safety Act Schedules 5/6/7, Ada Lovelace Institute Harms Framework 2022, MIT AI Risk Repository 700+ risks Slattery et al. August 2024, Stanford HAI AI Index Incident Database, AI Now Institute reports). Second, it functions as a case-law anchor binding specific high-profile incidents (Molly Russell, Sewell Setzer, Cambridge Analytica, Frances Haugen disclosures, Snowden, IWF CSAM reports, Mobley v Workday, Garcia v Character Technologies) to abstract harm types so that the ontology supports both deductive policy reasoning and inductive case-pattern recognition. Third, it provides the contrast surface against which competing framings (techno-utopianism, individual-choice responsibilisation, market-only solutionism) can be assessed, ensuring the ontology does not silently encode the assumption that digital systems are net-positive instruments awaiting only better users.

The Taxonomic Landscape

Four taxonomies dominate UK and international harm-cataloguing practice, each privileging different ontological commitments.

Ofcom Online Safety Act 2023 (UK): The Act creates a duties-of-care regime for user-to-user services and search services, distinguishing three priority categories. Schedule 5 lists priority illegal content: terrorism (Terrorism Act 2000/2006 offences); child sexual exploitation and abuse (CSEA: making/distributing indecent images of children, grooming, sexual communication with a child); encouraging or assisting suicide; people-trafficking; controlling or coercive behaviour in intimate relationships; serious unlawful immigration; sexual exploitation; harassment, stalking, threats, abuse; hate offences (stirring up racial/religious hatred); intimate-image abuse (so-called revenge pornography and synthetic intimate images per Online Safety Act Section 187 amending Sexual Offences Act 2003); drugs and psychoactive substances offences; firearms, knives and weapons offences; fraud (dishonest financial offences). Schedule 6 lists priority harms to children: pornographic content (Schedule 6 Part 1); content depicting child sexual abuse, racism, hate, bullying, violence promoting self-harm or suicide, instructions for dangerous physical stunts, harmful diet content. Schedule 7 lists user-empowerment categories adults can opt out of: suicide and self-harm content not crossing the criminal threshold, eating-disorder content, abusive content targeting protected characteristics. Phase 1 illegal-harms duties commenced 17 March 2025 following Ofcom’s Codes of Practice published December 2024. Phase 2 child-safety duties commenced 25 July 2025 requiring “highly effective age assurance” for services likely accessed by children. Phase 3 categorised-service duties (Category 1: largest user-to-user platforms; Category 2A: largest search; Category 2B: smaller user-to-user above threshold) entered implementation late 2025 with transparency-report and user-empowerment requirements.

Ada Lovelace Institute Harms Framework (2022): Distinguishes five harm types: (1) Allocative harms — when AI denies opportunities, resources, or information disproportionately (mortgage denial, hiring rejection, benefit termination); (2) Quality-of-service harms — when systems work worse for some groups than others (facial recognition failing on darker skin per Gender Shades Buolamwini/Gebru 2018, voice recognition failing on regional accents per Koenecke et al. 2020); (3) Representational harms — when systems reinforce stereotypes or denigrate identity (image search returning offensive results, generative models producing biased outputs); (4) Loss-of-liberty harms — when systems enable unjustified surveillance, detention, or restriction (wrongful arrests via facial recognition, predictive policing reinforcement); (5) Social-system harms — when aggregate deployment reshapes institutions and norms (filter bubbles, attention markets, deliberative-democracy degradation). This framework underpins much UK policy discourse and informed the Centre for Data Ethics and Innovation (CDEI, closed September 2023, functions absorbed by DSIT) approach.

MIT AI Risk Repository (Slattery et al., August 2024, arXiv:2408.12622): Aggregates 700+ unique risks extracted from 43 risk taxonomies and frameworks. Provides a causal taxonomy classifying by entity (human/AI), intent (intentional/unintentional), and timing (pre-deployment/post-deployment), and a domain taxonomy with seven domains: (1) discrimination and toxicity; (2) privacy and security; (3) misinformation; (4) malicious actors and misuse; (5) human-computer interaction; (6) socioeconomic and environmental harms; (7) AI system safety, failures and limitations. This is the most comprehensive AI-specific harm catalogue in 2024-2025 academic literature, hosted as a living resource at airisk.mit.edu and updated quarterly.

Council of Europe HUDERIA (2024): The Human Rights, Democracy and Rule of Law Impact Assessment methodology under the Framework Convention on Artificial Intelligence opened for signature 5 September 2024 — the first legally binding international AI treaty. HUDERIA structures impact assessment around Convention obligations (Articles 4-12) covering dignity, non-discrimination, privacy, transparency, accountability, equality, safe innovation, and rule-of-law preservation.

Components / Architecture

• The harm catalogue decomposes into eight overlapping categories, each with documented incidents, regulatory hooks, and measurement instruments.

1. Information Harms

Information harms comprise epistemic distortions introduced by algorithmic content production, ranking, and amplification. The category overlaps with Deepfakes and fraudulent content (sibling concept) but extends beyond synthetic media to encompass attention-economy distortions of which deepfakes are one instance.

Disinformation production at scale: Generative AI reduces the marginal cost of producing plausible disinformation to near-zero. NewsGuard tracked 1,150+ AI-generated unreliable news sites in 50+ languages by end-2024, up from 49 in May 2023. Fact-checking organisations (Snopes, Full Fact UK, FactCheck.org, AFP Fact Check) reported a 3-5× increase in submitted claims requiring assessment 2023-2025.

Deepfake-enabled fraud: The Hong Kong Arup deepfake CFO incident (January-February 2024) saw a finance employee transfer HK$200million(US$25.6 million) to scammers after a video-conference attended entirely by deepfaked colleagues including a synthetic CFO. The Slovakian Šimečka audio (29 September 2023, two days before parliamentary elections) circulated a fabricated recording purporting to show Progressive Slovakia leader Michal Šimečka discussing vote-buying with a journalist; Progressive Slovakia narrowly lost.

Algorithmic amplification of low-quality content: Engagement-optimisation ranking systematically over-weights content producing strong emotional reactions (outrage, fear, novelty). Frances Haugen’s Facebook Files disclosures (September 2021) included internal Facebook research showing that the 2018 Meaningful Social Interactions (MSI) ranking change increased angry and divisive content by ~5× over the prior News Feed algorithm, with internal staff documenting this in 2019 and senior leadership declining to reverse the change.

2. Psychological Harms

Psychological harms include algorithmic amplification of self-harm content, eating-disorder promotion, body-image distortion, addiction architectures, and attention fragmentation.

Molly Russell inquest: Molly Russell, aged 14, died by suicide in November 2017. After a five-year investigation, North London Coroner Andrew Walker ruled on 30 September 2022 that she “died from an act of self-harm whilst suffering from depression and the negative effects of online content”. The inquest examined ~16,300 pieces of content Molly had saved, liked or shared on Instagram and Pinterest in the six months before her death, of which 2,100 related to depression, self-harm or suicide; Meta executive Elizabeth Lagone testified that Meta considered the content “safe” by community standards, which the coroner explicitly rejected. The ruling — the first in UK history to formally name online content as a contributing cause of death — catalysed the Online Safety Act 2023 and remains the canonical citation for psychological harm from algorithmic amplification.

TikTok eating-disorder amplification: The Wall Street Journal “It’s a Numbers Game” series (December 2021) documented that newly-created accounts identifying as 13-year-old girls were served eating-disorder content within minutes of viewing weight-related videos. The Center for Countering Digital Hate “Deadly by Design” report (December 2022) found new teen accounts encountered self-harm content within a median 2.6 minutes and eating-disorder content within 8 minutes. TikTok announced policy changes 2023-2024 but third-party audits continue to identify gaps.

Body image and the Facebook Files: Internal Facebook research disclosed by Frances Haugen included slides reporting that 32% of teen girls said Instagram made body-image issues worse, and that for some teens Instagram-related social-comparison was “destabilising” and connected to suicidal ideation. The Wall Street Journal published the slides September 2021; Haugen testified to US Senate Commerce Subcommittee 5 October 2021 and UK Parliament Joint Committee on the Draft Online Safety Bill 25 October 2021.

Addiction architectures and dopamine loops: Tristan Harris (former Google Design Ethicist) and the Center for Humane Technology (founded 2018) document the deliberate porting of B.F. Skinner’s variable-ratio reinforcement schedules into product surfaces (pull-to-refresh, autoplay, infinite scroll, slot-machine-style notification batching). Harris’s Senate Commerce Committee testimony (25 June 2019) and the 2020 Netflix documentary The Social Dilemma brought addiction-architecture analysis to mainstream audiences. PMC meta-analyses (2021-2024) document structural grey-matter differences in problematic internet users analogous to substance-addiction neuroimaging signatures (Nature Molecular Psychiatry 2021).

3. Economic Harms

Gig-economy precarity: The UK Supreme Court ruled in Uber BV v Aslam ([2021] UKSC 5, 19 February 2021) that Uber drivers are “workers” under the Employment Rights Act 1996 entitled to minimum wage and paid leave, not independent contractors. The ruling triggered cascading litigation: Deliveroo riders’ collective-bargaining cases (UK Supreme Court ruled against riders November 2023 on Article 11 ECHR grounds), Bolt driver cases (ongoing 2024-2025), Just Eat Takeaway transition to employed-courier model 2023-2024.

Discriminatory hiring: Amazon scrapped its recruitment ML tool in 2018 after Reuters revealed it penalised CVs containing the word “women’s” or referencing all-women’s colleges, having been trained on a decade of mostly-male engineering CVs. HireVue discontinued facial-expression analysis November 2020 after an EPIC FTC complaint argued the technology constituted pseudoscientific deception. Mobley v Workday (Northern District California, filed February 2023 by Derek Mobley alleging algorithmic age, race, and disability discrimination in screening) was certified as a collective action 16 May 2025, the first such certification against an AI HR platform in US courts.

Algorithmic management and automated layoffs: Xsolla terminated 150 staff August 2021 after an algorithm flagged them as “unengaged” based on Slack and Jira activity. Amazon warehouse “time-off-task” surveillance has been cited in OSHA filings and a 2023 California Senate Bill 1167. Uber Eats facial-recognition discrimination cases brought by IWGB before UK Employment Tribunal 2024 cited Microsoft Face API false-rejection rates against Black couriers, leading to settlements 2024-2025.

Pricing discrimination and dynamic-price markets: Algorithmic pricing in ride-hail (Uber surge), e-commerce (Amazon dynamic pricing), insurance (Admiral pricing experiments 2022 ICO investigation), and gig labour-allocation (Deliveroo Frank algorithm, Just Eat courier-allocation) raises concerns about hidden discriminatory effects. The UK Centre for Data Ethics and Innovation Review of Online Targeting (2020) and ICO Age Appropriate Design Code enforcement of “transparency” obligations have begun to constrain opaque pricing personalisation, but enforcement remains thin.

Welfare-state algorithmic harm: The Dutch SyRI ruling (Netherlands District Court The Hague, 5 February 2020, NJCM v The Netherlands ECLI:NL:RBDHA:2020:865) found the System Risk Indication welfare-fraud detection system breached ECHR Article 8 due to opacity and disproportionality, leading to the wider Toeslagenaffaire (Childcare Benefits Scandal) in which an algorithm flagged some 26,000 families — many of dual nationality or with foreign-sounding names — as fraudsters, with the Rutte III cabinet resigning January 2021. UK Department for Work and Pensions (DWP) algorithmic-fraud-detection tools challenged by Big Brother Watch judicial review 2023-2024; A-level algorithm 2020 controversy (Ofqual algorithm downgrading 40% of teacher-assessed grades disproportionately affecting state-school students) led to algorithm withdrawal August 2020 after public outcry and Foxglove Legal pre-action.

4. Social Harms

Political polarisation: Beyond Haugen’s MSI disclosures, the NYU Stern Center for Business and Human Rights (2021) and Royal Society “Online Information Environment” report (January 2022) documented engagement-driven polarisation. The NESTA Centre for Collective Intelligence Design UK programme 2022-2024 investigated deliberative interventions.

Election interference: Cambridge Analytica/Facebook (Christopher Wylie whistleblower disclosure March 2018) saw the harvesting of 87M Facebook profiles via the “thisisyourdigitallife” app for targeting in the 2016 Trump and Vote Leave Brexit campaigns. The UK ICO fined Facebook £500,000 (October 2018, the maximum under pre-GDPR Data Protection Act 1998), FTC settled with Facebook for $5 billion (24 July 2019), and SCL Group entered administration. The US Senate Intelligence Committee Volume 2 (2019) documented Russian Internet Research Agency operations 2014-2017 targeting US elections.

Microtargeting and political advertising: The EU Digital Services Act (entered force February 2024 for very-large online platforms) and the EU Regulation 2024/900 on transparency and targeting of political advertising (entered force February 2024, fully applicable October 2025) prohibit political microtargeting using special-category data per GDPR Article 9. UK regulation lags: the Election Act 2022 introduced an imprint regime for digital political advertising but stopped short of microtargeting limits, with Electoral Commission and ICO sharing enforcement responsibilities. The 2024 UK general election (4 July 2024) saw the first large-scale Labour deployment of generative AI for ad personalisation; the Conservatives were subject to ICO scrutiny for use of Tory-developed data tools. Generative AI in elections: The Slovakian Šimečka audio, the 2024 New Hampshire Biden robocall deepfake (Steve Kramer indicted May 2024), Indonesia synthetic Suharto endorsement video January 2024, and India general-election 2024 deepfake-heavy campaigning catalysed FCC ban on AI-generated robocall voices February 2024 and EU AI Act Article 50 watermarking obligations from August 2026.

Misinformation amplification and source quality collapse: The Stanford Internet Observatory (closed June 2024 following political pressure), the University of Washington Center for an Informed Public, the Election Integrity Partnership (wound down 2024 after House Judiciary subpoenas), and academic groups tracking election-period misinformation have faced political and legal pressure 2023-2024 weakening the misinformation-research infrastructure. Murthy v Missouri (US Supreme Court, decided 26 June 2024) held plaintiffs lacked standing to challenge federal-government communications with platforms about misinformation, but the case chilled platform-government information sharing nonetheless. Open-source equivalent infrastructure (CheckMyChips, Logically, Full Fact UK, Newtral Spain) is increasingly the only systematic monitoring layer.

Community fragmentation and parasocial substitution: Increasing screen-time consumption substitutes for face-to-face interaction, with OECD Society at a Glance 2024 noting 27% reduction in average daily face-to-face socialising 2003-2023 across OECD members. The rise of AI companions (Replika, Character.AI, Talkie, Kindroid) introduces new categories of parasocial relationship; 2024-2025 saw early empirical work (Marshall et al. CHI 2024, Skjuve et al. 2023) documenting both loneliness-reduction effects and concerns about substitution for human relationship formation in adolescents.

5. Human Rights Harms

Mass surveillance: The Snowden disclosures (June 2013, via The Guardian and Washington Post) revealed NSA PRISM (collecting from Google, Microsoft, Yahoo, Facebook, Apple, etc.) and GCHQ Tempora (tapping submarine cables at Bude Cornwall). UK responses included the Investigatory Powers Act 2016 (“Snoopers’ Charter”) and CJEU Schrems II (16 July 2020, Case C-311/18) invalidating EU-US Privacy Shield over inadequate US-government access protections.

Predictive policing: PredPol/Geolitica shut down 2024 after The Markup’s October 2023 investigation analysed 7.8 million predictions across 38 US police departments, finding 84-95% of predictions reinforced patrol patterns in Black, Latino, and lower-income neighbourhoods. Durham Constabulary HART (Harm Assessment Risk Tool) retired 2020 after ALI research raised proportionality and accuracy concerns. NYPD Compstat continues but with reduced predictive-policing emphasis post-Floyd 2020.

Facial-recognition wrongful arrests (all US, all Black Americans, all involving Clearview AI or municipal facial-recognition deployments): Robert Williams (Detroit, January 2020); Nijeer Parks (Woodbridge NJ, February 2019); Porcha Woodruff (Detroit, February 2023, arrested while eight-months pregnant); Randal Reid (Louisiana, November 2022, held six days for a Jefferson Parish robbery despite never having visited the state). Settlements 2023-2025 ranged $300K-$4M.

EU AI Act Article 5 prohibitions (Regulation 2024/1689, entered force August 2024, prohibitions applicable from 2 February 2025): ban real-time biometric identification in public spaces for law enforcement (with narrow exceptions), social-scoring by public authorities or private actors, emotion recognition in workplaces and educational institutions, predictive policing solely based on profiling, untargeted scraping of facial images, manipulative subliminal techniques, and exploitation of vulnerabilities of specific groups.

6. Child Safety Harms

CSAM production and growth: The Internet Watch Foundation 2024 Annual Report documented 291,273 webpages confirmed actioned containing child sexual abuse material — a 92% increase over 2023. The IWF October 2024 update found 18% of newly-discovered CSAM in 2024 contained AI-generated imagery, with self-generated content (often from “sextortion” coercion) comprising 91% of CSAM involving 11-13-year-olds.

Sewell Setzer III and Garcia v Character Technologies: Sewell Setzer III, aged 14, died by suicide 28 February 2024 in Orlando, Florida, after months of intensive use of Character.AI, including roleplay with a “Daenerys” chatbot to which he had become emotionally attached. His mother Megan Garcia filed Garcia v Character Technologies Inc., Google LLC et al. in the Middle District of Florida on 22 October 2024 (Case 6:24-cv-01903) alleging strict product liability, negligence, deceptive trade practices, and wrongful death. 22 May 2025: Judge Anne Conway denied most of the defendants’ motion to dismiss, allowing product-defect, failure-to-warn, and unjust-enrichment claims to proceed — a landmark ruling that AI chatbots can be treated as defective products. Partial settlement reportedly reached January 2026 with safety commitments and undisclosed compensation.

Sextortion of boys: The FBI/DOJ joint advisory January 2024 noted at least 20 US suicides 2021-2024 linked to financial sextortion targeting boys aged 14-17, with perpetrators (often West African organised crime networks per UK NCA 2024 alert) coercing victims into sending images then demanding payment under threat of distribution.

UK regulatory frame: The ICO Age-Appropriate Design Code (Children’s Code) entered force 2 September 2021, requiring data-protection-by-design with 15 standards (high-privacy defaults, no nudge techniques, age-appropriate data minimisation). Online Safety Act Phase 2 child-safety duties commenced 25 July 2025, requiring “highly effective age assurance” for services likely to be accessed by children and proactive removal of priority harmful content. 5Rights Foundation (founded by Baroness Beeban Kidron) provides specialised child-rights advocacy.

7. Environmental Harms

Data-centre electricity: The IEA Electricity 2024 report (January 2024) projects global data-centre electricity demand will reach 1,000 TWh by 2026, doubling from ~460 TWh in 2022. AI workloads are the fastest-growing component. Google sustainability report 2024: 24.2 billion litres of water consumed 2023 (mostly evaporative cooling), 14.3 MtCO2e total emissions (up 48% since 2019, primarily from data-centre electricity). Microsoft sustainability report 2024: Scope 3 emissions up 30.9% since 2020. Meta sustainability report 2024: 2023 emissions 9.5 MtCO2e.

Training emissions: Strubell et al. (2019) “Energy and Policy Considerations for Deep Learning in NLP” established the methodology baseline: training BERT-base emits 0.65 tCO2e, training a Transformer-Big with neural architecture search emits 284 tCO2e. GPT-3 training (175B params, 2020) estimated 552 tCO2e. GPT-4-scale runs (2023-2024, 1-2T params) estimated 5,000-10,000 tCO2e per training run.

E-waste: The UN Global E-waste Monitor 2024 documented 62 million tonnes generated globally 2022, with recycling lagging at 22.3%. Embedded carbon of GPUs (Nvidia H100 ~150 kg CO2e per unit), TPUs, and consumer devices is largely unreported.

8. Labour Harms

Crowdwork: Mechanical Turk workers earned median **$2/hourwithoutbenefits,appealrights,orlabour-lawprotection\ast \ast perHaraetal.CHI2018studyof3.8milliontasks;subsequentupdates(Haraetal.2023)showmedianwagesstagnant.Argentineworkerslabellingautonomous-vehicledataviaSamaandScaleearn$1-3/hour.

Kenyan content moderation for OpenAI: Time magazine investigation “OpenAI Used Kenyan Workers on Less Than $2PerHourtoMakeChatGPTLessToxic"\ast \ast (BillyPerrigo,18January2023)documentedSamacontractorspaid\ast \ast$1.32-$2/hour to label graphic CSAM, bestiality, suicide, and violence content as RLHF-safety training data. Sama terminated the contract February 2023. Daniel Motaung v Meta/Sama Nairobi labour court action filed May 2022, Meta lost jurisdictional appeal September 2023, case ongoing 2025. Separate 184 Facebook moderators class action filed 2023.

Philippine and other Global South moderation: Manila is the world’s largest content-moderation hub by headcount, with TikTok, Meta, and YouTube outsourcing to Accenture, Telus International, and TaskUs. PTSD and trauma documented in Pia Lopez (NYU dissertation 2023) and the Coalition of Asian Content Moderators 2023 statement.

Use Cases / Major Families

• The harm catalogue is operationalised across four major use-case families:
• Policy formulation and parliamentary inquiry: UK Parliament Joint Committee on the Draft Online Safety Bill (2021), Lords Communications and Digital Committee inquiries on AI in news (2023-2024), House of Commons Science Innovation and Technology Committee AI Governance inquiry 2024 use harm taxonomies to scope evidence and structure recommendations.
• Regulatory enforcement: Ofcom Codes of Practice (December 2024 illegal harms, May 2025 child safety) cite specific incidents and harm types when setting “highly effective” thresholds for risk assessment, proactive technology deployment, and transparency reporting.
• Civil-society advocacy and litigation: Ada Lovelace Institute, Open Rights Group, Privacy International, Foxglove Legal, AWO, Big Brother Watch, 5Rights, and Center for Countering Digital Hate use the catalogue to ground subject-access requests, judicial reviews (Bridges v South Wales Police 2020 facial recognition), strategic litigation, and parliamentary briefings.
• Corporate AI risk management: ISO/IEC 42001:2023 AI Management Systems and NIST AI RMF Generative AI Profile (July 2024) cite harm taxonomies to scope risk identification, with EU AI Act high-risk systems requiring conformity assessment against fundamental-rights harms.

Academic Context

• Digital harm scholarship sits at the intersection of media studies, science and technology studies (STS), socio-legal studies, human-computer interaction, computer science, philosophy of technology, and political economy.
• Foundational works: Lawrence Lessig Code and Other Laws of Cyberspace (1999); Lawrence Lessig Code: Version 2.0 (2006) introducing the regulability-by-architecture framing; Shoshana Zuboff The Age of Surveillance Capitalism (2019) extending Polanyi to data; Cathy O’Neil Weapons of Math Destruction (2016) coining “WMD” for opaque, scaled, damaging algorithms; Safiya Umoja Noble Algorithms of Oppression (2018) on search-engine racism; Virginia Eubanks Automating Inequality (2018) on welfare-state algorithmic harm; Ruha Benjamin Race After Technology (2019); Frank Pasquale The Black Box Society (2015).
• Surveillance and policing: David Lyon Surveillance Studies (2007); Bernard Harcourt Exposed (2015); Andrew Guthrie Ferguson The Rise of Big Data Policing (2017).
• Attention economy and psychological harm: Tim Wu The Attention Merchants (2016); Johann Hari Stolen Focus (2022); Jean Twenge iGen (2017) and Generations (2023); Jonathan Haidt The Anxious Generation (March 2024) consolidating the “smartphone-rewiring-childhood” thesis with global anxiety/depression timelines.
• AI ethics and harm cataloguing: Kate Crawford Atlas of AI (2021) on materiality and labour; Meredith Broussard Artificial Unintelligence (2018) and More than a Glitch (2023); Slattery et al. AI Risk Repository arXiv:2408.12622 (August 2024); Solaiman et al. Evaluating the Social Impact of Generative AI Hugging Face / Partnership on AI June 2023; Weidinger et al. Sociotechnical Safety Evaluation of Generative AI Systems Google DeepMind 2023.
• Empirical platform research: Christian Sandvig et al. on auditing algorithms (2014); Joy Buolamwini & Timnit Gebru Gender Shades (FAT* 2018) showing facial-recognition disparities up to 34.7%; Sandra Wachter & Brent Mittelstadt counterfactual explanations and right to explanation (Oxford OII).
• UK and Northern English academic ecosystem:
  • Oxford Internet Institute (OII) founded 2001, Professor Philip Howard (computational propaganda), Professor Vili Lehdonvirta (digital economy and platform labour), Professor Sandra Wachter (algorithmic accountability and law)
  • Cambridge Leverhulme Centre for the Future of Intelligence (LCFI) founded 2016 with Professor Stephen Cave, Dr Kanta Dihal (narratives), Dr Eleanor Drage (gender and AI)
  • Cambridge Minderoo Centre for Technology and Democracy founded 2021 with Professor Gina Neff
  • UCL Institute for Innovation and Public Purpose (IIPP) Professor Mariana Mazzucato; UCL Knowledge Lab on digital education
  • Imperial College London Institute for Security Science and Technology (ISST) AI safety and infrastructure; Department of Computing Professor Toby Walsh-adjacent work on adversarial ML
  • Edinburgh Centre for Technomoral Futures founded 2020 with Professor Shannon Vallor (author of Technology and the Virtues 2016 and The AI Mirror 2024), part of the £79M Edinburgh Futures Institute
  • King’s College London Policy Institute AI-society interface; Department of Digital Humanities
  • LSE Media and Communications Professor Sonia Livingstone (children’s online experience, EU Kids Online network)
  • Manchester Digital Trust and Security with Cyber Security Centre and Digital Futures programme
  • Leeds Digital Futures Society and Bauman Institute (digital sociology heritage); School of Politics and International Studies critical AI scholarship
  • Sheffield AI Research Centre AI4Industry programme; iSchool research on platform labour
  • Newcastle Open Lab human-computer interaction with deep community-engagement methodology; School of Computing critical-data-studies
  • Lancaster Data Science Institute and Centre for Mobilities Research
  • Liverpool John Moores Centre for Advanced Internet Studies

Current Landscape (2026)

• Regulatory enforcement: Ofcom Online Safety Act Phase 1 illegal-harms regime operational since 17 March 2025, with Phase 2 child-safety duties since 25 July 2025. First enforcement notices issued late 2025 against small file-hosting and image-board services. Ofcom Annual Online Safety Report November 2025 documented 47 services issued information notices; transparency reports for Category 1 services due Spring 2026. The EU AI Act prohibitions on real-time biometric identification, social scoring, emotion recognition in workplaces, and predictive policing solely based on profiling have been applicable since 2 February 2025; AI Office investigations under way 2025-2026.
• Litigation: Garcia v Character Technologies partial settlement reported January 2026 with safety commitments. Mobley v Workday certified as collective action 16 May 2025 with discovery ongoing through 2026. UK Investigatory Powers Tribunal hearing IPT/22/55/CH on bulk personal datasets concluded December 2025 ruling against MoD/GCHQ on certain notice issues. Bridges v South Wales Police follow-on cases involving Live Facial Recognition continued 2024-2026.
• Industry response: Major platforms have published Online Safety Act compliance summaries (Meta, Snap, TikTok, X, Discord, Reddit, BlueSky, Mastodon hosts). Age-assurance markets have consolidated around Yoti, AU10TIX, Persona, Onfido (now Entrust), and Privately. AI-content labelling pilots (C2PA Content Credentials, Adobe Content Authenticity Initiative, IPTC Photo Metadata) progressing but adoption uneven.
• Civil society: Ada Lovelace Institute published “AI Assurance” report June 2025 and “Beyond Foundation Models” December 2025. 5Rights launched “Children’s Code Strategic Litigation Fund” 2024. Open Rights Group filed strategic-litigation challenges to Online Safety Act encryption provisions 2024-2025. Foxglove Legal continues platform-worker litigation including Uber Eats and Deliveroo cases.
• Academic research: AI Risk Repository (MIT) v2 expected mid-2026 extending the 700+ taxonomy. Stanford HAI AI Index 2026 (March 2026) documenting incident growth, model-collapse research, and harm-mitigation metrics. International AI Safety Report 2026 (Bengio chair, follow-up to 2025 inaugural report) commissioned by UK AISI/AI Action Summit Paris February 2025.

UK Context

• Regulators and government bodies:
  • Ofcom — Online Safety Act regulator with broad enforcement powers including blocking, fines up to £18M or 10% global turnover (whichever greater), senior-manager criminal liability for non-compliance with information notices
  • Information Commissioner’s Office (ICO) — UK GDPR/Data Protection Act 2018, Age-Appropriate Design Code (Children’s Code), AI auditing framework
  • Competition and Markets Authority (CMA) — Digital Markets Unit (DMU) established 2021, Digital Markets Competition and Consumers Act 2024 entered force January 2025 with Strategic Market Status designations for “very large” digital firms
  • Department for Science, Innovation and Technology (DSIT) — absorbed digital policy from DCMS February 2023; absorbed CDEI functions September 2023
  • UK AI Security Institute (AISI) — renamed from UK AI Safety Institute February 2025, evolved from Foundation Model Taskforce November 2023; pre-deployment evaluations of frontier models (OpenAI, Anthropic, Google DeepMind via April 2024 MoUs)
  • National Cyber Security Centre (NCSC) — joint Guidelines for Secure AI System Development November 2023 with US CISA and 17 nations
  • Investigatory Powers Commissioner’s Office (IPCO) and Investigatory Powers Tribunal (IPT) — oversight of state surveillance
• Civil-society research and advocacy (London-centric but increasingly distributed):
  • Ada Lovelace Institute founded 2018 (Nuffield Foundation), director Gavin Freeguard 2024-
  • doteveryone closed June 2020 with research absorbed by ALI
  • Demos Centre for the Analysis of Social Media (CASM)
  • Centre for Long-Term Resilience (CLTR) founded 2020 focusing on catastrophic risks
  • Open Rights Group (ORG) founded 2005 digital civil liberties
  • Privacy International founded 1990, strategic litigation on surveillance
  • Big Brother Watch civil liberties focus on facial recognition and police data
  • Foxglove Legal strategic litigation on algorithmic injustice (DWP benefit-cap, Uber Eats, A-level algorithm 2020)
  • 5Rights Foundation founded 2015 by Baroness Beeban Kidron, children’s digital rights
  • Center for Countering Digital Hate (CCDH) founded 2018, dual London/Washington base
  • Internet Society UK Chapter, AWO law firm on digital rights
• Northern English academic and industrial hubs:
  • Manchester — Alan Turing Institute partner, ARM (semiconductor IP), Co-op Group digital ethics, BBC R&D MediaCity, IPPR North policy
  • Leeds — Channel 4 headquarters since 2019, AHRC Centre for Cultural Value, Bauman Institute, Leeds City Council Smart Cities Catapult site
  • Sheffield — Department of Computer Science, Speech and Hearing Research Group, Sheffield Digital cluster, AMRC manufacturing AI
  • Newcastle — Open Lab HCI, National Innovation Centre for Data, Helix urban quarter
  • Liverpool — Liverpool Centre for Advanced Internet Studies, Knowledge Quarter
  • York — Department of Computer Science, Centre for Modern Studies (digital humanities and society)
  • Durham — Centre for the History of Medicine and Disease (digital-health-ethics work), Institute of Advanced Study
  • Lancaster — Data Science Institute, Centre for Mobilities Research, Imagination Lancaster

Future Directions (2026-2030)

• Generative AI harm research consolidation: Expansion of MIT AI Risk Repository to v2-v3 with deeper incident-database linkage (AIID Partnership on AI, OECD AI Incident Monitor AIM). Harmonisation of NIST AI RMF, ISO/IEC 42001, and ISO/IEC 23894:2023 with EU AI Act conformity-assessment requirements. Standard development of harm benchmarks (HELM, SafetyBench, ALERT, BBQ extensions) and red-teaming protocols.
• Child-safety regime maturation 2026-2028: Online Safety Act categorised-service duties fully operational from 2026 with first transparency reports and audits. Age-assurance market consolidation. Continued legal challenge to encryption provisions; Apple, Signal, and Meta positions. EU Child Sexual Abuse Regulation (CSAR) trilogue resolution expected 2026-2027 with detection-order architecture.
• AI in workplace surveillance: EU AI Act Article 5 emotion-recognition prohibition applicable from February 2025 expected to drive market shift to “wellbeing analytics” framings. UK Worker Protection (Amendment of Equality Act 2010) Act 2023 implications for algorithmic management. PRO Act US re-introduction prospects 2027.
• Environmental accountability: Compute-emissions reporting standards (CSRD scope 3, SEC climate disclosure) covering data-centre and training workloads. UK Energy Security and Net Zero Department review of AI compute-grid impact 2026-2027. Hyperscaler “AI factory” siting controversies (Microsoft Three Mile Island restart, AWS Talen nuclear PPA, Google fusion contracts).
• Global South labour standards: Continuation of Daniel Motaung Kenya case to substantive trial 2026. ILO Convention on Decent Work in the Digital Platform Economy (under discussion 2024-2026). Possible class-action expansion to Filipino and Latin American moderation workforces.
• Foundation-model harm attribution: Methodological advances on model-collapse (Shumailov et al. Nature 2024), data-poisoning attack practicality, prompt-injection at scale, and agent-harm enumeration. Pre-deployment evaluation standards from AISI, US AISI, EU AI Office, Singapore AIVF, Japan AISI converging 2026-2028.
• Democratic harm-mitigation: Constitutional AI, deliberative AI assistants, jury-based algorithmic-governance experiments (Citizens’ Assembly on Algorithms UK 2025), platform-design alternatives (BlueSky’s “stackable moderation”, Fediverse governance variations).

Research & Literature

• Harm taxonomies and frameworks:
  1. Slattery, P., Saeri, A.K., Grundy, E.A.C., Graham, J., Noetel, M., Uuk, R., Dao, J., Pour, S., Casper, S., & Thompson, N. (2024). The AI Risk Repository: A Comprehensive Meta-Review, Database, and Taxonomy of Risks From Artificial Intelligence. arXiv:2408.12622. MIT FutureTech and University of Queensland. 700+ unique risks across 43 source taxonomies.
  2. Ada Lovelace Institute (2022). Algorithmic Accountability for the Public Sector. Joint report with AI Now Institute and Open Government Partnership.
  3. Ada Lovelace Institute (2022). Looking before we leap: Expanding ethical review processes for AI and data science research.
  4. Shelby, R., Rismani, S., Henne, K., Moon, A., et al. (2023). Sociotechnical Harms of Algorithmic Systems: Scoping a Taxonomy for Harm Reduction. AAAI/ACM AIES 2023.
  5. Weidinger, L., Mellor, J., Rauh, M., et al. (DeepMind, 2021). Ethical and social risks of harm from Language Models. arXiv:2112.04359.
  6. Weidinger, L., et al. (2023). Sociotechnical Safety Evaluation of Generative AI Systems. arXiv:2310.11986.
  7. Bender, E.M., Gebru, T., McMillan-Major, A., Shmitchell, S. (2021). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? FAccT 2021. DOI: 10.1145/3442188.3445922.
• UK regulatory and policy sources: 8. Online Safety Act 2023 (c.50). UK Public General Acts. https://www.legislation.gov.uk/ukpga/2023/50 9. Ofcom (December 2024). Statement: Protecting people from illegal harms online — Codes of Practice. 10. Ofcom (May 2025). Statement: Protecting children from harms online — Codes of Practice and Guidance. 11. Information Commissioner’s Office (2020, in force 2 September 2021). Age Appropriate Design: A Code of Practice for Online Services. 12. UK Government (March 2023). A pro-innovation approach to AI regulation (White Paper, CP 815). 13. Department for Science, Innovation and Technology (February 2024). A pro-innovation approach to AI regulation: government response.
• EU and international frameworks: 14. Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). 15. Regulation (EU) 2022/2065 (Digital Services Act). 16. Council of Europe Framework Convention on Artificial Intelligence, Human Rights, Democracy and the Rule of Law (CETS No. 225, opened for signature 5 September 2024). 17. NIST (January 2023). AI Risk Management Framework 1.0 (NIST AI 100-1). 18. NIST (July 2024). Generative AI Profile (NIST AI 600-1). 19. ISO/IEC 42001:2023 — Information technology — Artificial intelligence — Management system.
• Foundational scholarship: 20. Zuboff, S. (2019). The Age of Surveillance Capitalism. PublicAffairs. 21. O’Neil, C. (2016). Weapons of Math Destruction. Crown. 22. Noble, S.U. (2018). Algorithms of Oppression: How Search Engines Reinforce Racism. NYU Press. 23. Eubanks, V. (2018). Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor. St. Martin’s. 24. Crawford, K. (2021). Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale UP. 25. Benjamin, R. (2019). Race After Technology: Abolitionist Tools for the New Jim Code. Polity.
• Empirical case studies: 26. Buolamwini, J., & Gebru, T. (2018). Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification. FAT* 2018, PMLR 81:77-91. 27. Senior Coroner Andrew Walker (30 September 2022). Regulation 28 Report to Prevent Future Deaths: Molly Russell. North London Coroner’s Court. 28. Internet Watch Foundation (April 2025). Annual Report 2024. https://www.iwf.org.uk 29. Perrigo, B. (18 January 2023). OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic. TIME. 30. The Wall Street Journal (September-October 2021). The Facebook Files series. 31. The Markup (October 2023). Crime Prediction Software Promised to Be Free of Biases. New Data Shows It Perpetuates Them (PredPol/Geolitica investigation).

Metadata

• Last Updated: 2026-05-16
• Review Status: Phase 6 enrichment — comprehensive editorial review
• Verification: Primary sources cited (Online Safety Act 2023 c.50, EU AI Act 2024/1689, Council of Europe AI Treaty CETS No. 225, MIT AI Risk Repository arXiv:2408.12622, NIST AI RMF 100-1, ISO/IEC 42001:2023, IWF Annual Reports, coroner reports, court filings); incident case references cross-checked against contemporaneous press coverage and court records
• Domain Correction: infrastructure → society. The original frontmatter placed this concept in the infrastructure domain alongside data-centres and networking topology; the concept is properly a society-domain catalogue of socio-technical harms. IRI, URI, same-as, and owl-class updated accordingly. Documented also in Provenance.
• Regional Context: UK regulators (Ofcom, ICO, CMA/DMU, DSIT, AI Security Institute), London civil society (Ada Lovelace, ORG, Privacy International, 5Rights, CCDH, Foxglove, Big Brother Watch), Northern English academic and industrial hubs (Manchester, Leeds, Sheffield, Newcastle, Liverpool, York, Durham, Lancaster), Oxford OII, Cambridge LCFI/Minderoo, UCL, Imperial, Edinburgh CTMF, KCL, LSE detailed
• Production-Ready: Complete OWL formal semantics; comprehensive content coverage (taxonomic landscape, eight harm categories with documented incidents, academic context, current 2026 regulatory/litigation status, UK context, 2026-2030 future directions, 31 academic/regulatory/empirical references); contrast surface against techno-utopianism, individual-choice and market-only framings made explicit
• Authority Score: 0.87 (foundational legal-regulatory citations; canonical incident references; active regulatory implementation 2025-2026; established civil-society and academic research base)

Provenance

• domain-correction: infrastructure → society (concept is a society-domain harm catalogue, not infrastructure; IRI/URI/same-as/owl-class updated)