Microsoft’s new responsible technology is leading the development of high-speed AI

Fully responsible, reliable technology is an almost impossible mandate in a speed-first technology environment – but that doesn’t mean some companies aren’t trying.

On the heels of the Trump administration’s national AI legislative framework on March 20, when “winning the AI ​​race” is still in the forefront, technology developers face the tension between the general principles of moving quickly and breaking things versus using strategies responsible technology frameworks from the beginning.

Moving forward has, in many cases, taken the driver’s seat, its costs already clear. Microsoft’s recognition that AI-generated code often sacrifices accessibility makes human oversight and replication a necessity.

For Jenny Lay-Flurrie, who became head of Microsoft’s Trusted Technology Group in February and has worked on access for most of her 21 years with the company, responsible development and deployment of technology is twofold: “How do we make sure we build it right? And how can we make sure it stays right?”

Microsoft launched the Trusted Technology Group in early 2025 and has since brought together all responsible technology initiatives under its umbrella, including Lay-Flurrie’s former accessibility directive.

While Microsoft has included its responsible technology under a top-down model, competitors such as Google maintain an engineering-led architecture led by core AI principles and specialized security councils. Strategies vary across major technologies, but Microsoft’s approach is one that has been reshaped since 2002, when Bill Gates issued a Trustworthy Computing memo that prioritized things like reliability over new feature development.

AI problems (and the people who fix them)

Lay-Flurrie’s foray into the more responsible tech space may be recent, but she says it follows the same general principles she’s familiar with, including fairness, transparency, inclusiveness and accountability. Microsoft works with the principle that “people should be responsible for AI” regardless of its results.

That’s why, when Microsoft realized that its AI was not properly representing blind people, its team immediately fixed the problem.

“Some of the pictures of blind people that were made came back with people wearing these ugly holes full of eyes,” she said. “These models were trained on a lot of things that exist in the community. Unfortunately, the community is not always the most cohesive place, so there are situations where we have to input data to train it.”

To do this, Microsoft purchased more than 20 million minutes of multimodal data from Be My Eyes, a non-profit accessibility platform that blind and partially sighted people can use for free to interact with live volunteers and AI, giving them an audio understanding of what they see. “They had a lot of videos of blind people using them with walking sticks and dogs and finding keys to the house, we ended up not revealing the details by blurring the faces and all that so we could train our models properly about blindness,” said Lay-Flurrie.

The process is rigorous, but Annie Brown, CEO and founder of Reliabl, a machine learning training software that works to reduce bias and increase the performance of AI models, said there is room for improvement.

“A lot of different details are part of it,” Brown said. “If you’re not paying attention to what’s going on in the metadata layer, which is how those images uploaded to your dataset are labeled, that’s going to create bias.”

Outside of the world-changing AI race, Microsoft is part of a broader movement of companies that are publicly sharing their responsible technology learning. Microsoft Learn is available for free to students, academics and developers and includes training modules on responsible AI principles and more. Brown recommends Microsoft learn from small social good organizations to see “how they bring inclusivity to AI.”

As for improvisation, Lay-Flurrie says it comes with the territory. “It’s clearly listening to the feedback, getting that, iterating, evaluating and solving those in the shortest possible time,” he said.

People vs. AI

Microsoft is a leading provider of business technology, which means that its own AI is fueling other companies that often make the decision to cut employees in lieu of advanced solutions. Microsoft itself is part of a broader wave of major tech layoffs, though it’s being characterized as a realignment of priorities rather than a mere replacement. The company has cut an estimated 15,000 jobs in sales, gaming and customer-facing divisions by 2025 and hired new staff elsewhere to focus on AI infrastructure.

Even as layoffs continue across industries, Lay-Flurrie says AI is already leveling the playing field for previously disadvantaged workers, including those with disabilities.

“The first community that had access to Copilot at Microsoft was our group of employees with disabilities,” he said. “In the Deaf community, captions, texts, meeting notes, sign language recognition, that gives independence. You don’t have to wait for a cartographer to be there to record what’s being said.”

In the neurodiverse community that discovered Copilot early on, it helped so much with the mental load that “they wouldn’t let me get the license back,” he said.

Diego Mariscal, CEO and founder of 2Gether-International (2GI), run by entrepreneurs with disabilities, recognizes that Microsoft has made a point of including people with disabilities. “The fact that Jenny’s position exists even at this level is proof of that,” he said. However, including people with disabilities at the decision-making table is important from the top down. “How can we ensure that, as AI evolves, disabled people are brought to the table, not from a charity perspective, but because doing so will ensure that technology and innovation are at the forefront and accessible to everyone?”