In February 2025, Jenny Lay-Flurrie took the helm of Microsoft's newly launched Trusted Technology Group, a centralized effort to weave responsible AI practices into the company's development pipeline from the ground up. For Lay-Flurrie, who has spent 21 years at Microsoft working in accessibility, the mandate is clear and dual-sided: "How do we make sure that we build it right? And how can we make sure that it stays right?"
The question reflects a real tension that defines AI development today. While governments like the Trump administration prioritize winning the AI race, companies face relentless pressure to move fast and iterate quickly. Yet speed without care leaves real damage. Microsoft's own discovery that AI-generated code often lacks accessibility features became a turning point—a moment when the company realized that responsible practices aren't a marketing feature but a structural necessity.
The Trusted Technology Group consolidates all of Microsoft's responsible tech initiatives under one umbrella, bringing together accessibility work, fairness protocols, transparency standards, and accountability measures. Lay-Flurrie grounds this approach in four principles: fairness, transparency, inclusiveness, and accountability. "People should be accountable for AI," she emphasizes, "regardless of its outcomes." It's a philosophy that traces back to Bill Gates's 2002 Trustworthy Computing memo, which chose reliability over rapid feature development—a principle the company is now reviving for the AI era.
The real test of these commitments shows up in the details. When Microsoft discovered that its AI models were generating imagery of blind people wearing "horrible full-on blindfolds"—a crude, dehumanizing stereotype—the company didn't brush it aside. Instead, it took action. The problem stemmed from training data: the models were learning from the biases embedded in society itself. To correct course, Microsoft purchased more than 20 million minutes of multimodal data from Be My Eyes, a nonprofit platform that blind and low-vision individuals use to connect with sighted volunteers and AI assistance. The data included video footage shot by blind people navigating everyday life with canes and dogs, finding keys in homes, and moving through spaces independently. Microsoft anonymized faces in this footage and used it to retrain its models toward more authentic and respectful representation.
It's painstaking work, but incomplete. Annie Brown, CEO of Reliabl, a machine learning software company, notes that diverse data alone isn't enough. "If you don't pay attention to what's happening at the metadata layer, which is how those images that were uploaded to your data set are labeled, that itself is going to create bias," she said. The conversation is evolving beyond volume toward vigilance.
What sets Microsoft apart is its willingness to share these learnings publicly. Microsoft Learn offers free training modules on responsible AI principles to students, academics, and developers worldwide. Lay-Flurrie sees improvement itself as continuous work: "It's listening clearly to the feedback, receiving that, iterating, testing and resolving those within as short of a period of time as we can."
There's an irony worth noting: as Microsoft invests in responsible AI, the company has also undergone significant restructuring, cutting roughly 15,000 jobs in sales, gaming, and customer-facing divisions in 2025. Yet Lay-Flurrie points to an overlooked benefit: AI is already creating pathways for workers who've long faced barriers—neurodiverse and disabled employees among the first to access tools like Copilot at Microsoft. The technology that displaces some workers is, simultaneously, opening doors for others. It's a reminder that the question isn't whether AI will shape the future, but whether humans will shape how.