Generate:Biomedicines

The generative models are the origin of the molecules rather than a filter applied to them. The Generate Platform runs a continuous generate, build, measure and learn loop: machine learning models generate protein sequences to answer a specific therapeutic question, including antibodies against defined epitopes and proteins recoded to avoid immune recognition, the sequences are made as real proteins at scale, measured, and the data trains the next round.

The company says more than 42,000 proteins have been generated, built and tested, and that GB-0895, its lead antibody, was jointly optimized by the platform for potency, half life and dosing interval. The laboratory exists to measure designs and feed the models.

Every design the models produce is built as a real protein and measured before a program advances, in a loop the company describes as machine learning driving generative design with "human insight embedded at every layer", and the clinical stage assets then pass through regulatory and trial oversight: an IND "study may proceed" was granted for GB-4362 and GB-5267 in December 2025, and the Phase 3 and Phase 1 studies run under protocol and site ethics oversight.

A threshold or approval gate between a model's output and a program commitment is not published.

The approach is published in papers. Chroma, the company's generative protein model, was published in Nature in November 2023 with a named author list, and the code and model weights were released for academic and non profit use, so outside researchers can run it and report where it fails. The platform pages describe what the models do: generating protein complexes including antibodies against defined epitopes, optimizing arbitrary protein properties, and recoding a therapeutic protein to retain function while reducing immune recognition, inside a generate, build, measure and learn loop.

What is not published is which models are in production now, how they are versioned, or how a collaborator learns that the system changed; Chroma's 2023 release describes the method rather than the current configuration, and the weights available to academic users are not the commercial platform.

The chain is short: the models are the company's own, and one of them is public. Chroma was built in house, published in Nature and released with its weights for academic and non profit use, and the rest of the platform is operated internally alongside the company's own protein production, assay and structural biology systems.

The hosting arrangement and subprocessors behind the platform are not named.

A platform designed candidate is in a named Phase 3 study with a disclosed timeline, and the platform's method is published. GB-0895, an antibody against TSLP designed on the platform and engineered for twice yearly subcutaneous dosing, is in Phase 3 in severe asthma (the SOLAIRIA study, with its own trial site) and in a Phase 1 study in COPD; the first Phase 3 patient was dosed on 26 January 2026. GB-4362, an antibody that binds free MMAE to reduce the toxicity of antibody drug conjugates, and GB-5267, an IL-18 armored CAR T therapy against MUC16, received IND "study may proceed" in December 2025, and GB-4362 has FDA Fast Track designation. The method behind the designs is in the peer reviewed literature: Chroma in Nature, November 2023, with weights released for academic and non profit use. Collaborations with Amgen and Novartis cover confidential programs first announced in January 2022 and September 2024.

No efficacy results are published, and Phase 3 entry is not efficacy. The company's platform economics projections (time and cost to proof of concept against industry benchmarks) are its own estimates rather than realized results and are not evidence here.

Generate's privacy notice (last updated 21 October 2024) covers people who visit its sites, download its software or otherwise interact with it, states that the information collected can include sensitive information such as health information, and says clinical trial participants receive a separate notice at the time they consent. Retention is stated broadly (as long as necessary for the purpose, or for the period of the research and a reasonable period after), and security as reasonable physical, technical and administrative measures.

The discovery workflow runs on protein structures, sequences and assay data from the company's own laboratories; whether a collaborator's sequences or assay results train models that serve other programs is not stated.

A privacy policy covers the service, with no business associate statement. Generate's privacy notice (last updated 21 October 2024) covers people who visit its sites, download its software or otherwise interact with it, names health information among the sensitive information it may collect, and says clinical trial participants receive a separate notice when they consent to enroll.

No statement of business associate status appears anywhere, and no scope position explains whether the discovery platform handles protected health information; the separate trial notice, which is where participant data is addressed, is given at consent rather than published.

The controls are described by the company alone, with no outside audit or test named. No SOC 2 report, ISO 27001 certificate, penetration testing cadence, trust center or security page appears on the company's site, and the privacy notice offers only that it maintains reasonable physical, technical and administrative measures while stating that it cannot warrant the security of information.

The registration statement filed ahead of the February 2026 listing sets out cybersecurity risk factors and discloses that the company has experienced attacks, security incidents and system failures, none of which it believes caused a material interruption. Disclosing that is candor, and a risk factor is written to inform investors of exposure rather than to evidence controls to a counterparty. The annual cybersecurity description, covering how risk is assessed, identified and managed and how the board oversees it, attaches to an annual report the company has not yet filed.

Generate presents the Generate Platform as research technology, and its regulatory activity sits with the medicines rather than the software. GB-0895 is in Phase 3 in severe asthma and Phase 1 in COPD; GB-4362 and GB-5267 received IND "study may proceed" in December 2025, and GB-4362 carries FDA Fast Track designation, a development and review mechanism rather than an approval or a finding of efficacy.

The company also publishes its position on expanded access, stating that trial participation is currently the only route to its investigational therapies, and none of its products has been approved yet.

Generate published Chroma, its generative protein model, in Nature in November 2023 and released the code with model weights for academic and non profit use, and the platform pages describe the generate, build, measure and learn loop. The central domain question for a model learned from a public structural archive is where it performs unevenly across protein families or fold classes.

A governance framework, evaluation results and a breakdown of that kind are not published.

Designs are measured before programs advance, with nothing standing behind them. The platform is described as a continuous loop in which generated sequences are built as real proteins at scale and measured for the characteristics and functions that matter, with the data feeding the next round; the company reports more than 42,000 proteins generated, built and tested. Chroma's publication in Nature and the release of its weights to academic and non profit users let outside researchers run the model and report where it fails, which is traceability of the method rather than a commitment about an output.

No error rate, design success rate or accuracy figure is published with a method and denominator, and no warranty, indemnity or remediation obligation is stated; the platform is not sold, so a counterparty's route runs through a collaboration agreement rather than published terms. The privacy notice (last updated 21 October 2024) covers personal information only.

Generate is a therapeutics developer running its platform internally, so the record systems that matter to a collaborator are research systems such as electronic lab notebooks, laboratory information systems and research data platforms. The platform's own build and measure systems are described in detail.

How a collaborator's data and the platform's results move between the two organizations is not published.

An internal platform, with location implied rather than committed. The Generate Platform is not deployed to customers: it runs on the company's own infrastructure, coupled to its laboratories in Somerville and its Boynton Yards and Andover sites, and collaborators' work comes to it. Nothing states where the models run or the data rests, whether outside cloud providers are used, or how one collaborator's material is separated from another's and from the company's own programs. The exception is Chroma, released openly, which academic and non profit users run wherever they choose.

The Generate Platform is not licensed or sold: the company develops its own pipeline and runs confidential collaboration programs with Amgen and Novartis, first announced in January 2022 and September 2024. As a listed company (Nasdaq: GENB, trading since February 2026) it reports its financial results quarterly, and Chroma's code and weights are released at no cost to academic and non profit users.

What a collaboration costs or pays is not published.

Coverage is named, with results behind part of it. The platform addresses protein therapeutics: antibodies against defined epitopes, functional antibodies that act on cell surface receptors, proteins optimized for arbitrary properties, and proteins recoded to reduce immune recognition.

The disclosed pipeline runs across immunology and inflammation (GB-0895 in severe asthma and COPD) and oncology (GB-4362 with antibody drug conjugates, GB-5267 in metastatic ovarian cancer, and preclinical work including next generation ADCs), with further confidential collaboration programs. Results sit behind the immunology work, where a platform designed antibody has reached Phase 3; the oncology and preclinical modalities are named at earlier stages. Small molecules are outside the platform's scope.