INTERNATIONAL BUSINESS MACHINES (IBM) CORPORATION | Report on AI bias at INTERNATIONAL BUSINESS MACHINES (IBM) CORPORATION

Status
Filed
AGM date
Previous AGM date
Proposal number
7
Resolution details
Company ticker
IBM
Lead filer
National Center for Public Policy
Resolution ask
Report on or disclose
ESG theme
  • Social
ESG sub-theme
  • Digital rights
Type of vote
Shareholder proposal
Filer type
Shareholder
Company sector
Technology
Company HQ country
United States
Resolved clause
Shareholders request that the Board of Directors of IBM issue a report, within the next year, on the methods used to eliminate bias from the Company’s artificial intelligence (AI) models, including an assessment of the risk that seeking to avoid “disparate impact” in outputs will undermine the accuracy of, and trust in, those outputs.
Whereas clause
Whereas: IBM has publicly stated that “AI must be designed to … promote inclusive representation.”1 Whereas: Real-world data often reflects actual outcome inequalities and correcting these disparities through model adjustments risks replacing accuracy with outcomes that are, in fact, less representative of reality. Whereas: Evidence of political bias in widely used LLM models exists. For example, Stanford researchers found that four broadly used AI models exhibited what was perceived as a “left-leaning slant” by bipartisan respondents.2 Whereas: The White House has issued an executive order specifically seeking to combat “woke” AI, committing “not to procure models that sacrifice truthfulness and accuracy to ideological agendas.”3 Whereas: IBM has stated that powerful AI technologies “must be transparent.”4 (1) https://www.ibm.com/design/ai/ethics/fairness/ (2) https://news.stanford.edu/stories/2025/05/ai-models-llms-chatgpt-claude-gemini-partisan-bias-research-study (3) https://www.whitehouse.gov/presidential-actions/2025/07/preventing-woke-ai-in-the-federal-government/ (4) https://newsroom.ibm.com/Principles-and-Practices-for-Building-More-Trustworthy-AI
Supporting statement
The requested report could consider: 1.How IBM distinguishes between inaccurate bias and real-world outcome disparities. 2.The safeguards IBM employs to ensure that corrections for bias do not distort data. 3.Whether disparate impact-driven fairness metrics are balanced against accuracy and reliability of AI outputs. 4.Potential business and reputational risks if AI outputs are perceived as less accurate due to over-correction to avoid disparate impact in outputs. We recognize IBM’s attempt to demonstrate leadership in promoting fairness in AI, including development of the open-source AI Fairness 360 toolkit and its asserted commitment to transparency.5 However, as IBM itself has noted, “without the right safeguards, AI could cause harm.”6 Those harms could arise not only from failing to address biased inaccurate outputs, but also from correcting accurate output in ways that reduce accuracy and reliability in the name of equity. As stewards of long-term shareholder value, we believe an assessment of this risk is necessary to ensure IBM remains trusted for transparency and accuracy. (5) https://research.ibm.com/blog/ai-fairness-360 (6) https://www.studocu.com/ru/document/aleksei-nikolaevich-kosygin-russian-state-university/froh-gnlfoxnbf/ibm-policy-lab-mitigating-bias-in-artificial-intelligence/121557100

DISCLAIMER: By including a shareholder resolution or management proposal in this database, neither the PRI nor the sponsor of the resolution or proposal is seeking authority to act as proxy for any shareholder; shareholders should vote their proxies in accordance with their own policies and requirements.

Any voting recommendations set forth in the descriptions of the resolutions and management proposals included in this database are made by the sponsors of those resolutions and proposals, and do not represent the views of the PRI.

Information on the shareholder resolutions, management proposals and votes in this database have been obtained from sources that are believed to be reliable, but the PRI does not represent that it is accurate, complete, or up-to-date, including information relating to resolutions and management proposals, other signatories’ vote pre-declarations (including voting rationales), or the current status of a resolution or proposal. You should consult companies’ proxy statements for complete information on all matters to be voted on at a meeting.