MIT AI Risk Repository

Browse AI risks

50 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 Also filtered by framework Liu2024 ×

50 entries

  1. 30.03.01 · Risk Sub-Category

    Fairness

    Injustice

    In the context of LLM outputs, we want to make sure the suggested or completed texts are indistinguishable in nature for two involved individuals (in the prompt) with the same relevant profiles but might come from different groups (where the group attribute is regarded as being irrelevant in this context)

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  2. 30.03.02 · Risk Sub-Category

    Fairness

    Stereotype Bias

    LLMs must not exhibit or highlight any stereotypes in the generated text. Pretrained LLMs tend to pick up stereotype biases persisting in crowdsourced data and further amplify them

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  3. 30.03.03 · Risk Sub-Category

    Fairness

    Preference Bias

    LLMs are exposed to vast groups of people, and their political biases may pose a risk of manipulation of socio-political processes

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  4. 30.07.03 · Risk Sub-Category

    Robustness

    Interventional Effect

    existing disparities in data among different user groups might create differentiated experiences when users interact with an algorithmic system (e.g. a recommendation system), which will further reinforce the bias

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  5. 30.02.00 · Risk Category

    Safety

    Avoiding unsafe and illegal outputs, and leaking private information

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  6. 30.02.01 · Risk Sub-Category

    Safety

    Violence

    LLMs are found to generate answers that contain violent content or generate content that responds to questions that solicit information about violent behaviors

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  7. 30.02.02 · Risk Sub-Category

    Safety

    Unlawful Conduct

    LLMs have been shown to be a convenient tool for soliciting advice on accessing, purchasing (illegally), and creating illegal substances, as well as for dangerous use of them

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  8. 30.02.03 · Risk Sub-Category

    Safety

    Harms to Minor

    LLMs can be leveraged to solicit answers that contain harmful content to children and youth

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  9. 30.02.04 · Risk Sub-Category

    Safety

    Adult Content

    LLMs have the capability to generate sex-explicit conversations, and erotic texts, and to recommend websites with sexual content

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  10. 30.06.00 · Risk Category

    Social Norm

    LLMs are expected to reflect social values by avoiding the use of offensive language toward specific groups of users, being sensitive to topics that can create instability, as well as being sympathetic when users are seeking emotional support

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  11. 30.06.01 · Risk Sub-Category

    Social Norm

    Toxicity

    language being rude, disrespectful, threatening, or identity-attacking toward certain groups of the user population (culture, race, and gender etc)

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  12. 30.06.03 · Risk Sub-Category

    Social Norm

    Cultural Insensitivity

    it is important to build high-quality locally collected datasets that reflect views from local users to align a model’s value system

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  13. 30.03.00 · Risk Category

    Fairness

    Avoiding bias and ensuring no disparate performance

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  14. 30.03.04 · Risk Sub-Category

    Fairness

    Disparate Performance

    The LLM’s performances can differ significantly across different groups of users. For example, the question-answering capability showed significant performance differences across different racial and social status groups. The fact-checking abilities can differ for different tasks and languages

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  15. 30.02.06 · Risk Sub-Category

    Safety

    Privacy Violation

    machine learning models are known to be vulnerable to data privacy attacks, i.e. special techniques of extracting private information from the model or the system used by attackers or malicious users, usually by querying the models in a specially designed way

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  16. 30.07.01 · Risk Sub-Category

    Robustness

    Prompt Attacks

    carefully controlled adversarial perturbation can flip a GPT model’s answer when used to classify text inputs. Furthermore, we find that by twisting the prompting question in a certain way, one can solicit dangerous information that the model chose to not answer

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  17. 30.07.04 · Risk Sub-Category

    Robustness

    Poisoning Attacks

    fool the model by manipulating the training data, usually performed on classification models

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  18. 30.01.00 · Risk Category

    Reliability

    Generating correct, truthful, and consistent outputs with proper confidence

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  19. 30.01.01 · Risk Sub-Category

    Reliability

    Misinformation

    Wrong information not intentionally generated by malicious users to cause harm, but unintentionally generated by LLMs because they lack the ability to provide factually correct information.

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  20. 30.01.02 · Risk Sub-Category

    Reliability

    Hallucination

    LLMs can generate content that is nonsensical or unfaithful to the provided source content with appeared great confidence, known as hallucination

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  21. 30.01.04 · Risk Sub-Category

    Reliability

    Miscalibration

    over-confidence in topics where objective answers are lacking, as well as in areas where their inherent limitations should caution against LLMs’ uncertainty (e.g. not as accurate as experts)... ack of awareness regarding their outdated knowledge base about the question, leading to confident yet erroneous response

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  22. 30.01.05 · Risk Sub-Category

    Reliability

    Sychopancy

    flatter users by reconfirming their misconceptions and stated beliefs

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  23. 30.07.02 · Risk Sub-Category

    Robustness

    Paradigm & Distribution Shifts

    Knowledge bases that LLMs are trained on continue to shift... questions such as “who scored the most points in NBA history" or “who is the richest person in the world" might have answers that need to be updated over time, or even in real-time

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  24. 30.04.00 · Risk Category

    Resistance to Misuse

    Prohibiting the misuse by malicious attackers to do harm

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  25. 30.04.01 · Risk Sub-Category

    Resistance to Misuse

    Propaganda

    LLMs can be leveraged, by malicious users, to proactively generate propaganda information that can facilitate the spreading of a target

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  26. 30.04.02 · Risk Sub-Category

    Resistance to Misuse

    Cyberattack

    ability of LLMs to write reasonably good-quality code with extremely low cost and incredible speed, such great assistance can equally facilitate malicious attacks. In particular, malicious hackers can leverage LLMs to assist with performing cyberattacks leveraged by the low cost of LLMs and help with automating the attacks.

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  27. 30.04.03 · Risk Sub-Category

    Resistance to Misuse

    Social-Engineering

    psychologically manipulating victims into performing the desired actions for malicious purposes

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  28. 30.04.04 · Risk Sub-Category

    Resistance to Misuse

    Copyright

    The memorization effect of LLM on training data can enable users to extract certain copyright-protected content that belongs to the LLM’s training data.

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  29. 30.01.03 · Risk Sub-Category

    Reliability

    Inconsistency

    models could fail to provide the same and consistent answers to different users, to the same user but in different sessions, and even in chats within the sessions of the same conversation

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  30. 30.05.02 · Risk Sub-Category

    Explainability & Reasoning

    Limited Logical Reasoning

    LLMs can provide seemingly sensible but ultimately incorrect or invalid justifications when answering questions

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  31. 30.05.03 · Risk Sub-Category

    Explainability & Reasoning

    Limited Causal Reasoning

    Causal reasoning makes inferences about the relationships between events or states of the world, mostly by identifying cause-effect relationships

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  32. 30.06.02 · Risk Sub-Category

    Social Norm

    Unawareness of Emotions

    when a certain vulnerable group of users asks for supporting information, the answers should be informative but at the same time sympathetic and sensitive to users’ reactions

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  33. 30.07.00 · Risk Category

    Robustness

    Resilience against adversarial attacks and distribution shift

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  34. The ability to explain the outputs to users and reason correctly

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  35. 30.05.01 · Risk Sub-Category

    Explainability & Reasoning

    Lack of Interpretability

    Due to the black box nature of most machine learning models, users typically are not able to understand the reasoning behind the model decisions

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  36. 30.01.00.a · Additional evidence

    Reliability

  37. 30.01.02.a · Additional evidence

    Reliability

    Hallucination

  38. 30.01.05.a · Additional evidence

    Reliability

    Sychopancy

  39. 30.01.05.b · Additional evidence

    Reliability

    Sychopancy

  40. 30.02.03.a · Additional evidence

    Safety

    Harms to Minor

  41. 30.02.05 · Risk Sub-Category

    Safety

    Mental Health Issues

    unhealthy interactions with Internet discussions can reinforce users’ mental issues

    From Trustworthy LLMs: A Survey and Guideline for Evaluating Large Language Models’ Alignment (Liu2024)

  42. 30.02.06.a · Additional evidence

    Safety

    Privacy Violation

  43. 30.03.00.a · Additional evidence

    Fairness

  44. 30.03.00.b · Additional evidence

    Fairness

  45. 30.03.01.a · Additional evidence

    Fairness

    Injustice

  46. 30.03.03.a · Additional evidence

    Fairness

    Preference Bias

  47. 30.03.03.b · Additional evidence

    Fairness

    Preference Bias

  48. 30.04.03.a · Additional evidence

    Resistance to Misuse

    Social-Engineering

  49. 30.04.03.b · Additional evidence

    Resistance to Misuse

    Social-Engineering

  50. 30.06.01.a · Additional evidence

    Social Norm

    Toxicity

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