MIT AI Risk Repository

Browse AI risks

242 risk entries extracted from 74 frameworks, coded by domain, subdomain, causal entity, intent and timing. Filter, then export the current selection with its licence and citation attached.

Reset

242 entries · page 2 of 5

  1. 62.08.06 · Risk Sub-Category

    Direct Harm Domains (content safety harms)

    Dangerous content (e.g., CBRN)

  2. "Beyond the risks associated with AI technology and its applications, and the legal challenges arising from its development, it is crucial to consider other long- term issues posed by the deployment of increasingly advanced generative AI models. These risks to society, sometimes referred to as “systemic risks,”537 encompass several key areas: the potential for excessive market concentration, the impacts on employment, environmental consequences, and broader risks to humanity."

    From Regulating under Uncertainty: Governance Options for Generative AI (G'sell2024)

  3. "Societal and Cultural - Harms affecting the functioning of societies, communities and economies caused directly or indirectly by the use or misuse technology systems."

    From A Collaborative, Human-Centred Taxonomy of AI, Algorithmic, and Automation Harms (Abercrombie2024)

  4. 62.09.02 · Risk Sub-Category

    Direct Harm Domains (societal harm)

    Economic harm

  5. 70.03.00 · Risk Category

    Economic Risks

  6. 62.13.01 · Risk Sub-Category

    Negative Externality Domains (Other harms from AI development and use)

    Societal inequality (individuals and companies who develop the best AIs get disproportionately powerful)

  7. 62.13.02 · Risk Sub-Category

    Negative Externality Domains (Other harms from AI development and use)

    Geopolitical harms (potential for conflict due to power imbalances)

  8. 62.11.02 · Risk Sub-Category

    Negative Externality Domains (Manufacturing of AI Hardware)

    Human rights harms from exploitation of human labour

  9. In response to the multitude of new risks associated with generative AI, papers advocate for legal regulation and governmental oversight. The focus of these discussions centers on the need for international coordination in AI governance, the establishment of binding safety standards for frontier models, and the development of mechanisms to sanction non-compliance. Furthermore, the literature emphasizes the necessity for regulators to gain detailed insights into the research and development processes within AI labs. Moreover, risk management strategies of these labs shall be evaluated. However,

    From Mapping the Ethics of Generative AI: A Comprehensive Scoping Review (Hagendorff2024)

  10. 19.04.04 · Risk Sub-Category

    Social AI Risks

    Lack of knowledge and social acceptance regarding AI

  11. 19.06.05 · Risk Sub-Category

    Legal AI Risks

    Capturing future AI development and their threats with appropriate mechanism

  12. 62.11.01 · Risk Sub-Category

    Negative Externality Domains (Manufacturing of AI Hardware)

    Environmental harms from exploitation of natural resources

  13. 62.12.01 · Risk Sub-Category

    Negative Externality Domains (Running AI Hardware)

    Environmental harms from energy usage

  14. 62.07.02 · Risk Sub-Category

    Direct Harm Domains (system and operational)

    Operational harms (financial markets)

  15. 62.21.00 · Risk Category

    Agency

    "This section catalogs the risk sources and risk management measures related to agentic AI systems. We categorize these into the following groups: goal- directedness, deception, situational awareness, self-proliferation, and persuasion"

    From Risk Sources and Risk Management Measures in Support of Standards for General-Purpose AI Systems (Gipiškis2024)

  16. 72.05.00 · Risk Category

    Model Capabilities

  17. 62.07.03 · Risk Sub-Category

    Direct Harm Domains (system and operational)

    Operational harms (critical infrastructure)

  18. 62.07.04 · Risk Sub-Category

    Direct Harm Domains (system and operational)

    Operational harms (other physical systems e.g., transport)

  19. 62.15.08 · Risk Sub-Category

    Model Development

    Fine-tuning related (Excessive or overly restrictive safety-tuning)

    "Excessive safety training or safety tuning can impair the performance of AI systems, leading to overly cautious behavior. As a result, these systems may refuse to answer entirely safe prompts which are partially similar to harmful ones [27]."

    From Risk Sources and Risk Management Measures in Support of Standards for General-Purpose AI Systems (Gipiškis2024)

  20. Being a multifaceted concept, the term 'transparency' is both used to refer to technical explainability as well as organizational openness. Regarding the former, papers underscore the need for mechanistic interpretability and for explaining internal mechanisms in generative models. On the organizational front, transparency relates to practices such as informing users about capabilities and shortcomings of models, as well as adhering to documentation and reporting requirements for data collection processes or risk evaluations.

    From Mapping the Ethics of Generative AI: A Comprehensive Scoping Review (Hagendorff2024)

  21. 02.11.01 · Risk Sub-Category

    Not-Suitable-for-Work (NSFW) Prompts

    Insults

  22. 02.11.02 · Risk Sub-Category

    Not-Suitable-for-Work (NSFW) Prompts

    Crimes

  23. 02.11.03 · Risk Sub-Category

    Not-Suitable-for-Work (NSFW) Prompts

    Sensitive Politics

  24. 02.11.04 · Risk Sub-Category

    Not-Suitable-for-Work (NSFW) Prompts

    Physical Harm

  25. 02.11.05 · Risk Sub-Category

    Not-Suitable-for-Work (NSFW) Prompts

    Mental Health

  26. 02.11.06 · Risk Sub-Category

    Not-Suitable-for-Work (NSFW) Prompts

    Unfairness

  27. 05.14.00 · Risk Category

    Evaluation - Auditing

    Closely related to other clusters like AI safety, fairness, or harmful content, papers stress the importance of evaluating generative AI systems both in a narrow technical way as well as in a broader sociotechnical impact assessment focusing on pre-release audits as well as post-deployment monitoring. Ideally, these evaluations should be conducted by independent third parties. In terms of technical LLM or text-to-image model audits, papers furthermore criticize a lack of safety benchmarking for languages other than English.

    From Mapping the Ethics of Generative AI: A Comprehensive Scoping Review (Hagendorff2024)

  28. 05.19.00 · Risk Category

    Miscellaneous

    While the scoping review identified distinct topic clusters within the literature, it also revealed certain issues that either do not fit into these categories, are discussed infrequently, or in a nonspecific manner. For instance, some papers touch upon concepts like trustworthiness, accountability, or responsibility, but often remain vague about what they entail in detail. Similarly, a few papers vaguely attribute socio-political instability or polarization to generative AI without delving into specifics. Apart from that, another minor topic area concerns responsible approaches of talking abo

    From Mapping the Ethics of Generative AI: A Comprehensive Scoping Review (Hagendorff2024)

  29. 10.07.00 · Risk Category

    Injustice

  30. "What can be evaluated in a technical system and its components'...The following categories are high-level, non-exhaustive, and present a synthesis of the findings across different modalities"

    From Evaluating the Social Impact of Generative AI Systems in Systems and Society (Solaiman2023)

  31. "what can be evaluated among people and society"

    From Evaluating the Social Impact of Generative AI Systems in Systems and Society (Solaiman2023)

  32. "Fig 3 shows that technological, data, and analytical AI risks are characterised by the loss of control over AI systems, whereby in particular the autonomous decision and its consequences are classified as risk factors since they are not subject to human influence (Boyd & Wilson, 2017; Scherer, 2016; Wirtz et al., 2019). Programming errors in algorithms due to the lack of expert knowledge or to the increasing complexity and black-box character of AI systems may also lead to undesired AI results (Boyd & Wilson, 2017; Danaher et al., 2017). In addition, a lack of data, poor data quality, and bia

    From Governance of artificial intelligence: A risk and guideline-based integrative framework (Wirtz2022)

  33. 20.03.02 · Risk Sub-Category

    AI Society

    Social acceptance and trust in AI

    "Social acceptance and trust in AI is highly interconnected with the other challenges mentioned. Acceptance and trust result from the extent to which an individual’s subjective expectation corresponds to the real effect of AI on the individual’s life. In the case of transparent and explainable AI, acceptance may be high but if an individual encounters harmful AI behavior like discrimination, acceptance for AI will eventually decline (COMEST, 2017).

    From The Dark Sides of Artificial Intelligence: An Integrated AI Governance Framework for Public Administration (Wirtz2020)

  34. 21.01.00 · Risk Category

    Data-level risk

  35. 21.02.00 · Risk Category

    Model-level risk

  36. "An essential factor in preventing accidents and maintaining low levels of risk lies in the organizations responsible for these technologies."

    From An Overview of Catastrophic AI Risks (Hendrycks2023)

  37. 22.03.02 · Risk Sub-Category

    Organizational Risks (Accidental)

    Organizational Factors can Reduce the Chances of Catastrophe

    "Some organizations successfully avoid catastrophes while operating complex and hazardous systems such as nuclear reactors, aircraft carriers, and air traffic control systems [92, 93]. These organizations recognize that focusing solely on the hazards of the technology involved is insufficient; consideration must also be given to organizational factors that can contribute to accidents, including human factors, organizational procedures, and structure. These are especially important in the case of AI, where the underlying technology is not highly reliable and remains poorly understood"

    From An Overview of Catastrophic AI Risks (Hendrycks2023)

  38. 26.01.00 · Risk Category

    Transparency

    "Ability to provide responsible disclosure to those affected by AI systems to understand the outcome"

    From Summary Report: Binary Classification Model for Credit Risk (AIVerify2023)

Informational only, not legal advice. Verify every claim against the linked official sources and consult qualified counsel before acting.