OpenAI, Google, and Anthropic Unveil Specialised Medical AI Capabilities
Earlier this month, OpenAI, Google, and Anthropic, three tech giants, made headlines by announcing specialised medical AI capabilities within days of each other. This simultaneous launch hints more at competitive pressure than mere coincidence. However, none of these releases have yet been approved as medical devices or cleared for clinical use. Moreover, they are not yet available for direct patient diagnosis, despite their marketing language suggesting a transformation in healthcare.
What Do These New Capabilities Offer?
OpenAI kicked off the announcements on January 7, introducing ChatGPT Health. This system enables U.S. users to connect their medical records through partnerships with b.well, Apple Health, Function, and MyFitnessPal. Google followed suit, releasing MedGemma 1.5 on January 13, which expands its open medical AI model to interpret three-dimensional CT and MRI scans. This also includes whole-slide histopathology images. Anthropic wasn’t far behind, announcing Claude for Healthcare on January 11. This offers HIPAA-compliant connectors to CMS coverage databases, ICD-10 coding systems, and the National Provider Identifier Registry.
All three companies are targeting the same issues in the workflow—prior authorisation reviews, claims processing, and clinical documentation. They are doing so using similar technical approaches but with different go-to-market strategies.
Developer Platforms, Not Diagnostic Products
Despite the different deployment and access models, all these systems have notable architectural similarities. They all use large language models, fine-tuned on medical literature and clinical datasets. Privacy protections and regulatory disclaimers are prominently emphasised, and each system is positioned as a support to clinical judgement, not a replacement for it.
Benchmark Performance vs Clinical Validation
There have been substantial improvements in medical AI benchmark results across all three releases. However, the gap between test performance and real-world clinical deployment remains significant. These benchmarks measure performance on curated test datasets, not clinical outcomes in practice. Translating benchmark accuracy into clinical utility is a complex process, particularly in medical AI where errors can have life-threatening consequences.
Regulatory Pathway Remains Unclear
The current regulatory framework for medical AI tools is ambiguous, particularly in the U.S where the FDA’s oversight depends on the intended use of the software. As yet, none of these tools have received FDA clearance. Liability questions are equally unresolved, with existing case law providing limited guidance in the event of a patient suffering harm due to delayed care informed by these AI tools. This regulatory ambiguity affects adoption timelines in markets where healthcare infrastructure gaps might otherwise accelerate implementation.
Administrative Workflows, Not Clinical Decisions
Despite the potential of these tools to dramatically impact patient outcomes, their real-world application is currently focused more on administrative workflows – such as billing, documentation, and protocol drafting – rather than direct clinical decision support. This is likely a strategic move to focus on areas where errors are less immediately dangerous.
While the technology for sophisticated medical reasoning tools is available, and accessible for as low as $20 monthly subscriptions, the transformation of healthcare delivery heavily depends on addressing regulatory, liability, and workflow integration complexities.
For more information, check out the source here.
