The acquisition brings Torch’s four-person team into OpenAI and folds its core technology straight into the health product unveiled in January 2026. 

Torch gives OpenAI a ready-made system for pulling together scattered medical data at a moment when the company wants to enhance its focus on personal health tools.

Torch had been building what it described as “a medical memory for AI, unifying scattered records into a context engine.” The idea is to take health information spread across clinics, labs, wearables and wellness apps, and make it usable in one place. 

That work now sits at the heart of ChatGPT Health, which allows users to securely link medical records and daily health data inside the chatbot.

While OpenAI did not disclose the price, reports vary. Some put the value near $100 million in equity, others closer to $60 million. Either way, the structure points to an acqui-hire. The team joins; the product becomes infrastructure.

This development lands just over a year after a very different ending for the same founders. Torch’s team met while working at Forward Health, a high-profile clinic startup built around automated care. 

Forward raised close to $400 million before shutting down abruptly in late 2024, laying off staff and closing its doors. Torch’s sale shows how fast fortunes can turn in health technology, where ideas outlive companies.

ChatGPT Health itself is standing carefully. OpenAI says it is a secure, separate space within ChatGPT, designed to help people organise information, prepare questions and understand records, not to replace doctors. More than 260 physicians were involved in building safeguards around how responses are delivered.

With Torch in-house, OpenAI wants to solve one of the hardest problems in digital healthcare; making sense of messy, fragmented data without losing context or trust. 

The product, called Claude for Healthcare, is designed to take on the heavy administrative and research workload that slows down medical care, while also giving patients a better way to understand their own health data. 

At its core, the new product allows healthcare organisations to connect Claude to industry databases. These connections let the system search, verify and organise medical and policy information that clinicians and insurers usually spend hours tracking down. 

Anthropic says this will reduce delays in processes such as prior authorisation, where doctors must justify treatments before insurers approve payment.

“Clinicians often report spending more time on documentation and paperwork than actually seeing patients,” Mike Krieger, Anthropic chief product officer said.

Rather than replacing doctors, the company is pushing Claude into the background work that clogs up healthcare. Submitting forms, checking coverage rules, matching diagnosis codes and assembling appeal documents are all tasks the system is meant to handle faster. 

For insurers and providers, this could mean quicker decisions and fewer backlogs. For patients, it could mean less waiting.

Claude for Healthcare connects directly to systems such as the US Centres for Medicare and Medicaid Services coverage database, the ICD-10 coding system, the National Provider Identifier registry and PubMed’s vast research library. 

With these links in place, the system can cross-check policy regulations against patient records, flag missing information and prepare reports that staff would normally compile by hand.

Anthropic is also adding specialised features aimed at interoperability and approvals. One tool focuses on FHIR, the standard used to move data between healthcare systems, while another provides a configurable template for prior authorisation reviews. 

The idea is to reduce errors that creep in when staff juggle multiple platforms under time pressure.

Beyond administration, the company is adding to Claude’s functions in life sciences. New integrations are intended to support clinical trials, regulatory submissions and research analysis, areas where speed and accuracy can tell whether new treatments reach the public sooner.

Patients are part of the plan too. In the US, subscribers on higher-tier plans can choose to connect Claude to personal health records, lab results and fitness data through platforms such as Apple Health and Android Health Connect. 

When enabled, the system can summarise medical history, explain test results in plain language and help users prepare questions for their doctors. Anthropic stresses that users stay in control of what is shared and that health data is not used to train its models.

The development comes as large numbers of people already turn to conversational systems to discuss health issues. OpenAI has said that around 230 million people talk about their health with ChatGPT each week, a fact that patients are filling gaps where access to clinicians is limited. 

Both companies, however, warn that these tools are not a substitute for professional medical advice.

A direct response to how people already use the service, OpenAI says more than 230 million users ask health and wellness questions every week. 

Until now, those conversations sat beside everyday chats. With Health, they are ring-fenced. The company says this separation is meant to stop sensitive health details from appearing in unrelated discussions, while still allowing users to return to them when needed.

The Health section operates as its own environment. If someone begins discussing a medical concern in a regular chat, the system is designed to prompt a move into Health, where added privacy applies. 

At the same time, limited context from general chats, such as lifestyle habits or fitness goals, may be used to make health discussions more relevant. The flow works one way only. Health information does not feed back into standard conversations.

A major part of the rollout is data connection. Users will be able to link medical records and wellness apps, including Apple Health, Function and MyFitnessPal, so conversations are grounded in personal information rather than general advice. 

OpenAI says these health chats will not be used to train its models, and that the feature uses extra layers of encryption and isolation because of the sensitivity of the data involved.

Fidji Simo, OpenAI’s chief executive for applications, described the product as an answer to long-standing problems in healthcare, including high costs, limited access, overbooked doctors and poor continuity of care. 

The aim, she said, is to help people feel more prepared and informed when dealing with their own health, not to replace medical professionals.

However, systems like ChatGPT generate responses based on patterns, not on an understanding of truth, and can sometimes produce inaccurate information. OpenAI acknowledges this risk. 

In its own terms, the company states that it is “not intended for use in the diagnosis or treatment of any health condition.”

ChatGPT Health has been developed with input from hundreds of doctors across dozens of specialities, according to OpenAI. Their feedback has impacted how the feature explains results, flags potential risks and encourages follow-up with clinicians when necessary. 

Even so, the company stresses that the tool is meant to support everyday understanding, not clinical decision-making.

Access will begin with a limited group of users in the coming weeks, with a wider rollout planned after further testing. Some integrations, including medical record connections, will initially be available only in the United States.