Compromising privacy or security by correctly inferring sensitive information
AI systems that memorize and leak sensitive personal data or infer private information about individuals without their consent. Unexpected or unauthorized sharing of data and information can compromise user expectation of privacy, assist identity theft, or cause loss of confidential intellectual property.
Anticipated risk: "Privacy violations may occur at inference time even without an individual’s data being present in the training corpus. Insofar as LMs can be used to improve the accuracy of inferences on protected traits such as the sexual orientation, gender, or religiousness of the person providing the input prompt, they may facilitate the creation of detailed profiles of individuals comprising true and sensitive information without the knowledge or consent of the individual."(p. 218)
Supporting Evidence (1)
Example: "Notably, risks may arise even if LM inferences are false, but believed to be correct. For example, inferences about a person’s sexual orientation may be false, but where this information is shared with others or acted upon, it can still cause discrimination and harm."(p. 218)
Part of Risk area 2: Information Hazards
Other risks from Weidinger et al. (2022) (25)
Risk area 1: Discrimination, Hate speech and Exclusion
1.2 Exposure to toxic contentRisk area 1: Discrimination, Hate speech and Exclusion > Social stereotypes and unfair discrimination
1.1 Unfair discrimination and misrepresentationRisk area 1: Discrimination, Hate speech and Exclusion > Hate speech and offensive language
1.2 Exposure to toxic contentRisk area 1: Discrimination, Hate speech and Exclusion > Exclusionary norms
1.1 Unfair discrimination and misrepresentationRisk area 1: Discrimination, Hate speech and Exclusion > Lower performance for some languages and social groups
1.3 Unequal performance across groupsRisk area 2: Information Hazards
2.1 Compromise of privacy by leaking or correctly inferring sensitive information