Kintsugi shuts down: voice-based depression detection startup fails to obtain FDA approval

California-based startup Kintsugi is shutting down after seven years of developing artificial intelligence capable of detecting signs of depression and anxiety disorder from patient speech. The reason for closure is the inability to obtain timely approval from the U.S.

Food and Drug Administration (FDA). Most of the technology will be made publicly available. The problem Kintsugi was addressing is real and large-scale.

Mental health disorder diagnosis still relies primarily on questionnaires and clinical interviews — unlike physical diseases, where laboratory tests and imaging have long been used. The startup proposed a different approach: instead of analyzing speech content, it studied how a person speaks — pace, pauses, voice modulations, intonation patterns. These acoustic markers, according to the company's hypothesis, correlate with depression and anxiety.

The technology looked promising: voice analysis could be conducted remotely, quickly, and without special equipment. This opened possibilities for primary screening in telemedicine, monitoring patient status between doctor visits, and providing support in regions with acute shortages of psychiatrists. The company attracted investment and spent years refining algorithms.

However, medical regulations in the U.S. represent one of the most difficult barriers for AI startups.

The FDA requires clinical evidence of effectiveness and safety, and the approval process can stretch for years. For Kintsugi, the combination of regulatory timelines and, apparently, financial constraints proved critical. Nevertheless, the story does not end in complete failure.

Public disclosure of the technology gives it a chance at a second life. According to The Verge, some developments could be applied outside medicine — specifically, for detecting synthetic audio and deepfakes. This direction is rapidly gaining relevance as the quality of AI voice generation grows.

The Kintsugi case clearly demonstrates a key contradiction in medical AI: technology may work, but proving it to regulators within the required timeframe and budget is achievable only by a few. Voice-based diagnosis of mental disorders as a field will not disappear — it will continue to be developed by other players, including those relying on Kintsugi's now-open research.