In a technology industry that prioritizes speed, delivering fully responsible and reliable technology is a near-impossible imperative. But that doesn’t mean some companies aren’t trying.
Following the Trump administration’s March 20 enactment of a national AI legislative framework that prioritizes “winning the AI race,” technology developers are facing a tension between a shared ethos to move quickly and break things and strategically implementing responsible technology frameworks from the outset.
In many cases, moving forward has taken over the driver’s seat, and the cost is becoming clear. Microsoft’s own recognition that AI-generated code often abandons accessibility makes human oversight and iteration mandatory.
For Jenny Lay-Flurrie, who took over as head of Microsoft’s Trusted Technology Group in February and has worked on accessibility for most of her 21 years at the company, the responsible development and deployment of technology is two-fold. “How do you make sure you build it correctly? And how do you make sure it’s maintained correctly?”
Microsoft launched the Trusted Technology Group in early 2025 and has since brought all of its responsible technology efforts under its umbrella, including Lay-Flurrie’s previous accessibility directive.
While Microsoft centralizes responsible technology under a top-down model, competitors such as Google maintain a more engineering-driven architecture based on core AI principles and dedicated safety councils. Techniques vary among major technology companies, but Microsoft’s approach has been reshaped since 2002, when Bill Gates released his Trustworthy Computing memo, which prioritized things like reliability over the development of new features.
The problem with AI (and the people solving it)
Lay-Flurrie’s foray into the broader field of responsible technology may be recent, but she says she follows the same general principles as before: fairness, transparency, inclusivity and accountability. Microsoft operates on the principle that humans are responsible for AI, regardless of the outcome of the AI.
That’s why when Microsoft realized its AI wasn’t accurately representing visually impaired people, her team set out to solve the problem.
“In the images of blind people that were generated, we had these scary, completely blind people coming back,” she says. “These models were trained using a lot of material that exists in society. Unfortunately, society isn’t always the most inclusive place, so in some cases you need to inject data to train the society.”
To accomplish this, Microsoft purchased over 20 million minutes of multimodal data from Be My Eyes, a nonprofit accessibility platform. The platform is free to use for visually impaired people to connect with live volunteers and AI to provide audio insights into what they’re seeing. “We had a lot of video footage of visually impaired people using canes, dogs, and finding keys in their homes. We anonymized the data by blurring things like faces so we could better train the model about visual impairments,” Ray-Fleury said.
While this process is robust, there is room for improvement, says Annie Brown, CEO and founder of machine learning training software Reliabl, which works to minimize bias and maximize performance in AI models.
“More diverse data is just one part of it,” Brown said. “If you don’t pay attention to what’s happening at the metadata layer, how images are labeled as they’re uploaded to your dataset, you’re creating bias in and of itself.”
Despite the AI race changing the world, Microsoft is part of a broader movement of companies publicly sharing responsible technology learning. Microsoft Learn is free for students, academics, and developers and includes training modules on responsible AI principles and more. Brown also recommends learning from smaller social good organizations to see how Microsoft is “bringing inclusivity to AI.”
As for improvements, Ray Fleury said it comes with the region. “We are clearly listening to feedback, taking it, iterating, testing, and resolving it in the shortest possible time,” she said.
Human vs. AI
Microsoft is a leading provider of enterprise technology. That means Microsoft’s own AI is inspiring other companies, which often make decisions to cut employees instead of advanced solutions. Microsoft itself is part of a broader wave of large-scale tech layoffs, but it’s clear that it’s more of a reprioritization than a complete replacement. The company cut about 15,000 jobs in sales, gaming, and customer care in 2025, and hired new talent elsewhere to focus on AI infrastructure.
Ray-Fleury says AI is already leveling the playing field for previously marginalized workers, including those with neurodiversity and disabilities, even as layoffs continue across the industry.
“The first community to access Copilot at Microsoft was our disability employee group,” she said. “For the deaf community, captions, transcripts, meeting notes, and sign language recognition provide independence. They don’t have to wait for a cartographer to be there to transcribe what is said.”
For the neurodiverse community that received CoPilot early on, it was so effective at reducing cognitive load that they “didn’t let us get our licenses back,” she said.
Diego Mariscal, CEO and founder of 2Gether-International (2GI), a global startup accelerator run by and for entrepreneurs with disabilities, acknowledges that Microsoft values the participation of people with disabilities. “The fact that Jenny’s status exists at this level is proof of that,” he said. Still, including people with disabilities in decision-making is important, both top-down and bottom-up. “As AI evolves, how do we make sure that people with disabilities can participate, not just from a philanthropic perspective, but so that the technology and innovation becomes even more cutting-edge and accessible to everyone?”
