Microsoft and Amazon Launch Consumer Health AI Ahead of Independent Safety Validation
The Development
Microsoft launched Copilot Health on March 18, adding a dedicated health space inside its Copilot app where users can connect medical records and ask clinical questions. Two days prior, Amazon announced its Health AI tool—previously exclusive to One Medical members—would become broadly available. Both join OpenAI's ChatGPT Health, released in January. The scale of demand is not in question: Microsoft reports receiving 50 million health-related questions daily, with health ranking as the top topic on its Copilot mobile app. OpenAI's own HealthBench shows GPT-5 outperforming prior models, and Stanford's MedHELM currently ranks GPT-5 highest among evaluated models. However, a Mount Sinai study published this month found ChatGPT Health sometimes over-recommends care for mild conditions and misses emergencies. All six independent academic researchers interviewed by MIT Tech Review raised concerns that these products are launching before external safety validation is complete.
Our Take
Three major platforms now carry health AI products to mass audiences, and the evaluation infrastructure to validate them doesn't exist yet at scale. Companies are citing internal benchmarks—HealthBench was designed and released by OpenAI—while independent frameworks like Stanford's MedHELM have documented limitations around multi-turn conversations. Google's AMIE passed rigorous human-trial standards but isn't being released precisely because of remaining safety gaps. The competitive pressure is obvious: if 50 million daily health queries are already happening on a general-purpose chatbot, a branded health product captures that intent and the data it generates. But for marketers building healthcare campaigns on or adjacent to these platforms, the liability picture is murky. Brands placing health-category advertising near AI-generated medical guidance—or co-branding with these tools—are doing so without third-party confirmation that the underlying product performs safely across diverse user populations.
The Exploit
Action Item
If your brand runs health or pharma campaigns on Microsoft or Amazon properties, audit your adjacency settings in DV360 and Amazon DSP before Q2 2026 upfront commitments lock in placements near Copilot Health and Health AI surfaces.