IBM’s watsonx.governance takes aim at AI auditing

With regulation on the rise, IT leaders must ensure AI model safety and integrity. Big Blue believes it has an edge, offering audit trails, real-time monitoring, and platform flexibility.

IBM is betting big on its toolkit for monitoring generative AI and machine learning models, dubbed watsonx.governance, to take on rivals and position the offering as a top AI governance product, according to a senior executive at IBM.

The toolkit, released as part of IBM’s watsonx generative AI platform last May, is uniquely differentiated by its audit trail capability, said Heather Gentile, director of product management at IBM’s data and AI software division.

The capability, according to Gentile, enables watsonx.governance customers to create a repository for logging details throughout a model’s lifecycle, such as the rationale behind a certain model choice or which stakeholder had what involvement in the model’s lifecycle.

Gentile claims that watsonx.governance provides a more “holistic approach or solution” compared to rival solutions as they either don’t have the audit trail capability or they don’t offer real-time monitoring services.

What is watsonx.governance and how does it work?

watsonx.governance is a toolkit for governing generative AI and machine learning models. It focuses on three core areas of documentation: compliance, risk management, and model lifecycle management — processes IBM says are intertwined.

Documentation begins with the initial phase of model lifecycle management and continues through every change and metric essential to the model adoption process, including its use case.

The toolkit comes with a best practices workflow that IBM pushes across the enterprise in combination with the company’s existing governance policy, according to Gentile.

“We’re basically operationalizing [an enterprise’s] best practices so that regardless of who the stakeholder is in the process of creating an AI use case, the toolkit will require the stakeholder to submit pieces of information as they move through the process of adopting a particular model for that task or use case,” Gentile said.

Once a use case is approved for use, IBM conducts a risk analysis based on the enterprise’s AI usage and governance policy, suggesting the type of monitoring the enterprise should employ for that particular use case, Gentile added.

Post risk analysis, the enterprise then compares models to see which model best suits the use case at hand, Gentile said.

“We are capturing the performance information of those models, in fact sheets. We are actually taking a baseline of each of the models that the enterprise is comparing, and also creating an audit trail of what model was chosen to support the use case,” Gentile said, adding that the audit trail is maintained right from the beginning of the process.

The audit trail also includes information from the next phase when the model enters production, after which it is continuously monitored in real-time, according to IBM.

The toolkit will record and notify the enterprise if the model raises an alert or is not performing well or needs to be updated, IBM added.

Why is AI auditing so important?

Gentile believes the audit trail capability sets watsonx.governance apart, especially in the wake of AI regulations that mandate enterprises to audit their models to ensure compliance.

Major economies such as the US, UK, EU, and India are already working on regulations and frameworks for ensuring the safety of large language models (LLMs). Last month, for example, the US and UK signed an agreement to work closely in developing evaluation suites for LLMs that underpin many AI systems.

Last month also saw the EU signing the world’s first comprehensive law to govern AI. According to the EU AI Act’s final text, the law aims to promote the “uptake of human-centric and trustworthy AI, while ensuring a high level of protection for health, safety, fundamental rights, and environmental protection against harmful effects of artificial intelligence systems.”

To that end, IBM has developed an EU AI Risk Assessment capability within watsonx.governance to identify gaps in compliance with the EU AI Act, providing suggestions on how to remedy them, Gentile said.

The company is planning to use the same framework to incorporate regulatory requirements from other jurisdictions, including the US, UK, and India.

This framework, according to Gentile, will include a continuous monitoring capability to ensure models comply with regulations steadily.

What is IBM’s strategy with watsonx.governance?

Although it was launched with the watsonx platform, IBM plans to sell watsonx.governance as a separate offering.

“Our main roadmap initiative is AI everywhere,” Gentile said, adding that the toolkit can be accessed on any model across any platform or framework as its implementation or integration doesn’t require an enterprise to “rip and replace” their current technology.

“We have the ability to govern any third-party or open-source model wherever it resides, including on-premises … or public and private cloud,” Gentile said, explaining that IBM’s SDKs and APIs facilitate this.

Platform support for watsonx.governance includes Amazon SageMaker, Amazon Bedrock, Google Vertex, and Microsoft Azure.

IBM’s go-to-market strategy for watsonx.governance may attract enterprises, said analysts Keith Kirkpatrick, of The Futurum Group, and Maribel Lopez, of Lopez Research.

“Having a cross-cloud AI governance strategy makes IBM more attractive to CIOs and administrators because it’s a multicloud world,” Lopez said. “Having said that, buyers will still look to their existing cloud providers for governance as well. In some cases, IBM could offer an overlay audit-like function to make sure that the other cloud’s governance is up to par.”

Governance discussions, according to Lopez and Kirkpatrick, will keep IBM in the loop with buyers, which is essential in a competitive market.

IBM’s focus on governance and the modularity of its watsonx offerings provide the vendor an edge over the others, Lopez added. “The flexibility provides IBM the opportunity to sell a specific product to a buyer that may be using another party for their primary cloud services,” she said.

Eric Johnson, director at professional services firm West Monroe, agreed, saying IBM’s toolkit may gain a market advantage due to its “impartial” nature.

watsonx.governance is generally available in two tiers — a free trial and an Essentials tier for individuals and proofs of concept, priced at US$0.60 per resource unit. An enterprise/production tier is slated for future availability.

SUBSCRIBE TO OUR NEWSLETTER