Lifebit launches AI Scientist Studio for governed health data

Lifebit said it launched AI Scientist Studio at HumanX Amsterdam. Researchers can build and deploy custom AI scientist agents inside secure Federated Trusted Research Environments, so frontier models run where governed health and genomic data already live instead of shipping patient-level records out to third-party AI tools.

HEALTH desk — a national biobank or a health ministry will not paste a patient’s genome into a chatbot that sits in someone else’s cloud, because the rules that already lock that record forbid the move. Lifebit’s announcement puts the agent inside the same Trusted Research Environment that holds the data, so the other choices are to skip a top-end model or to ship the patient-level record out.

What the lede says the launch is. Lifebit, called a leading provider of secure biomedical data platforms and Trusted Research Environments, has announced the launch of AI Scientist Studio at HumanX Amsterdam. A Trusted Research Environment, or TRE, is a locked workspace where researchers analyze sensitive health data under rules, instead of downloading the records. Federated, here, means the work runs at each place that holds a copy, and the patient-level rows stay there. Those glosses are this desk’s. Leading is the release’s adjective. The new platform enables researchers to build and deploy custom AI scientist agents within secure Federated TREs, so organizations can use frontier AI models on governed health data while ensuring data never leaves the source. A frontier model, here, is a top-end general model from a major lab. An agent, here, is software that can take the next step on a research task, not only answer one question. Those glosses are this desk’s. Rather than requiring researchers to transfer data into external AI environments, the release says Lifebit brings AI to the data, so analysis takes place where governed datasets already reside. Those sentences are the release’s. This desk did not move a record.

Dr Maria Dunford, as a quote, not as a survey of who can use which tool. Dr Maria Dunford, CEO of Lifebit, said AI Scientist tools are here, and they are the future of health data analysis, but the world’s most important health data still cannot use them. She said researchers are constrained by pre-built tools, while existing AI Scientists require sensitive data to move outside governed environments. She said AI Scientist Studio changes this by bringing any AI model to the data, enabling researchers to build their own AI Scientists and run them securely, in a federated manner, across health and genomic data without compromising control, sovereignty, or oversight. The future, any model, and without compromising are her sentences. They are not a test this desk ran. Sovereignty, here, is her word for keeping control of the data and the machines. That gloss is this desk’s. This desk did not interview her.

The two barriers the next paragraph names, and whose reach that is. The release says AI scientist platforms are proliferating, but the world’s most valuable health data remains largely out of reach. National health systems, biobanks, and pharmaceutical companies operate under strict privacy and governance requirements that prevent patient-level data from moving to third-party AI Scientist tools. A biobank, here, is a store of biological samples and the records that go with them. Patient-level means the row is about a person, not a total. Those glosses are this desk’s. Researchers are also constrained by pre-built tools they cannot easily adapt or replace. The release says AI Scientist Studio removes both barriers: researchers can build their own best-in-class AI Scientists and bring the AI to the data, not the data to the AI. Best-in-class is the release’s adjective. It is not a ranking this desk made. Largely out of reach, and removes both barriers, are the release’s claims. This desk did not audit a health system’s rules.

What the platform is built on, and which jobs the release names. Built on Lifebit’s Federated TRE architecture, AI Scientist Studio applies the Five Safes framework to agentic AI. Agentic AI, here, means AI that takes actions, not only a chat reply. The Five Safes is a checklist used for access to sensitive data. The release names the framework and does not print the five words. Those glosses are this desk’s. Researchers can bring their own or choose any foundation model, scientific tool, and knowledge base, then build specialist agents for cohort discovery, deep phenotyping, analysis, and evidence generation. A foundation model is a large general model a lab already trained. A knowledge base is reference material the agent can use. A cohort is a defined group of patients. Deep phenotyping is a detailed read of a patient’s traits from records and tests. Those glosses are this desk’s. The four jobs are the release’s list. The release does not print product names for those agents. This desk did not build one.

Where the agent runs, and what is allowed to leave. Agents execute on compute inside each data custodian’s governed environment, inheriting user permissions and operating federated across sites. A custodian, here, is the organization that holds the records and sets the rules. Inheriting user permissions means the agent can see only what that signed-in person can see. Those glosses are this desk’s. Patient-level data never moves. Chat history, outputs, and complete audit trails remain within the environment, and only approved results leave through Lifebit’s Airlock controls. An audit trail is the log of what was asked and what came back. Airlock is the release’s name for the control that lets an approved result out and keeps the rest inside. Those glosses are this desk’s. Never moves, complete, and only approved are the release’s words. This desk did not open a log.

The second Dunford quote, and the sentence about the meeting. Dunford added that as governments, health systems, and life sciences organizations adopt frontier AI, they increasingly face the challenge of doing so without surrendering sovereignty over sensitive data and infrastructure. She said major AI vendors have made significant moves into scientific research this year, but the prevailing model still assumes that data comes to the AI. She said Lifebit is inverting that model, bringing frontier AI directly to sensitive health and genomic data. Significant moves, prevailing model, and inverting are her sentences. The release does not name those vendors. Do not invent them. The next sentence says Dr Maria Dunford, CEO of Lifebit, will be discussing AI Scientist Studio at HumanX Amsterdam between 22-24 September, exploring the growing role of AI scientist agents in biomedical research and how organizations can apply frontier AI within governed health-data environments. The lede says the launch is at HumanX Amsterdam. Will be discussing, and the 22-24 September window, are this sentence’s. On a release stamped 07:00 GMT on 24 Sep, that window includes days already on the calendar and that same day. This desk did not treat the sentence as a transcript, and it did not attend.

Who the release says already uses the infrastructure, and whose count 275 million is. AI Scientist Studio builds on Lifebit’s existing federated secure infrastructure, already used by organizations including the UK’s Genomics England, the UK’s NIHR BioResource, Canada’s Canpath, Singapore’s Ministry of Health, and Boehringer Ingelheim, among others. Including, and among others, mean the release is not offering a complete list. NIHR is the National Institute for Health and Care Research. That expansion is this desk’s. The release prints NIHR and does not spell it out. The wire spells Canpath with a lowercase p. Lifebit’s platform supports the secure management and analysis of more than 275 million patient records across 30+ countries, providing a distributed foundation for governments, health systems, biobanks, and life sciences organizations to apply advanced AI to sensitive biomedical data at scale. More than 275 million is the company’s figure for records on the platform. Thirty-plus is more than 30 countries. Those translations are this desk’s. This desk did not count the records or the countries.

Who the about box says the company is, and who to call. Lifebit provides secure, cloud-native technology that enables organizations to unify, govern, and analyse biomedical data at scale. Analyse is the about box’s spelling. Cloud-native, here, means the software is built to run in a cloud service. That gloss is this desk’s. Its Trusted Research Environment platform supports leading public and private sector organizations to accelerate discovery while maintaining the highest standards of security, compliance, and data governance. Highest standards, and leading, are the about box’s words. They are not a certificate this desk checked. The author contact is Thorben Seeger, Chief Business Development Officer, Lifebit, thorben@lifebit.ai. schema.org names that same person as author. This desk did not email that address.

What the card shows, and what it does not. The card is the EIN launch graphic: a dark navy field, a Lifebit wordmark, and the stacked headline this desk reads as “The World’s Most Important Health Data Can’t Use Frontier AI. Until Now: Lifebit Launches AI Scientist Studio.” The right side is an abstract mark. It is not a product screen. The card has no date text. The release’s sentences about the Five Safes, Airlock, the four research jobs, the customer names, and the 275 million figure are not printed as labels on this card. Do not read the mark as a diagram of those controls. The schema image on the wire is a medium JPEG at img.einnews.com, 300 by 168. The page’s og:image meta names a Facebook-sized PNG. This filing does not ship those two files. The large PNG at img.einpresswire.com is the source image. Its last-modified header was Wed, 23 Sep 2026 16:14:11 GMT, the day before the stamp. That header is the image file’s. It is not a second announcement.

What the product page adds, and what it does not let you file as the wire. The page at lifebit.ai/ai-scientist-studio/, read 25 Sep 2026, does not print HumanX, 07:00 GMT, the Dunford quotes, NIHR BioResource, Canpath, Singapore’s Ministry of Health, or Boehringer Ingelheim. A line says AI Scientists that work on the data that cannot move, and that you deploy AI Scientists you build, inside your secure environment, with any model, and data never leaves the source. A line says trusted by the custodians of 275M+ patient records across 30+ countries. The wire’s figure is more than 275 million patient records across 30+ countries. The page’s line says custodians of those records. Do not collapse the two sentences into one audit. The page names Cohort Scientist, Feasibility Scientist, Target Scientist, Evidence Scientist, Safety Scientist, Phenotyping Scientist, Harmonization Scientist, and a build-your-own path. The wire’s list is cohort discovery, deep phenotyping, analysis, and evidence generation. It does not print those product names. A feasibility line says counts across sites, no data movement, and minutes, not months. A models line names GPT, Gemini, Claude, open models, or your own. The wire says bring your own or choose any foundation model. It does not print those product names. A Five Safes line says Safe Settings, Safe Outputs, and Safe Projects hold when the researcher is an agent. That is three labels. The wire names the framework and does not print them. A line says one blocked workflow, your data, your model, running in 6 to 8 weeks, and live in weeks, not quarters. Another line says the scientist you describe is ready in minutes. Six to eight weeks, and minutes, are both on this page. The wire prints neither clock. A frequently asked question says it launches with early-access founding partners this quarter. The wire says Lifebit has announced the launch. It does not say early access, founding partners, or this quarter. The page does not name the quarter. A quote from Dr. Arun Sujenthiran of Flatiron Health, and a Genomics England case-study line from Chris Wigley, former CEO, are on this site. They are not sentences in the 24 Sep release. The footer prints a London address at 207 Regent Street and a New York address at 228 East 45th Street, hello@lifebit.ai, and a list of security marks including FedRAMP Authorized, ISO/IEC 27001, SOC 2 Type II, HITRUST, HIPAA, GDPR, Cyber Essentials Plus, and NHS DSPT. Those addresses and marks are the page’s. The 24 Sep release does not print them. This desk did not book a demo.

What the newsroom card adds, and which older rows it does not let you file as today. The news index at lifebit.ai/news/, read 25 Sep 2026, leads with a press-release card: Sep 24, 2026, announced at HumanX Amsterdam, the same headline, and the model-agnostic line that data never leaves the source. Read the release goes to the EIN article. That card is a pointer to the wire. It is not a second announcement with new numbers. Media inquiries on the newsroom are hello@lifebit.ai. The wire’s contact is thorben@lifebit.ai. Do not collapse the two addresses. Lower on the same index, this desk read older rows: 18 Sep 2026, Digital Health, Lifebit to provide a research platform for NIHR BioResource; 3 Sep 2026, Contract Pharma, Boehringer Ingelheim and Lifebit extend a clinical data partnership; 2 Jul 2026, Genomics England selects Lifebit; 28 May 2026, a press release that Lifebit and CanPath establish a Canadian framework. The index spells CanPath with a capital P. The 24 Sep wire spells Canpath with a lowercase p. Those older rows are not sentences in today’s release. This desk did not open those earlier articles.

Plain English for the rest of the card. A Trusted Research Environment is the locked room where the records already sit. Federated means the model runs in that room, at each site, instead of the rows being copied out. An agent is software that can do the next research step. Airlock is the door that lets an approved result out and keeps the chat, the outputs, and the log inside. The Five Safes is the checklist the release says it applies. The release does not list the five. The jobs on the wire are finding a group of patients, describing them in detail, running the analysis, and writing up the evidence. More than 275 million is Lifebit’s figure for patient records on the platform, across more than 30 countries. The names on the wire are Genomics England, NIHR BioResource, Canpath, Singapore’s Ministry of Health, and Boehringer Ingelheim, plus others the release does not name. 07:00 GMT is 7:00 a.m. UTC and 3:00 a.m. Eastern. 07:47:18 UTC is the page’s modified clock, forty-seven minutes and eighteen seconds later. Dr Maria Dunford is chief executive. Thorben Seeger is chief business development officer. The card is the launch headline. This filing is the 24 Sep announcement.

PRIMARY here: Lifebit’s 24 Sep 2026 EIN Presswire, article 944250050, stamped September 24, 2026, 07:00 GMT, datelined London — Tier A PRIMARY, the company’s own announcement. STATUS PRIMARY. The launch at HumanX Amsterdam, the Federated TRE sentence, both Dunford quotes, the two-barriers paragraph, the Five Safes name, bring-your-own or any foundation model, the four jobs, permissions, Airlock, the 22-24 September discussion sentence, Genomics England, NIHR BioResource, Canpath, Singapore’s Ministry of Health, Boehringer Ingelheim, more than 275 million patient records, 30+ countries, the about box, and thorben@lifebit.ai are that release’s. The named scientist products, GPT, Gemini, Claude, the three Five Safes labels, 6 to 8 weeks, minutes, early-access founding partners this quarter, the Flatiron quote, the Wigley line, the street addresses, the security-mark list, and hello@lifebit.ai are the product page’s or the newsroom’s. The newsroom’s 18 Sep, 3 Sep, 2 Jul, and 28 May rows are older items. NOT claimed: that this desk attended HumanX, that the discussion sentence is a transcript, that the vendors in the second quote were named, that 275 million or 30 countries was counted here, that Canpath and CanPath were reconciled, that early access this quarter is a sentence on the wire, that the product names or the model names are on the wire, that the card’s abstract mark is an Airlock diagram, a login, a demo this desk booked, a stock tip, or investment advice. The card is the EIN launch graphic. Distinct from the already-filed hims-ai-weight-loss, basalt-health-series-a, trilliant-health-oria, and veeva-study-builder-agent.