Superluminal Medicines

The company consistently presents three co-equal pillars, structure based drug discovery, protein dynamics and machine learning, and two of those three are physics disciplines that long predate modern machine learning. Structure based design and molecular dynamics were producing drugs before generative models existed, and part of the structural and design work here runs on licensed third party software, Schrodinger LiveDesign, rather than on the company's own models. Without the machine learning, what would remain is a computational chemistry company with cryo-electron microscopy capability, which is a real business.

Against that, the company's stated differentiator is a claim about models: receptor focused co-folding models, de novo design across diverse three dimensional pocket conformations, and predictive absorption and toxicology, aimed at modeling the specific structural change that produces a therapeutic effect. The lead program expresses that directly, a biased agonist designed to activate only the signaling pathways responsible for the intended effect, which is a design problem the physics alone does not solve. Central, then, but not solely so.

The architecture is right and the mechanism behind it is not published. The platform is described as predict, design and test, with experimental validation named as a component rather than implied, and the company says its agentic cryo-electron microscopy generates hundreds of empirical receptor structures, so a physical checkpoint exists between what the models propose and what the company claims.

What is missing is everything that would make it a control rather than a description. No hit rate, pass rate or round by round data is published showing how often designs fail as well as how often they succeed. Nothing states who selects which designs are synthesized, at what confidence a prediction is not acted on, what happens when a prediction and an assay disagree, or which judgments require a chemist rather than allowing one. Oversight is asserted through the shape of the workflow without a mechanism attached to any step of it.

The approach and the suppliers are both named, which stands out.

The platform is called Hyperloop and its components are stated rather than gestured at: structure based drug discovery, protein dynamics modeling, generative chemistry and machine learning, in silico pharmacokinetic and toxicology prediction, and experimental validation, arranged as a predict, design and test architecture. The September 2026 financing announcement adds more: agentic cryo-electron microscopy capabilities the company says generate hundreds of empirical receptor structures, receptor focused co-folding models, de novo small molecule design across diverse three dimensional pocket conformations, predictive absorption and toxicology models, and receptor biophysics. The stated aim is to model the specific structural change that produces a therapeutic effect and design selectively for it, which the lead program expresses as a biased agonist activating only the signaling pathways responsible for the intended effect.

The suppliers are named too: Schrodinger LiveDesign for molecular design and data analysis and CDD Vault for assay data management.

What is absent is everything beyond naming. No model is identified, no architecture, model class or parameter count is given, the training data behind the co-folding models is not characterized by size or source, no benchmark result is published, and no version identifier or update practice exists for any component.

Substantial partial disclosure, resting on named dependencies rather than on policy. Two external systems are identified with their specific functions: Schrodinger LiveDesign for digital molecule design and shared analysis of experimental and virtual data, and CDD Vault for protocol definition and assay data organization. Naming the actual tools a discovery stack runs on shows that part of the structural and design capability is licensed rather than built.

The model layer itself is described as the company's own: receptor focused co-folding models, de novo small molecule design and predictive absorption and toxicology models, with no third party model provider named or implied between the structures and the designs.

What remains undisclosed is the proprietary half. There is no training data provenance for the generative and predictive models, no compute or hosting provider, no foundation model base named for the co-folding work, and no open source component inventory. Nvidia is an investor across all three rounds, which implies a compute relationship that is nowhere stated as one, and nothing describes what governs a partner's target information once it reaches the platform.

A named customer and vendor reported results, with no method behind them.

The customer is substantial. In August 2025 Eli Lilly signed a collaboration worth up to $1.3 billion covering undisclosed receptor targets in cardiometabolic disease and obesity, structured as upfront and near term payments, an equity investment, development and commercial milestones, and tiered royalties, with Lilly taking exclusive rights to compounds arising from the work. A $60 million Series B closed on 3 September 2026, led by BVF Partners with Deep Track Capital and Perceptive Advisors joining, bringing disclosed funding to $213 million.

The pipeline is now named. The lead program is a selective, biased MC4R agonist for rare genetic forms of obesity and hypothalamic obesity, with a Phase 1 trial stated to begin by the end of 2026 and potential later application in Prader-Willi syndrome and in combination with GLP-1 therapies. The company states the candidate demonstrated high selectivity and a favorable safety profile in preclinical studies.

None of that is checkable. No figure, denominator, method or comparator accompanies the preclinical claims, no paper, preprint or dataset is published, no molecule has entered a human study, and the Phase 1 start is forward looking. The company describes itself as transitioning to a clinical stage biotechnology company rather than being one. A sophisticated buyer paying twice is meaningful, and it is not the same as a measurement anyone else can check.

Without a privacy policy, terms of use or cookie policy anywhere on the site, nothing describes how information moves through this system: there is no document of any kind to set a retention window, a training use position or a handling commitment.

No protected health information is in scope today. The separation question that does apply is sharper than the usual version, because of how close the parties sit. Superluminal runs a wholly owned pipeline in the same target family and the same therapeutic area, cardiometabolic disease and obesity, that its Lilly collaboration covers, and it does that work inside Lilly's own incubator building while Lilly holds equity in it. Nothing published states how the wholly owned programs are walled from the partnered ones, what the platform retains from collaboration work, or how target selection avoids collision between a partner's undisclosed targets and the company's own. Physical colocation with the partner makes the question more pressing rather than less. Nothing indicates a problem and nothing addresses it either.

Safety engineering, meaning guardrails, hallucination mitigation and how a safety event is handled, is not published either.

No status is stated and no privacy document exists. The site carries no legal documents at all: no privacy policy, no terms of use and no cookie policy appear anywhere in the navigation or the footer. There is not even a website notice that might fall short of covering the platform.

The obligation may not yet have arisen. The platform operates on receptor structures, molecular dynamics and assay measurements, with no protected health information in scope and no covered entity relationship. Still, nothing published says plainly that the platform processes no patient data, and a statement to that effect, with a document behind it, is within easy reach.

The question is about to become live. The company states that its lead program, a selective biased MC4R agonist for rare genetic forms of obesity and hypothalamic obesity, is expected to begin a Phase 1 trial by the end of 2026, and a sponsor running its own trial acquires patient data obligations that nothing published anticipates.

The site now blocks automated reading, so a privacy document may have been added since.

Nothing is published. There is no security page, no trust center, no certification and no attestation anywhere across the navigation or the footer. Beyond that, no legal documents exist on the site at all, with no privacy policy, no terms of use and no cookie policy, so there is not even the standard disclaimer language that usually constitutes a company's entire published security posture.

The exposure is concrete. Superluminal holds undisclosed receptor target information for one of the largest pharmaceutical companies in the world under a collaboration worth up to $1.3 billion, it runs a wholly owned pipeline in the same therapeutic area, and it operates from inside that partner's own incubator building. An attestation covering the environment where target and design data sit is the first assurance a partner's security review would ask for, and nothing of the kind is published.

The platform makes no device claim and is scoped accordingly. The models and the structural biology capability are not offered as a medical device and carry no clearance, which is the appropriate position for what they are.

Nothing is cleared and nothing adverse appears. Every program is preclinical on the public record. The lead asset, a selective biased MC4R agonist for rare genetic forms of obesity and hypothalamic obesity, is stated to begin a Phase 1 trial by the end of 2026, and no investigational new drug clearance, trial registration, first dosing announcement or designation has been published for it. The company describes itself as transitioning to a clinical stage biotechnology company rather than being one.

At least one commercial drug pipeline database attributes an unrelated clinical stage compound and its trial results to this company. Nothing Superluminal has disclosed supports that attribution.

Nothing is published on how model behavior is governed or tested. The site runs to about, science, pipeline, careers and press sections and none of them addresses governance, evaluation practice, failure modes or responsible use. There is no model card, no validation methodology, no benchmark and no independent audit, and no body of published method work stands in the gap, because this company has published no papers, preprints or datasets at all.

The relevant failure mode is concrete and unaddressed. A platform built to predict which structural change produces a therapeutic effect will have target classes and conformational states where it performs poorly, and the company's own framing makes that sharper rather than softer: it claims agentic cryo-electron microscopy generating hundreds of empirical receptor structures and design across diverse pocket conformations, which is a claim about coverage of a space whose boundaries are never stated. No statement exists of where the approach degrades or how the company would know.

The one result offered, that the lead candidate showed high selectivity and a favorable safety profile in preclinical studies, carries no figure, denominator or comparator.

Mechanisms exist that let an output be challenged, and two of them are named rather than implied. The platform's architecture puts experimental validation in as a component, so a physical checkpoint sits between what the models propose and what the company claims, and the data behind it lands in named third party systems, Schrodinger LiveDesign for design and analysis and CDD Vault for assay data, which are auditable environments rather than a spreadsheet. The Lilly collaboration structure adds a commercial check: the partner takes exclusive rights to resulting compounds and would test them itself before developing them.

Nothing stands behind the output. No hit rate, pass rate, accuracy figure or error rate is published for the design or prediction models with a method and a denominator, and the preclinical claims for the lead candidate, high selectivity and a favorable safety profile, carry no numbers. No guarantee, indemnity or remediation obligation is published for a partner whose program follows a prediction that does not hold, and the terms that would allocate that risk sit inside a collaboration agreement that is not public.

No clinical record integration exists. There is no electronic health record touchpoint, no clinical workflow surface and no provider connection, because the company designs small molecules against receptor targets rather than serving health systems. That will remain true when its lead program enters a Phase 1 trial, which runs through a clinical site rather than through anything it offers.

On the research systems of record a partner would run, nothing is named either. No partner electronic laboratory notebook, laboratory information management system, research data platform or cloud environment is named as connected, and no connector or application programming interface is offered. The company does name two research systems of record it runs itself, Schrodinger LiveDesign for molecular design and analysis and CDD Vault for assay data management, which is specific, but that is its own stack rather than a connection a partner could use. Nothing published describes how data moves between Superluminal and Eli Lilly under a collaboration in which the partner takes exclusive rights to the compounds.

The platform runs for Superluminal rather than being deployed to anyone, so there is no tenancy model, hosting option or isolation commitment to evaluate. Nothing is published about where partner or program data is held, under what separation, with what retention or with which subprocessors, and no cloud or compute provider is named anywhere, though the company's design and assay data sit in two named third party systems, Schrodinger LiveDesign and CDD Vault, whose own hosting is not described here.

The one deployment fact that is disclosed is physical and consequential. The company's registered address is inside Lilly Gateway Labs on Necco Street in Boston, the incubator operated by its investor and largest commercial partner. Sharing a building with a counterparty is a data adjacency question in its own right, and nothing published addresses how systems, records or access are separated, which matters because the wholly owned pipeline and the partnered work sit in the same therapeutic area.

Nothing is licensed or installed, so no price is published and none would apply. What a prospective partner can see before contact is the deal structure, and it is set out in some detail.

The August 2025 Eli Lilly collaboration is stated at up to $1.3 billion and broken into its components: upfront and near term payments, an equity investment, development and commercial milestones, and tiered royalties on net sales, with Lilly holding exclusive development and commercialization rights to compounds arising from the work.

Those are contract values rather than prices, and no individual amount is given, so the upfront, the equity and the milestone tranches are all unsized and $1.3 billion is a ceiling that may never be reached. Equity funding is disclosed plainly, with leads and syndicates named: $33 million in seed in August 2023, $120 million in Series A in September 2024, and $60 million in Series B on 3 September 2026 led by BVF Partners, bringing the total to $213 million.

One entanglement colors the commercial signal. Eli Lilly is an investor in the company, its largest commercial partner, and its landlord, since Superluminal's registered address is inside Lilly Gateway Labs. Nvidia is also both an investor and a widely used technology supplier in AI drug discovery. None of that makes the validation false, but it is not arm's length.

Narrow by design and explicit about the boundary, with the specifics now named. The work is small molecules against membrane proteins, starting with G protein coupled receptors and with a stated intention to extend to ion channels and transporters. On the partnered side the therapeutic focus is cardiometabolic disease and obesity under the Eli Lilly collaboration, against undisclosed targets. On the wholly owned side the lead program is a selective biased MC4R agonist for rare genetic forms of obesity and hypothalamic obesity, with Bardet-Biedl syndrome named, potential later application in Prader-Willi syndrome, and a stated route into the broader population in combination with GLP-1 therapies. Six small molecule programs were reported overall, and the wider frame is endocrine and cardiometabolic disease.

Nothing here reaches biologics, other target classes, translational work or clinical development, and the company does not claim otherwise. The choice is defensible on its own terms, since the undrugged fraction of this receptor family is large enough to occupy a company indefinitely.

Coverage is stated clearly, with nothing validating it yet. No program has reached a human study, no result is published for any of them, and the partnered targets are undisclosed, so nothing published separates where the platform has been shown to work from where it is being applied.