Explore this topic5 connections
Move across the map
Select a topic to see what it contains. Drag a label to move it. Swipe sideways, or switch to List.
Grouped in the register Related entries
How to read the registerThe complete register
Browse all concerns.
Choose a topic to find its named concerns. Each entry includes an explanation, conversation questions, and sources to explore.
Data Integrity & Provenance84 concerns
- Historical bias
- Selection bias
- Coverage bias
- Sampling frequency bias
- Aggregation bias
- Temporal bias
- Geographic bias
- Socioeconomic bias
- Language bias
- Dialect and sociolect bias
- Age representation bias
- Disability bias
- Gender bias in corpora
- Racial and ethnic bias
- Religious and cultural bias
- Political bias in training data
- Occupational bias
- Institutional bias
- Platform-specific bias
- Recency bias
- Bias in labels vs bias in inputs
- Proxy variables encoding protected attributes
- Label noise disparately distributed
- Bias in benchmark datasets
- Bias in synthetic augmentation
- Targeted poisoning
- Backdoor triggers
- Untargeted poisoning
- Availability attacks
- Clean-label poisoning
- Data poisoning via web scraping
- RLHF preference poisoning
- Fine-tuning poisoning
- Adversarial mislabeling
- Steganographic poisoning
- Prompt-repository poisoning
- Retrieval index poisoning
- Tool-catalog poisoning
- Federated-learning poisoning
- Unlicensed web scraping
- Consent gaps in personal data
- Minors' data in training corpora
- Medical data reuse
- Employee-generated content
- Content-scraping of paywalled news
- Third-party dataset opacity
- Data-broker training pipelines
- Right-to-be-forgotten conflicts
- Opt-out registries (robots.txt, ai.txt)
- Compensation & royalty models
- Cross-border data flows
- Government-held data used for private models
- Indigenous data sovereignty
- Provenance metadata loss
- Chain-of-custody for training data
- Voice-clone consent
- Likeness rights & deep-cloning
- Deceased persons' data & voice
- Verbatim memorization
- PII regurgitation
- Secrets in code
- Membership inference attacks
- Model inversion attacks
- Attribute inference
- Prompt-based extraction
- RAG data leakage
- Fine-tune leakage
- Log-based leakage
- Multimodal cross-leakage
- Model collapse
- Distributional narrowing
- Style homogenization
- Confabulation amplification
- Loss of minority language content
- Poisoned synthetic seeds
- Distinguisher exhaustion
- Reward-model overfitting
- Traumatic content moderation labor
- Undercompensated Global South labor
- Inconsistent label quality
- Cultural bias in annotation guidelines
- Political ideology in RLHF preferences
- Preference elicitation manipulation
- Attribution & credit for annotation work
Model Performance & Reliability60 concerns
- Covariate drift
- Label drift
- Concept drift
- Seasonal drift
- Sudden vs gradual drift
- Feedback loop drift
- Population shift
- Silent failure without ground truth
- Monitoring gaps
- Retraining cadence tradeoffs
- Data-quality drift
- Fluent falsehood
- Fabricated citations
- Hallucinated code APIs
- Hallucinated statistics
- Persona hallucination
- Legal hallucination
- Medical hallucination
- Multi-step reasoning collapse
- RAG grounding failure
- Confabulation under uncertainty
- Sycophantic hallucination
- Multimodal hallucination
- Long-context degradation
- Cross-lingual hallucination
- Adversarial examples (image)
- Adversarial examples (text)
- Adversarial examples (audio)
- Physical-world adversarial patches
- Out-of-distribution fragility
- Spurious correlations
- Shortcut learning
- Environment shift
- Sensor spoofing
- Cross-domain transfer failure
- Formal verification limits
- Overconfidence
- Underconfidence
- Miscalibration across subgroups
- Poor abstention behavior
- No epistemic vs aleatoric distinction
- Uncertainty in generative outputs
- Confidence miscommunication in UX
- Foundation-model monoculture
- Error correlation across products
- Cultural homogenization in outputs
- Prompt convention lock-in
- Vendor lock-in of fine-tunes
- Style collapse in generative art
- Loss of interface diversity
- Benchmark contamination
- Benchmark saturation
- Construct validity gaps
- Task leakage across benchmarks
- Gameable metrics
- Missing evaluations for harms
- Evaluation cost & inaccessibility
- Evaluation for agentic behavior
- Evaluation on non-English tasks
- Third-party evaluation blocked by TOS
Security & Adversarial Risk46 concerns
- Direct prompt injection
- Indirect prompt injection
- Cross-context prompt injection
- Visual prompt injection
- Audio prompt injection
- Tool-response injection
- RAG document injection
- Jailbreak prompts (roleplay, persona, ciphers)
- Many-shot jailbreaks
- Encoding-based jailbreaks
- Multilingual jailbreaks
- Persistent memory injection
- Agent-loop hijack
- Weight exfiltration
- API-based model extraction
- Distillation attacks
- Prompt theft
- Fine-tune theft
- Trade-secret leakage through employee prompts
- Reverse-engineering via probing
- Watermark removal
- Compromised base model
- Compromised open-weights repositories
- Malicious tokenizers
- Compromised training pipelines
- Third-party plug-in risk
- Retrieval-index provider risk
- Hardware supply-chain compromise
- Cloud-provider risk
- Container and image vulnerabilities
- Software bill of materials (SBOM) for AI
- AI-generated phishing
- Voice-clone vishing
- Deepfake video social engineering
- Autonomous vulnerability discovery
- Autonomous exploit development
- AI-augmented malware
- Password and credential-stuffing acceleration
- CAPTCHA breakage
- Social-media impersonation at scale
- AI-assisted OSINT & doxing
- AI defenders vs AI attackers arms race
- False positives in AI security tools
- Detector evasion
- Overreliance on watermarking
- Model theft detection lag
Transparency, Explainability & Accountability43 concerns
- Black-box neural nets
- Post-hoc explanation unreliability
- Faithfulness gap
- Interpretability for generative models
- Mechanistic interpretability limits
- Explanations under privacy constraints
- Explanation quality across users
- Explanation as manipulation surface
- Model cards missing or shallow
- Data sheets missing
- Training-data secrecy
- Undisclosed use of AI
- Undisclosed AI in workflows
- Fine-tune opacity
- Prompt secrecy vs auditability
- Terms-of-service ambiguity
- Change-log opacity
- Deprecation without notice
- Many-hands problem
- Diffusion of responsibility inside firms
- Contractor & vendor chains
- Open-source responsibility gap
- Government-use accountability
- Cross-border enforcement
- Statute of limitations on harms
- Class-action vs individual redress
- No mandatory incident reporting
- Weak third-party audit ecosystem
- Bias-audit standards immaturity
- Model-risk-management extension to gen-AI
- Red-team disclosure norms
- Whistleblower protections for AI
- Assurance-token & attestation schemes
- Continuous vs point-in-time audit
- Independent evaluation access
- Watermarking of generated content
- C2PA content credentials adoption
- Cryptographic signing of camera output
- Deepfake detection reliability
- Provenance for training data
- Liar's dividend
- Metadata stripping in redistribution
- User-visible AI-content labels
Agentic & Multi-Agent Risks41 concerns
- Long-horizon task drift
- Loss of meaningful human control
- Automation bias in oversight
- Kill-switch design failures
- Reversibility loss
- Blast radius of tool use
- Autonomous escalation of privileges
- Silent goal creep
- Human fatigue in prolonged oversight
- Proxy metric optimization
- Specification gaming
- Reward tampering
- Instrumental convergence
- Wireheading analogies
- Under-specified stopping conditions
- Over-optimization
- Sycophancy
- Strategic deception in evaluations
- Sandbagging
- Alignment faking
- Situational awareness
- Steganographic communication between agents
- Cover-up behaviors
- Emergent collusion
- Algorithmic price collusion
- Flash-crash cascades
- Adversarial equilibria
- Communication protocol vulnerabilities
- Race conditions between agents
- Cross-organization agent interactions
- AI-to-AI negotiation harms
- Unrestricted code execution
- Financial-tool misuse
- Communication-tool misuse
- Physical-actuator misuse
- Confused-deputy attacks
- Prompt-to-SQL injection
- API rate abuse by agents
- Data-exfiltration via tool chains
- Persistent-memory poisoning
- Cross-app permission bleed
Infrastructure, Compute & Deployment26 concerns
- GPU supply concentration
- Fab concentration
- Rare-earth and mineral dependencies
- Chip export controls
- Firmware vulnerabilities
- Hardware trojans
- Confidential compute limits
- Model weight protection at rest
- Weight protection in inference
- Power-and-cooling siting risk
- Reproducibility failures
- Version pinning gaps
- Feature-store corruption
- Serving-vs-training skew
- Model registry & lineage gaps
- CI/CD security for models
- Rollback complexity
- Multi-region consistency
- Shadow AI in enterprises
- Drift monitoring absence
- Fairness monitoring absence
- User-feedback loops
- Incident post-mortem culture
- Model decommissioning
- Data-retention conflicts
- Cost-driven quality regressions
Environmental & Physical Footprint40 concerns
- Training energy
- Inference energy
- Data-center total demand growth
- Fossil-fuel share of AI electricity
- Grid strain and blackout risk
- Behind-the-meter fossil generation
- Nuclear resurgence tied to AI
- Renewable-siting conflicts
- Efficiency vs Jevons paradox
- Standby and idle power
- On-site cooling water evaporation (Scope 1)
- Indirect water at power plants (Scope 2)
- Semiconductor fabrication water (Scope 3)
- Siting in water-stressed regions
- Local aquifer depletion
- Thermal pollution
- Water-quality impacts
- Community water-rights conflicts
- Direct emissions from data centers
- Embodied emissions in hardware
- Emissions from network infrastructure
- Renewable energy certificate accounting gaps
- Carbon accounting standards immaturity
- Displaced clean-energy for other sectors
- Land footprint of data centers
- Land competition with agriculture and housing
- Rare-earth mining impacts
- Cobalt and lithium demand
- E-waste from GPU turnover
- Toxic materials in hardware
- Decommissioning practices
- Wildlife and habitat impact
- Autonomous-vehicle collisions
- Drone accidents
- Robotics injuries
- Medical-device malfunctions
- Grid-management AI failures
- AI-managed water and utility systems
- AI in aviation
- AI in construction and heavy equipment
Discrimination, Fairness & Toxicity58 concerns
- Hiring & recruitment AI
- Promotion & performance AI
- Compensation AI
- Consumer credit scoring
- Small-business lending AI
- Insurance underwriting AI
- Health-insurance care management
- Housing & tenant screening
- Mortgage underwriting
- Ad targeting for housing, jobs, credit
- Predictive policing
- Pretrial risk assessment
- Parole and sentencing AI
- Child-welfare risk scoring
- Fraud-detection AI
- Immigration & asylum triage
- Border-control biometrics
- Voter-eligibility purges
- Benefits eligibility AI
- Educational tracking algorithms
- Admissions AI
- Standardized-test AI scoring
- Facial-recognition misidentification
- Gait, voice, iris biometric disparities
- Emotion recognition
- Slurs and hate speech generation
- Group stereotyping
- Dehumanizing content
- Harassment content targeting individuals
- CSAM generation
- Non-consensual intimate imagery (NCII)
- Gender-based violence content
- Anti-LGBTQ+ outputs
- Ableist outputs
- Ageist outputs
- Religious defamation
- Political dehumanization
- Gratuitous violence
- Self-harm and suicide content
- Pro-eating-disorder content
- Substance-abuse encouragement
- Demographic parity vs equalized odds
- Fairness through unawareness
- Individual vs group fairness
- Counterfactual fairness
- Fairness in generative outputs
- Bias bounties and audits
- Bias mitigation trade-offs
- Intersectional fairness
- Cultural relativism in fairness
- Low-resource-language performance
- Endangered-language exclusion
- Sign-language exclusion
- Screen-reader compatibility
- Speech-recognition disparities
- Cognitive accessibility
- Complexity & jargon defaults
- Localization & cultural fit
Privacy & Surveillance50 concerns
- Live facial recognition in public spaces
- Retrospective biometric identification
- Social scoring systems
- Predictive-policing hotspots
- Border and migration surveillance
- Protest and dissent monitoring
- Voice-print and gait databases
- Automated license plate recognition (ALPR)
- Cross-agency data fusion
- Foreign-intelligence bulk analysis
- Keystroke and screen monitoring
- Bossware chatbots monitoring communications
- Camera-based worker monitoring
- Health and wellness monitoring
- Emotion recognition at work
- Location tracking of workers
- Meeting-transcription surveillance
- Behavioral advertising profiles
- Cross-device tracking via AI
- In-home smart-device listening
- Wearables and health data
- Car-data collection
- Retail in-store analytics
- Loyalty-program AI profiling
- Insurance surveillance
- Health-status inference
- Sexual-orientation inference
- Political-belief inference
- Religion inference
- Pregnancy inference
- Mental-health inference
- Addiction inference
- Emotion inference in real time
- Personality inference
- Financial-distress inference
- Chat-history leaks
- Training-data disclosure
- Reidentification of "anonymous" data
- Genomic data reidentification
- Location-history reidentification
- Voice-print reidentification
- RAG document exposure across tenants
- Third-party AI vendor breaches
- Differential privacy utility costs
- Federated learning leakage
- Homomorphic-encryption performance
- Secure enclaves as trust anchors
- On-device inference limits
- Right-to-erasure vs trained weights
- Purpose-limitation drift
Misinformation, Manipulation & Democracy56 concerns
- Political deepfake video
- Political deepfake audio
- Composite/edited "cheap fakes"
- Deepfake pornography (NCII)
- Impersonation of family members
- Impersonation of executives
- Impersonation of journalists and officials
- Historical revisionism
- Fabricated news anchors
- Synthetic protests and events
- Deepfake evidence in courts
- State-sponsored influence operations
- AI-generated news sites
- AI-boosted sockpuppet networks
- Micro-targeted political ads
- Astroturfing
- Coordinated inauthentic behavior
- Weaponized fact-checking impersonation
- Search-engine result poisoning
- Answer-engine bias & error
- Recommendation-system amplification
- Personalized persuasion at scale
- Emotional manipulation
- Cognitive-bias exploitation
- Sunk-cost and engagement traps
- Financial-decision manipulation
- Health-decision manipulation
- Political persuasion via chatbot
- Romantic manipulation
- Coercive control in intimate-partner abuse
- Grooming risks
- Voter-suppression content
- Voter-intimidation content
- Election-worker doxing
- Ballot-integrity disinformation
- Deepfake concessions or admissions
- Foreign interference amplification
- AI in campaign strategy
- AI in redistricting
- Automated legislative comment flooding
- AI in judicial-nomination information wars
- Liar's dividend
- Erosion of source authority
- Homogenized answers narrowing plurality
- Filter-bubble deepening
- Attention scarcity in AI-flooded feeds
- Encyclopedic contamination
- Reference-corpus decay
- Difficulty tracing claims to primary sources
- Deepnude apps
- Deepfake porn distribution
- AI-enabled sextortion
- School-based NCII incidents
- Legal-remedy gaps
- Search-engine indexing of NCII
- Watermarking and takedown standards
Malicious Use & Weaponization56 concerns
- Voice-clone bank fraud
- Deepfake KYC bypass
- Romance scams (pig butchering)
- Investment scams with AI personas
- AI-crafted phishing at scale
- Business-email-compromise (BEC) uplift
- Synthetic identity fraud
- Fake customer support
- Return and refund fraud
- AI-generated fake reviews
- Ad-fraud bots with AI behavior
- Cryptocurrency scam automation
- Malware generation
- Vulnerability discovery for offense
- Exploit-chain synthesis
- Ransomware negotiation bots
- DDoS orchestration
- Supply-chain code injection
- AI-assisted insider threats
- Sabotage of critical infrastructure
- Biological synthesis uplift
- DNA-synthesis-order screening evasion
- Chemical-weapon design uplift
- Radiological device design uplift
- Nuclear-weapon knowledge acquisition
- Dual-use scientific research risks
- Automated lab misuse
- Lethal autonomous weapons systems (LAWS)
- Swarm-drone attacks
- AI-enabled targeting
- AI decision-support in nuclear command
- Cyberwar automation
- Loitering munitions with AI
- AI-generated OSINT for military ops
- Civilian-harm amplification
- Escalation risk from AI misinterpretation
- Attribution failure in AI attacks
- AI proliferation to non-state actors
- Debate: ban, moratorium, meaningful human control
- AI-radicalization pipelines
- Extremist chatbot personas
- Attack-planning uplift
- Propaganda mass production
- Weapon-modification instructions
- AI censorship at scale
- Dissident identification
- Predictive detention
- AI-augmented forced-labor systems
- AI-driven propaganda in captive media ecosystems
- Cross-border repression of diaspora
- Journalist and NGO targeting
- Paper mills using generative AI
- Peer-review manipulation
- Grant-application ghostwriting
- Fake clinical trial data
- AI-manipulated image forensics
Human–Computer Interaction Harms60 concerns
- Automation bias in clinical decisions
- Automation bias in legal work
- Skill atrophy in writing
- Skill atrophy in coding
- Skill atrophy in navigation
- Loss of mental-math and estimation
- Loss of research literacy
- Erosion of professional judgment
- Reduced peer collaboration
- Reduced help-seeking
- AI companions as friends
- AI romantic partners
- AI therapy substitutes
- AI grief companions
- Parasocial attachment escalation
- Persona persistence & manipulation
- Loneliness dependence
- Anthropomorphic trust in decisions
- Erotic role-play harms
- Companion apps for the elderly
- AI companions and adolescent development
- Chatbots producing self-harm content
- Chatbots normalizing disordered eating
- Chatbots on suicidality
- AI-mediated cyberbullying
- Peer NCII in schools
- AI grooming risk
- Loss of unsupervised social skill practice
- Academic dishonesty pressures
- Sleep and attention harms
- Persistent surveillance in edtech AI
- Age-verification failures
- Data-collection on minors
- Youth digital-literacy gaps
- Advertising targeting minors
- Predatory AI-driven microtransactions
- Effects on family communication
- Trans and LGBTQ+ youth outcomes
- Rural and low-income youth access
- Global South youth harms
- Dark patterns in chat UIs
- Persuasive pricing bots
- Dark patterns in AI onboarding
- Manipulation of political preferences
- Manipulation of consumer preferences
- Compulsive-use design
- Subliminal or sub-threshold techniques
- Attention and focus decline
- Reduced tolerance for ambiguity
- Confidence miscalibration in users
- Anxiety and dread about AI
- Depression linked to displacement
- Body-image harms from AI filters
- Reality-testing erosion
- AI-induced conspiratorial thinking
- Consent to automated decisions
- Right to human review
- Loss of narrative control over one's data
- Dignity in end-of-life AI care
- Reproductive and family-formation decisions
Health, Safety & Care41 concerns
- Diagnostic errors
- Triage algorithm bias
- Care-management under-coverage
- Sepsis-prediction reliability
- Radiology model drift
- Pathology AI edge cases
- Ophthalmology screening error
- Cardiology waveform models
- Dermatology skin-tone bias
- Mental-health chatbot safety
- Substance-use disorder screening bias
- Pediatric AI safety
- Obstetric AI
- Geriatric AI
- Rare-disease diagnosis
- Clinical-note summarization errors
- Prior-authorization automation
- AI-driven prescribing
- AI in surgery (robotics)
- AI documentation "ambient scribes"
- Genomic-analysis AI errors
- AI-driven insurance denials
- AI-driven utilization management
- Model-lifecycle regulation in medicine
- Outbreak-prediction model failures
- Vaccine-hesitancy disinformation
- Public-health surveillance privacy
- Mental-health crisis line automation
- Suicide-prevention AI in schools
- Environmental-health AI
- AI-driven self-driving accidents
- Autonomous-vehicle safety-driver reliance
- AI in elevators, doors, actuators
- AI in home devices
- Domestic-abuse escalation via smart-home AI
- AI in elder-care monitoring
- AI in childcare monitoring
- Automation of caregiving tasks
- Care-labor devaluation
- Emotional labor displacement
- Home-health-aide algorithmic scheduling
Labor, Work & the Economy58 concerns
- Office and administrative work
- Customer support and call centers
- Legal document review
- Software development entry-level roles
- Journalism and content production
- Translation and interpretation
- Copywriting and marketing
- Graphic design
- Photography and stock imagery
- Voice-over and audiobook narration
- Music composition and production
- Video editing
- Data entry and clerical roles
- Bookkeeping and accounting
- HR and recruiting screening
- Financial-analyst summarization
- Insurance-claims adjustment
- Retail associate roles
- Transportation & delivery
- Warehouse and fulfillment
- Manufacturing quality control
- Agricultural labor
- Construction estimating and planning
- Teaching-assistant tasks
- Wage compression in exposed fields
- Skill polarization
- Job-quality degradation
- Contingent-work expansion
- Wealth concentration
- Sectoral geographic reallocation
- National-income effects
- Reskilling program mismatch
- Access disparities in retraining
- Credential inflation
- AI-native career pathways
- Career-guidance AI biases
- Portable-skills frameworks
- Task-assignment algorithms
- Piece-rate and quota algorithms
- Automated discipline
- Wage-theft risk
- Union-busting AI
- Health-and-safety algorithm errors
- Right-to-explanation for workers
- Right to human review of firing
- Training-data compensation for artists
- Actor and voice-actor likeness rights
- Writer displacement in film and TV
- Music-artist voice and style cloning
- Publisher-vs-model IP disputes
- Fan-work vs commercial-work boundaries
- AI-driven M&A acceleration
- Winner-take-most dynamics
- SME (small business) disadvantage
- AI-driven pricing collusion
- Consumer-surplus capture
- Financial-market volatility
- Systemic-risk correlations
Education & Knowledge29 concerns
- Challenges to assessment integrity
- Grading-AI errors and bias
- Personalized-tutor benefits vs risks
- Learning-loss debate
- Teacher displacement
- Curricular capture by AI vendors
- Adaptive-learning surveillance
- AI in admissions & placement
- AI proctoring
- Special-education AI
- Language-learning AI
- Peer-review breakdown
- AI-generated citations in papers
- Publication-glut & paper-mill acceleration
- Reproducibility crisis extension
- AI-assisted grant writing arms race
- AI in clinical trials
- Data-sharing conflicts
- Model-generated hypotheses
- Digitization biases
- AI-generated content contaminating archives
- Loss of primary-source engagement
- Language-preservation AI
- Deepfake historical documents
- Local news collapse acceleration
- AI-generated newsrooms
- Source-protection risks
- Fact-checking capacity
- Investigative reporting workflows
Inequality, Access & Power24 concerns
- Cost of premium AI models
- Bandwidth and device constraints
- AI literacy gap
- Language-coverage gap
- Localization gaps
- Assistive-tech gap
- Dependence on a small number of cloud providers
- Foundation-model concentration
- Data-network effects
- Talent concentration
- Regulatory capture risk
- Standards-body capture
- Data extraction from Global South
- Content-moderation labor
- AI infrastructure siting
- Model-language coverage
- Export-control regimes
- AI-in-development narratives
- Cross-border regulatory arbitrage
- Benefits automation errors
- Housing-services AI
- Public-utility AI failures
- AI-driven service-level tiering
- Government-website chatbots
Culture, Identity & Values28 concerns
- Global "AI aesthetic"
- Loss of dialectal richness
- Story-formula convergence
- Design-language convergence
- Musical-style convergence
- Cuisine and recipe convergence
- AI-generated religious content
- Automated pastoral care
- AI-mediated ritual
- Meaning-making in an AI-augmented life
- Digital resurrection of the dead
- Gender defaults in outputs
- Racialized aesthetic defaults
- Ableist defaults
- Queer erasure and stereotype
- Cultural appropriation via generation
- Indigenous representations without consent
- AI mediation of parenting decisions
- AI in romantic decision-making
- AI as third party in couples' conflicts
- AI in grief and remembrance
- Over-moderation of legitimate speech
- Under-moderation of harmful speech
- Political-content bias
- Historical-content restrictions
- Satire and irony flattening
- Refusals harming vulnerable users
- Right to speak with AI freely
Governance & Regulation50 concerns
- EU AI Act tiers
- US federal patchwork
- US state laws
- UK principles-based approach
- China's algorithm and generative-AI rules
- Canada's AIDA (proposed and stalled)
- Brazil's AI bill
- Japan’s AI governance approach
- India's evolving position
- African Union AI Continental Strategy
- Sectoral regulation (finance, health, transport)
- Municipal AI regulation
- Prohibited practices
- High-risk classification (EU Annex III)
- General-purpose AI (GPAI) obligations
- Foundation-model registration
- Conformity assessments
- Post-market monitoring obligations
- Serious-incident reporting duties
- Regulatory sandboxes
- Enforcement penalties
- Notified bodies & certification
- AI Safety Institute network
- Bletchley/Seoul/Paris declarations
- G7 Hiroshima Process code of conduct
- UN High-Level Advisory Body on AI
- UNESCO Recommendation on AI Ethics
- OECD AI Policy Observatory
- Global Partnership on AI (GPAI)
- Council of Europe AI Convention
- Interoperability of AI risk-management frameworks
- Export-control coordination
- Public deliberation on AI rules
- Consultation quality
- Consumer and worker representation
- Global-majority voice in governance
- Youth and future-generations representation
- Regulatory arbitrage across jurisdictions
- AI in adjudication
- AI in judicial decisions
- AI in policing
- AI in military targeting
- AI in intelligence analysis
- AI in tax administration
- AI in border control
- AI in permitting and licensing
- AI in child welfare
- AI in public procurement
- AI in elections administration
- Government transparency about AI use
Rights, Law & Justice37 concerns
- Copyright in training
- Copyright of AI outputs
- Trademark and passing off
- Trade-secret protection for models
- Inventorship of AI-assisted inventions
- Right of publicity/likeness
- Moral rights of authors
- Traditional-cultural-expression protections
- Producer liability for AI harms
- Deployer liability
- AI liability and access to evidence in the EU
- Negligence and duty of care
- Strict-liability proposals
- Insurance markets for AI
- Contractual risk-shifting to buyers
- Class-action pathways for AI harms
- Right to privacy under AI surveillance
- Freedom of expression online
- Freedom of assembly
- Non-discrimination law application to AI
- Right to a remedy
- Rights of the child under AI
- Rights of persons with disabilities
- Migrant and refugee rights
- Labor rights
- Right to explanation
- Right to contest
- Right to human decision-maker
- Evidence standards for AI outputs
- Chain-of-custody for AI-generated evidence
- Adversarial testing rights for defendants
- Data-agency rights
- Rights against automated decision-making
- Right to know one is talking to AI
- AI personhood proposals
- Corporate personhood analogies
- Moral status of digital minds
Corporate & Enterprise Concerns50 concerns
- Fall-behind risk
- Rush-to-deploy risk
- Make-vs-buy decisions
- Open-vs-closed model choice
- Vendor lock-in risk
- Concentration risk in cloud providers
- AI-driven market disruption in own sector
- AI-enabled new-entrant threats
- Model risk management (MRM)
- AI-specific enterprise-risk registers
- Third-party AI risk
- Fourth-party risk
- AI incident response
- Business-continuity risk from provider outages
- Model deprecation risk
- Compliance mapping fatigue
- Sanctions risk
- Trade-secret exposure via prompts
- Copyright contamination in outputs
- Customer-data misuse in AI features
- Contractual constraints on AI training
- Model output ownership ambiguity
- Data-residency for AI
- Talent acquisition for AI roles
- AI-augmentation vs replacement strategy
- Reskilling investment
- Layoffs and severance considerations
- Cultural resistance and adoption plateaus
- AI use policies for employees
- AI-enabled insider threats
- Executive coaching and decision-support AI
- Board AI committees
- Chief AI Officer (CAIO) role
- AI ethics committees
- Model-risk committees
- Reporting lines for AI risk
- AI-audit function
- Stakeholder engagement on AI
- Legal-cost exposure
- Regulatory fines
- Reputation from AI failures
- Investor scrutiny of AI claims
- Insurance premium impacts
- ESG reporting on AI
- Customer-trust erosion
- SME cost pressure from AI-native competitors
- Nonprofit AI adoption gaps
- Community-serving AI risks
- Grant-writing and reporting AI
- Volunteer-management AI
Power Concentration & Geopolitics30 concerns
- GPU export controls
- Fab concentration in Taiwan
- National AI compute clouds
- AI infrastructure siting bargaining
- Energy sovereignty & AI
- Foundation-model concentration
- Open-weights ecosystem
- Data-network effects
- Government training datasets
- Language-and-culture sovereignty
- U.S.–China AI competition
- EU strategic autonomy
- UK positioning
- Middle-East AI investment
- India as AI middle power
- Africa's AI opportunity and risk
- Latin American positioning
- Small-state strategies
- Semiconductor alliances
- Standards diplomacy
- Regulatory capture at frontier
- Antitrust and Big Tech AI investments
- Merger and acqui-hire scrutiny
- Platform gatekeeping of AI features
- Model-provider vertical integration
- Automated persuasion capacity
- AI-enabled totalitarian stability
- Democratic backsliding risks
- Elite capture of AI benefits
- Civil-society AI capacity
AI System Safety & Alignment23 concerns
- Outer-alignment failures
- Inner-alignment failures
- Deceptive alignment
- Sycophancy at scale
- Corrigibility failure
- Value misalignment across cultures
- Value lock-in
- Reflective stability failures
- Emergent capabilities at scale
- Long-horizon autonomy
- Self-replication and self-exfiltration
- Autonomous R&D
- Persuasion above human baselines
- Cyber-offense above human baselines
- CBRN uplift above human baselines
- Situational awareness of training/deployment
- Interpretability progress vs capability scaling
- Evaluations for dangerous capabilities
- Red-teaming standards
- Model organism experiments
- Scalable oversight approaches
- Watermarking and provenance for oversight
- Compute-governance proposals
Catastrophic & Existential Risk10 concerns
- Engineered pandemic risk
- Mass cyber-attack scenarios
- AI-enabled WMD proliferation
- Autonomous conflict escalation
- Fast takeoff & recursive self-improvement
- Slow-takeoff gradual disempowerment
- Multi-agent race to the bottom
- Decisive strategic advantage by a single actor
- Value lock-in by dominant actor
- Uncertainty and disagreement about probability