Compromise of privacy by obtaining, leaking or correctly inferring sensitive information

36 of the world's largest companies we reviewed acknowledge the AI risk "Loss of privacy", including Salesforce, IBM and SAP.

Which companies acknowledge this risk?

• Salesforce: “data masking is the process that replaces sensitive Personally Identifiable Information (PII) or proprietary business data with non-identifiable tokens before the prompt is sent to the LLM.” (Trusted AI: Key Principles | Salesforce)
• IBM: “ChatGPT suffered a bug and exposed titles and active users' chat history to other users... even more private data... first and last name, email address... last four digits of their credit card number” (Foundation models: Opportunities, risks and mitigations)
• SAP: “data may accidentally become embedded within a GenAI model during training or development. As a result, GenAI training data is at risk of exfiltration...” (SAP Generative AI Cybersecurity Strategy)
• Alphabet (Google): “the increased risk of inadvertent disclosure of confidential information or personal data in connection with the utilization of AI technologies” (Alphabet - Form 10-K 2024 Annual Report)
• Meta: “We employed different techniques to reduce memorization... we red-teamed the model for memorization of information about private individuals and took steps to mitigate the model against those risks.” (Expanding Our Open Source Large Language Models Responsibly)
• Mastercard: “may inadvertently disclose confidential information and/or breach intellectual property, privacy or other rights” (Mastercard 2024 Annual report (10-K))
• Nomura Research Institute: “respond not only to the fundamental risks of AI such as "information divulgence," "bias and unfairness," and "invasion of privacy” (NRI - ESG Databook 2025)
• Microsoft: “The team also used red teaming to test the model's ability to infer sensitive attributes from users' voices. They found that while the model inferred sensitive attributes in 27% of test prompts” (Responsible AI Transparency Report | Microsoft)
• Palantir: “The notice-and-consent paradigm for managing privacy risks has failed in the face of complex and sophisticated uses and transfers of personal information...” (Palantir Australia Response to Supporting Responsible AI ...)
• Amazon: “Our set of risk categories includes the following: sensitive-data exfiltration” (Amazon Nova and our commitment to responsible AI)
• Baidu: “Inappropriate or controversial data practices... including the inappropriate collection of data and potential data misuse” (Baidu 2025 Annual Report)
• QUALCOMM: “As AI systems collect and analyze vast amounts of data, it is essential to protect individuals' privacy rights and ensure the security of sensitive information.” (Shaping the future of AI responsibly | Nasdaq | Qualcomm)
• Visa: “The increased risk of inadvertent disclosure of confidential information or personal data in connection with the utilization of AI technologies” (Visa 2025 Annual report)
• Cisco: “Applications of AI often use personal data that could impact individual privacy and civil liberties if not managed properly.” (Cisco Principles for Responsible Artificial Intelligence)
• Citigroup: “Handling of personally identifiable data by the LLM requires strong safeguards to prevent leakage, and ensure privacy and security.” (Delivering Improved Experience Through Generative AI)
• Eli Lilly: “people, their activities, and health state can be gleamed from insignificant-looking data captured by commercially available devices including cell phones and smart watches.” (Lilly AI Policy and Implications for Digital Health and Drug Development)
• Fortinet: “If customer data is used to train AI based systems and such data is not adequately anonymized, this may lead to breach of sensitive information and loss of customer trust.” (Fortinet 2024 Annual Report)
• Novartis: “If we do not have sufficient rights to use the data... we may also incur liability through the violation of... privacy... rights” (Novartis Annual Report 2025 - Form 20-F)
• ServiceNow: “We identify rare but high-impact failure scenarios, such as... leaking personal data.” (How We Protect High-Risk AI Systems | ServiceNow)
• T. Rowe Price: “We are also cognizant of privacy and security risks, as large volumes of data are consumed in training and using AI models.” (Using Artificial Intelligence to Enhance Our Investment Processes | T Rowe Price)
• Tencent: “data security, privacy breaches ... are real challenges facing existing large-scale model applications” (Large Model Safety and Ethics Research Report 2024 (announcement))
• Amgen: “we may experience adverse effects on our business, including data breaches... unintentional disclosure of personal data” (Amgen 2024 Annual Report Letter and 10-K)
• Apple: “The risks of inadvertent disclosure of personal data can increase with the introduction of new and complex technologies, such as artificial intelligence features, further exacerbating such risks.” (Apple 2025 Annual Report on Form 10-K)
• Blackstone Group: “For example, a user may input confidential information, including material non-public information or personal identifiable information, into AI Technology applications, resulting in such information becoming part of a dataset that is accessible by third-party AI Technology applications and users, including our competitors.” (2024 Blackstone 10-K filing)
• Intuitive Surgical: “with the prolific use of AI technologies, there is an increased risk of unauthorized or accidental disclosure. For example, our employees, third-party service providers, strategic partners, or other contractors or consultants may input inappropriate or confidential information into an AI system (in particular, a system that is managed, owned, or controlled by a third party), thereby compromising our business operations.” (Intuitive Surgical - Form 10-K 2025 Annual Report)
• PNC Financial Services: “AI poses unique privacy challenges: the risk of feeding personal information into an AI system” (PNC Protecting Customer Privacy & Security)
• Samsung: “major risks related to generative AI (... personal information leakage ...)” (Samsung Electronics' Commitment to AI Ethics Principles)
• AT&T: “may... release private or confidential information” (AT&T INC. Form 10-K 2024 Annual Report)
• EssilorLuxottica: “the rising reliance on Artificial Intelligence introduces new risks, particularly concerning data privacy and information disclosure.” (Essilor Luxottica - 2024 Universal Registration Document)
• Goldman Sachs: “may produce output or take action ... that result in the release of private, confidential or proprietary information” (Goldman Sachs 2024 Form 10-K)
• Jabil: “if any of our contractors, consultants, vendors or service providers use any third-party AI-powered software or other tools in connection with our business... it may lead to the inadvertent disclosure of our confidential information through its incorporation into publicly available training sets” (Jabil Form 10-K 2025 Annual Report)
• Kenvue: “the use of artificial intelligence tools may compromise our confidential or sensitive information, result in unauthorized processing of personal data” (Kenvue Annual Report 2024)
• Mitsubishi UFJ Financial: “we recognize the risks associated with the inappropriate use of AI, such as information leaks” (Announcement of the MUFG AI Policy)
• PSEG: “These limitations or failures could result in reputational damage, unauthorized disclosure of data, and legal liabilities.” (PSEG Annual Report 2024)
• Roche: “This reduces the risk of personal identification when AI models are created.” (Roche Artificial Intelligence (AI) Ethics Principles)
• RTX: “Improper use of artificial intelligence could also lead to data breaches” (RTX Annual Report 2024)