A Startup's Demise Highlights the FDA's AI Problem

Kintsugi, a California startup, spent seven years building artificial intelligence to identify depression and anxiety from speech patterns. This week, it closed its doors. The company, unable to secure FDA clearance before funding ran out, is releasing most of its technology to the public.

The core idea was to move beyond questionnaires. Kintsugi's software analyzed how a person speaks—pacing, tone, rhythm—to find vocal markers associated with mental health conditions. Published research indicated its performance was comparable to standard screening tools. The goal was to offer a scalable, objective check that could integrate into healthcare and workplace wellness programs.

That ambition hinged on regulatory approval. Kintsugi pursued a 'De Novo' classification for novel, low-risk devices. Founder Grace Chang described a lengthy process of educating regulators on AI, a system the FDA's framework isn't built for. Unlike a static surgical tool, AI models are designed to evolve, creating tension with rules for fixed devices. Government shutdowns added further delays.

As capital dwindled, the team rejected what Chang called predatory financing offers. Instead, they opted to open-source the technology. This move isn't without controversy. Experts note public models often lack the rigorous documentation regulators require, potentially complicating future approval efforts. There are also ethical questions about misuse outside clinical settings, though Chang believes the practical risk of abuse is lower than the risk the tool will simply go unused.

Not everything was released. The company is retaining some proprietary technology, including a system for detecting AI-generated or manipulated voices—an unexpected byproduct of its research with potential applications in security and fraud prevention. That field, notably, operates free of FDA oversight.

Kintsugi's story underscores a persistent mismatch between the pace of AI innovation and the timelines of medical regulation. Chang hopes others will continue the work, but without systemic adaptation, similar clashes seem inevitable.

Source: The Verge