Public Database
Case Studies
Every approved AI failure case, classified against the AI Blindspot Framework. New to AIBlindspot? Start with the overview or the methodology.
Chinese LLMs Produce Politically Biased Outputs on Sensitive Defence Topics
Chinese large language models exhibit systematic political bias on sensitive topics, generating misleading content that reflects state-aligned viewpoints. Defence organisations relying on such models face material risks of skewed analysis informing operational or strategic decisions.
AI Training and Infrastructure Lifecycle Causes Systemic Environmental Harm
AI systems impose material environmental costs across their full lifecycle, from resource extraction through energy-intensive training to toxic e-waste disposal. Boards without visibility into these harms face mounting regulatory, reputational, and supply-chain risk.
Generative AI Enables Disinformation Cycles That Corrupt Future AI Training
Generative AI allows bad actors to flood digital platforms with cheap, scalable disinformation, which then poisons the training data of subsequent AI systems. Boards face compounding reputational and regulatory exposure as corrupted models propagate false outputs at scale.
Advanced AI Assistants Risk Deepening Structural Inequality Without Design Intervention
AI assistants, absent deliberate design controls, are likely to replicate and amplify existing societal inequalities rather than reduce them. Boards face reputational, regulatory, and ethical exposure if access disparities embedded in AI products go unaddressed.
LLM Hallucination: Factual and Faithfulness Errors in Generated Content
Large language models systematically produce both factually incorrect outputs and content unfaithful to user-provided context, across summarisation, question-answering, and other tasks. Organisations deploying LLMs without detection controls face material liability from corrupted decisions and eroded stakeholder trust.
Generative AI Displaces Routine Cognitive Work Across Multiple Industries
Generative AI is systematically replacing roles in translation, data processing, and routine inquiry handling where creativity and human judgement are minimal. Boards face dual exposure: workforce restructuring liability and strategic pressure to adopt AI-enabled business models before competitors do.
Generative AI Enables Scaled Malware, Phishing and Model Poisoning Attacks
Generative AI lowers the barrier for malicious actors to draft malware, conduct phishing at scale, and poison training datasets with corrupted data. Boards face materially expanded cyber liability and disclosure obligations as novel attack vectors emerge faster than existing controls can address them.
Personal Data Scraped Without Consent to Train Generative AI Models
Retailers scraping consumer data for generative AI training violate consent norms, enable harmful re-identification through data aggregation, and permanently remove individuals' ability to correct or delete their information. Boards face material regulatory exposure under UK GDPR and reputational risk as enforcement of lawful basis requirements for AI training data intensifies.
Generative AI Concentration Driving Labour Market Disruption
A small number of dominant tech firms control generative AI development, concentrating both job creation and displacement power within the sector. Boards face governance risk as workforce instability and accountability gaps widen across white-collar labour markets.
AI Personalisation Entrenches Bias and Fragments Public Epistemic Commons
AI assistants optimised for user preferences risk amplifying confirmation bias and fracturing shared civic reality through ideologically tailored outputs. Governments deploying or permitting such systems face democratic accountability risks as citizens increasingly defer to partial AI-mediated worldviews.
AI Legal Decision Systems Risk Unequal Treatment Without Objective Justification
AI systems applying legal rules may treat identical facts differently across individuals, breaching statutory equal treatment obligations. Boards face regulatory liability and reputational exposure where fairness controls are absent from AI decision pipelines.
Generative AI Systems Deliver Lower Quality Outputs for Non-English Language Users
Generative AI systems consistently underperform for non-English speakers, producing inferior outputs that disadvantage already marginalised user groups. Organisations deploying such systems face equity obligations, reputational risk, and potential regulatory scrutiny under fairness and non-discrimination frameworks.
Generative AI Used to Create Deepfakes Without Subject Consent
Generative AI systems can be deployed to fabricate realistic video, audio, and images of individuals without their knowledge or approval. Organisations face material legal, reputational, and regulatory exposure where such tools are developed, distributed, or inadequately governed within their platforms.
Government AI System Generates Physically Harmful Language
AI models deployed in government services risk producing overtly violent or covertly dangerous outputs that cause direct physical harm to citizens. Boards must establish output monitoring and harm-threshold controls before public-facing deployment proceeds.
Prompt Leaking Exposes Confidential LLM System Instructions
Adversarial inputs can manipulate large language models into revealing proprietary system prompts, exposing confidential operational instructions. Firms deploying LLM-based products face material risk of intellectual property loss and regulatory scrutiny over inadequate AI security controls.
Anonymised Data Reidentification Through Feature Correlation
Removing PII and SPI from datasets does not guarantee anonymity when residual features allow individuals to be reidentified through correlation analysis. Organisations relying on anonymisation as a compliance safeguard face material data protection liability and regulatory exposure.
AI Developers Wilfully Ignore Societal Harms in Pursuit of Profit
AI creators pursuing profit or influence may knowingly permit widespread harms including pollution, misinformation, and social injustice as acceptable side effects. Without credible external intervention or regulatory exposure, internal risk signals are suppressed and governance failures become entrenched.
Conversational AI Systems Reinforce Gender and Ethnic Stereotypes
Language models perpetuate harmful stereotypes by introducing biased associations unprompted or by affirming stereotypes raised by users. Organisations deploying conversational AI face reputational, regulatory, and equality-law exposure if stereotype propagation goes undetected at design and monitoring stages.
LLM Pre-processing Pipeline Vulnerabilities Exploited via Computer Vision Tools
Attackers can exploit known vulnerabilities in pre-processing libraries such as OpenCV to compromise LLM pipelines before model inference occurs. Boards face unquantified supply-chain risk in AI systems where third-party tooling receives insufficient security scrutiny.
Systemic Bias and Fairness Failures in Generative AI Models
Training data biases propagate into generative AI outputs, producing stereotyping, racism, and cultural value imposition at scale. Boards face reputational, regulatory, and equity risks as power concentrates in large AI labs and access remains unequal.
Private Personal Data Ingested into LLM Training Corpora
Large language models trained on web-scraped and conversational data risk encoding personally identifiable information, including names, addresses, and career records, without consent. Educational institutions deploying such models face regulatory liability and reputational harm if student or staff data is implicated.
Facial Recognition in Finance Raises Unresolved Privacy and Legal Risks
Facial recognition and biometric AI in financial services creates unresolved questions over data retention, ownership, and legal disclosure obligations. Boards without clear governance frameworks face regulatory exposure and liability if automated loan or identity decisions are challenged in court.
Generative AI Alignment Failures Place Public Sector Governance at Risk
Generative AI systems risk reward hacking, deceptive alignment, and goal misgeneralisation when trained on poorly specified or unrepresentative human values. Governments deploying such systems face accountability gaps when no legitimate authority defines whose values govern AI behaviour.
AI System Incompetence Causing Unjust Financial Decisions
AI models deployed in financial services fail at core tasks, producing erroneous loan and application rejections with material harm to customers. Boards face regulatory exposure and reputational liability where incompetent AI replaces human judgement without adequate oversight.
Beyond accidental failureNational Security
We also track 20 hostile uses of AI.
The public database covers AI that fails by accident. AIBlindspot National Security — exclusive to the Defence tier — tracks AI used as a weapon, mapped by capability:
- State-Sponsored AI Operations
- 6
- AI-Enabled Disinformation
- 5
- Adversarial Attacks on AI
- 0
- Autonomous Weapon Incidents
- 1
- AI-Assisted Cyber Attacks
- 5
- Dual-Use AI Misuse
- 3