Unfair discrimination and misrepresentation

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

Which companies acknowledge this risk?

• SAP: “The most well-known example of discrimination by intelligent software is the exclusion of historically underrepresented groups in job application processes. The historical data with which the AI is trained may reflect the biased selection criteria of the past. Therefore, it is theoretically possible that the AI may adopt and reproduce these biases for its own selection process.” (The Ethical Side of AI at SAP | SAP News Center)
• Salesforce: “Safety: Mitigate bias… For example, an LLM can be guided to prevent the use of gender identity, age, race, sexual orientation, socioeconomic status, and other variables.” (How Salesforce Shapes Ethical AI Standards in the Agent Era)
• IBM: “Training an AI system on data with bias, such as historical or societal bias, could lead to biased or skewed outputs that can unfairly represent or otherwise discriminate against certain groups or individuals.” (Foundation models: Opportunities, risks and mitigations)
• Microsoft: “Identify and prioritize demographic groups, including marginalized groups, that may be at risk of being differentially impacted by the system... [and] designed to do so in a manner that minimizes disparities in outcomes for identified demographic groups” (Microsoft Responsible AI Standard v2 General Requirements)
• Nomura Research Institute: “NRI recognizes that the accuracy and representative bias of the data used for AI training, etc., as well as social biases inherent in the data, may cause errors or biases in the results of AI judgments... [and] strives to ensure that no individual or group is unfairly discriminated against in the use of AI.” (Nomura Research Institute 2025 Integrated Report)
• Palantir: “applying facial recognition technology to predict sexual orientation from facial features ... implicates sensitive category inferences that are morally, scientifically, and legally questionable ... potentially negative feedback loops, and subtly compounding disparate impacts” (Palantir AI Ethics)
• Intel: “these models learn from vast amounts of data available on the internet, which can contain biases present in society and may inadvertently perpetuate and amplify these biases” (Intel Responsible AI - E Book - Rev 2.6)
• UnitedHealth: “proactively assessing the performance of our AI solutions for potential bias and disparate impact on vulnerable groups” (UnitedHealth Group 2024 Sustainability Report - Our Mission in Action)
• Novartis: “Ensure AI training Data is representative and free from bias that could lead to discrimination against certain groups or individuals.” (Ethical Use of Data and Technology Policy)
• Alphabet (Google): “safeguards to mitigate unintended or harmful outcomes and avoid unfair bias… while minimizing inaccurate overgeneralizations and biases.” (Google - OECD HAIP Report)
• Amazon: “a dataset... prompts that test for bias across profession, gender, race, religion, and political ideology... In 2023, we introduced TANGO... tests for misgendering of transgender and nonbinary (TGNB) people” (Amazon Nova and our commitment to responsible AI)
• Mastercard: “AI algorithms may generate... unfair, biased or discriminatory outcomes” (Mastercard 2024 Annual report (10-K))
• Baidu: “governance models to address a range of ethics issues, including... bias, discrimination” (Baidu 2025 Environmental, Social and Governance Report)
• Citigroup: “if training data contains bias, the model will learn and reproduce that bias... This bias can lead to content being created that reinforces cultural, racist or gender-based stereotypes... its outputs could discriminate against certain groups” (Delivering Improved Experience Through Generative AI)
• National Australia Bank: “The deployment of AI technologies may also amplify existing risks across other risk categories, particularly if new applications are introduced without adequate safeguards for... discrimination” (NAB 2025 Annual Report)
• ServiceNow: “AI systems may not perform as intended or may produce outcomes, such as unreliable or biased results” (ServiceNow 2025 Form 10-K Annual Report)
• Abbott Laboratories: “technologies that inadvertently perpetuate biases in the data used to train them, leading to disparities in health outcomes” (Abbott 2024 Global Sustainability Report June 2025)
• Cisco: “AI creates the potential for harmful human bias or stereotypes to become ingrained in or amplified by technological systems.” (Cisco Principles for Responsible Artificial Intelligence)
• Commonwealth Bank of Australia: “it also introduced bias LLMs often fixated on negative traits when presented with fuller personas, assuming financial struggles... LLMs stereotype populations” (Artificial intelligence)
• Eli Lilly: “The algorithms and models utilized in AI systems may have limitations, including biases” (Eli Lilly 2024 Annual Report)
• Fortinet: “The use of AI also brings ethical issues related to privacy, surveillance and consent of use, as well as potential for bias and discrimination.” (Fortinet 2024 Annual Report)
• Goldman Sachs: “may produce output or take action ... that reflect biases included in the data on which they are trained” (Goldman Sachs 2024 Form 10-K)
• Intuit: “AI algorithms that we use may be flawed or may be based on datasets that are biased or insufficient.” (Intuit 2025 Annual Report)
• L'Oreal: “L'Oréal wishes to promote the responsible use of AI, notably to avoid algorithmic bias.” (L'Oreal - CODE OF ETHICS)
• Meta: “we cannot guarantee that third parties will not use such AI technologies for improper purposes, including... furthering bias or discrimination” (Meta - Form 10-K 2025 Annual Report)
• NVIDIA: “AI software products that we or our partners offer rely on training data that may originate from third parties and new training methods, and the resulting products may contain unknown or undetected defects and errors, or reflect unintended bias.” (NVIDIA 2025 Annual Report and 10-K)
• Prosus: “Such inherent risks range from... biased outcomes and potential for discrimination or unjust treatment for some groups of users.” (Prosus Annual Report 2025)
• Samsung: “we will work to avoid discrimination against any individual or group based on gender, ethnicity, nationality, income, sexual orientation, political or religious belief or other factors. ... ensure that AI does not promote or reinforce unfair bias.” (Samsung Electronics' AI Ethics Principles | AI Ethics | Tech for All ...)
• Wells Fargo: “With limited or untested data, algorithms can learn false assumptions about various demographics, how people speak, or even where they're located.” (How Wells Fargo builds responsible artificial intelligence)
• Alnylam Pharmaceuticals: “ensure that they do not cause harm, or result in discrimination or injustice... We strive to avoid unfair bias... We make sure that the algorithms we use are not biased (e.g. using an inadequate data set)” (Responsible Use of Artificial Intelligence)
• American Express: “flaws in models or datasets that may result in biased ... results” (American Express 2024 Annual Report)
• Amgen: “assessing potential impacts on individual rights, fairness, transparency, and risks such as algorithmic bias” (Amgen Personal Information Protection Policy)
• Apple: “We developed customized datasets to mitigate culture-specific risks and biases and stereotypes in model outputs.” (Updates to Apple's On-Device and Server Foundation Language Models Framework)
• AT&T: “But what if a new algorithm to skirt outages kept favoring urban customers at the expense of rural ones? Not on purpose - but as an unintended consequence.” (AT&T AI Guiding Principles)
• Johnson & Johnson: “external influences, such as biases that exist in healthcare, may affect them... we can identify when and why it may produce biased results.” (Doing the Right Thing - AI & Ethics - Johnson & Johnson)
• Kenvue: “We are actively working to mitigate algorithmic bias and ensure that automated decision-making does not negatively impact vulnerable populations” (Kenvue 2024 Healthy Lives Mission Report)
• Merck: “Ethical and legal challenges may arise, including biases or discrimination in AI outcomes” (Merck - 2025 Annual Report)
• Mitsubishi UFJ Financial: “issues of fairness that must be avoided when having AI make decisions, as AI may judge based on the gender, race, or other personal attributes of the subjects involved” (MUFG Human Rights Report 2024)
• Shopify: “AI decisions or outputs... may... lead to unintentional bias or discrimination;” (Shopify Inc. Annual Report on Form 10-K (Fiscal Year 2025))
• T-Mobile US: “AI may have unintended consequences, including inaccurate or biased responses.” (T-Mobile Use Systems with Care)
• Uber: “These and other deficiencies could also undermine the decisions, predictions or analysis AI applications produce, or lead to unintentional bias and discrimination, subjecting us to competitive harm, legal liability, and brand or reputational harm.” (Uber 10-K 2025 Annual Report)
• Alibaba: “it will favor top sellers while creating an unfair market for smaller merchants.” (Alibaba's CTO On Everything You Wanted To Know About AI Ethics (But Were Too Afraid To Ask))
• AstraZeneca: “AI systems can amplify biased and discriminatory decision making” (AstraZeneca Risk Supplement 2024)
• Constellation Software: “The use of AI in our businesses may generate risks with respect to ... perceived or actual bias, discrimination” (Constellation Software - Q4 2025 MD&A)
• EQT: “they also introduce challenges such as cybersecurity vulnerabilities, regulatory scrutiny, and risks related to data governance and AI bias.” (EQT 2024 Annual report)
• HDFC Bank: “Artificial intelligence algorithms may be flawed, for example if datasets contain biased information” (HDFC - Form 20-F 2025 Annual Report)
• Hilton Worldwide: “The use of artificial intelligence can lead to unintended consequences, including ... unintended biases and discriminatory outcomes” (Hilton 2024 10-K Annual Report)
• JPMorgan Chase: “incorrect, biased or misleading results or content generated by artificial intelligence, leading to harmful outcomes, including discrimination in lending practices against vulnerable populations” (JPMorganChase 2024 10-K Annual report)
• National Bank of Canada: “The growing use of AI in the Bank's operational processes exposes it to new risks, particularly discriminatory bias” (National Bank of Canada Annual Report 2025)
• Novo Nordisk: “Developers and owners should strive to minimize bias and maximize fairness in AI systems... evaluate if those assumptions minimize any bias of the developer or a bias that is present in the data” (Data ethics | Novo Nordisk data ethics principles)
• Philip Morris International: “This workstream will aim to mitigate, amongst others, risks such as: ... Amplification of bias” (The state of AI at PMI | PMI - Philip Morris International)
• PNC Financial Services: “AI poses unique privacy challenges: ... the risk that AI could identify or treat an individual differently based on the available information.” (PNC Protecting Customer Privacy & Security)
• Renesas Electronics: “Renesas continuously monitors its Data and AI Systems deployed for decisions or any other output that may unfairly disadvantage individuals, groups or communities” (Renesas Data and AI Principles | Renesas)
• ResMed: “Datasets may be insufficient or contain biased information... [FTC] concern regarding AI and bias... such bias could lead to unfair and deceptive practices” (ResMed 2025 Annual Report (10-K))
• Royal Bank Of Canada: “dealing with bias - which could engender inequalities in decision-making... there are risks of bias being reflected in the data sets leveraged by machine learning. For example, there are cases where racial bias has been detected in AI systems” (Responsible and Explainable AI: Exploring the Future of Trading)
• T. Rowe Price: “AI models... may produce output... that reflect biases included in the data on which they are trained or introduced during the training or fine tuning process” (T. Rowe Price Group 2025 Annual Report (10-K, FY2025))
• Tencent: “refine model performance to reduce biases” (Tencent environmental, social and governance report 2024)
• Xcel Energy: “We will deploy and monitor AI solutions with the goal of identifying and eliminating potential bias and discrimination in our AI-enabled technology.” (Xcel Energy AI Guiding Principles)