IBM is facing shareholder pressure to reveal more about how it manages bias in its Granite AI models, but the company argues it already provides extensive transparency through public reports and open-source documentation. At its April 28 shareholder meeting, IBM will defend its current approach against a resolution demanding a formal report on bias elimination methods within the next year.

What Does IBM Say About Its AI Bias Practices?

IBM's official response to the shareholder proposal emphasizes that the company has been transparent since releasing its first Granite model. The company points to technical reports, model cards, and other documentation as evidence of openness. IBM also highlights its participation in Stanford University's Foundation Model Transparency Index (FMTI), a benchmarking initiative that measures how openly companies disclose information about their foundation models, including data sources, training methods, evaluation metrics, risks, and governance practices.

Beyond documentation, IBM argues that its enterprise customers have the tools they need to address bias concerns themselves. "IBM models are smaller and targeted towards enterprise clients and use cases," the company stated. "These models are not general purpose, consumer-facing models. Therefore, our open-source models are built in a manner that allows our clients to build an AI solution that will address their specific needs".

Why Are Experts Skeptical of IBM's Position?

While industry analysts acknowledge that IBM discloses more bias information than most competitors, they argue the company is quietly shifting responsibility to customers. The core issue, experts say, is not a lack of transparency but a fundamental flaw in how the industry approaches bias mitigation.

"IBM's stance deserves to be taken seriously, but not at face value. The company is right to say that bias mitigation, fairness frameworks, and governance controls are already built into its AI systems. That is not in dispute. In fact, compared to much of the market, IBM has been more deliberate than most in turning responsible AI from a set of principles into something operational," said Sanchit Vir Gogia, chief analyst at Greyhound Research.

Sanchit Vir Gogia, Chief Analyst at Greyhound Research

However, Gogia added that when IBM points to customer fine-tuning as a solution, "it is quietly acknowledging a limitation. It is admitting that whatever happens at the model layer is not the end of the story. It is only the beginning of it".

Gogia

The Real Problem: Where Bias Actually Originates

The shareholder debate reveals a deeper industry-wide problem. Most AI vendors, including IBM, measure and attempt to fix bias after models are trained. But researchers argue this approach is fundamentally limited because bias often originates in the training data itself, long before any post-training adjustments can address it.

"The proponent frames disparate-impact correction as a threat to accuracy, but the real issue is that IBM, and every major model provider, is measuring bias at the output layer when most of it originates upstream. You cannot fairness-tune your way out of a training data problem," explained Noah Kenney, principal consultant for Digital 520.

Noah Kenney, Principal Consultant for Digital 520

Kenney noted that IBM's transparency efforts are genuine. "FMTI scores, model cards, FairIQ, Equi-tuning, FairReprogram, and the Granite transparency posture are all real, and more than most of their peers publish. The gap is not disclosure. The gap is that the industry has converged on post-hoc mitigation as the dominant paradigm, and post-hoc mitigation has diminishing returns once a model is trained".

Kenney

How Should Companies Address AI Bias? Key Approaches

• Upstream Data Curation: Address bias at the source by carefully selecting and cleaning training data before models are built, rather than trying to fix bias after training is complete.
• Transparency and Disclosure: Publish detailed information about data sources, training methods, and known limitations through technical reports, model cards, and participation in benchmarking initiatives like Stanford's FMTI.
• Customer-Facing Tools: Provide enterprise clients with fine-tuning capabilities and governance frameworks so they can adapt models to their specific use cases and address domain-specific bias concerns.
• Independent Audits and Standardized Benchmarks: Subject AI systems to external review and establish industry-wide standards for measuring and reporting bias, rather than relying solely on vendor self-assessment.
• Post-Deployment Monitoring: Continuously measure model performance after deployment to catch emerging bias issues and provide customers with tools to monitor and adjust their implementations.

Is Eliminating Bias Even Possible?

Some experts question whether the goal of eliminating bias is realistic. The definition of "unbiased" itself is subjective; what one stakeholder considers fair, another may view as unfair.

"I truly don't think eliminating bias is possible, and that's not an IBM problem. The whole market is operating the same way in that any model trained on human-generated data carries the biases of whoever made it, which is exactly the same as humans would do. What vendors can do, IBM included, as they do a bunch of this already, is measure it, disclose it, monitor it after deployment, and give customers tools to adapt," said Mike Leone, VP and principal analyst at Moor Insights and Strategy.

Mike Leone, VP and Principal Analyst at Moor Insights and Strategy

Leone added that anyone promising to completely eliminate bias is "telling you what you want to hear." Instead, the industry should focus on mitigation, measurement, and transparency.

Leone

The shareholder resolution highlights a tension facing enterprise AI adoption. Companies want assurance that their AI systems are fair and trustworthy, but vendors argue that responsibility ultimately lies with the customers who deploy and fine-tune the models. As AI becomes more central to business operations, this accountability gap will likely remain a flashpoint for investors, regulators, and customers alike.