Lila Sciences

Nothing here exists without the models. The company describes its system as an operating system for science in which models generate the hypotheses, proprietary instruments run the experiments and results feed back in real time, and its own framing puts the model at the centre and the robotics in service of it. There is no consulting business, no contract research offering and no product that would survive the removal of the machine learning. Whatever else is uncertain about this vendor, that the artificial intelligence is the whole proposition is not in question.

The highest claimed autonomy in this index paired with the least documented oversight, which is why this sits below vendors claiming far less. The company states plainly that its operating system executes the entire scientific method autonomously, from hypothesis through experiment to learning, and describes humans as guiding rather than performing. Nothing published describes the guiding.

There is no human approval gate, no escalation path, no statement of what a scientist can veto or when the system stops and asks. Neighbours in this lane that publish round by round hit rates from inside their loops grade higher here on far more modest autonomy claims.

What keeps this from the floor is evidence that people are in fact involved: the company's own account of its cell therapy work credits three named scientists with the design, so the practice is more supervised than the marketing implies. That is a reassurance drawn from a blog post rather than a disclosed operating model.

The least technically disclosed vendor graded in this lane, and the claims are the largest. Trademarked names stand where descriptions should be: Scientific Superintelligence, AI Science Factory, an operating system for science. No architecture, model class, parameter scale, training data, version history, evaluation suite or technical report exists, and two dedicated passes located no paper or preprint of any kind.

The company's homepage states that its model consistently outperforms other models across scientific domains with superior capabilities in complex analysis and reasoning, naming no benchmark, no comparator, no task and no source. A comparative superiority claim with nothing behind it is weaker than saying nothing, because it invites a reader to assume a measurement that has not been shown to exist. Neighbours in this lane that publish nothing about their models at least do not claim to have beaten anyone.

One named dependency and a large silence around it. The June 2026 collaboration with Nvidia identifies the BioNeMo agent toolkit as an external framework the company builds on, which is a specific and checkable disclosure, and the Department of Energy work names the national laboratory and university institutions involved. That is more than several neighbours offer.

Everything else is absent: no compute or hosting provider stated beyond what the Nvidia relationship implies, no foundation model base named for the large language models the company says it has built, no training data provenance, no open source inventory and no account of whether the scientific models are trained from scratch or adapted. One relationship to read carefully rather than at face value: Nvidia appears both as a technology partner and, on press reporting, as an investor, so it is not an arm's length supplier.

The widest gap between claim and evidence in this lane. The claims are large and specific: thousands of discoveries, genetic medicine constructs outperforming commercially available therapeutics, hundreds of antibodies, peptides and binders, and a blog dated 28 August 2026 stating that platform discovered messenger RNA yielded a CAR-T therapy that beat a leading marketed therapy in non human primates, designed by three scientists in months.

Two dedicated passes on 29 August 2026 located no paper, no preprint, no dataset and no third party confirmation for any of it, and the primate account reached the world through a company blog and social media amplification rather than a scientific venue. Unpublished primate data is normal at this stage in this field, so the issue is not the absence of a paper but the absence of anything checkable behind claims of this size.

What is real and independent sits outside the therapeutics story: Phase I awards under the United States Department of Energy Genesis Mission in July 2026, alongside Caltech, Lawrence Berkeley National Laboratory, Northwestern and Argonne, which is a competitive federal selection the company did not grade itself.

No protected health information is in scope. Two questions apply and neither is addressed anywhere. The first is autonomy at the bench: this is a system the company says executes the entire scientific method autonomously, generating its own hypotheses and running its own experiments in physical laboratories, across chemistry and biology, at a company that also serves an aerospace and defence vertical.

Nothing published states what the system will not attempt, what controls sit between a generated hypothesis and a physical experiment, or how the boundary is enforced across domains. The second is particular to this business model and sharper than the partner separation issue elsewhere in this lane. Catalyst places the platform inside a partner's existing research programmes, while Creation uses the platform to originate discoveries and found new companies.

What a Catalyst engagement contributes to Creation output is the obvious question and no published statement addresses ownership, segregation or use limits between the two. Recorded as located absences: nothing indicates a problem, and nothing answers the questions either.

Graded neutrally because the obligation does not currently arise. The work described is discovery and preclinical, operating on molecular and experimental data rather than protected health information, with no covered entity relationship and no business associate agreement published or required.

Recorded in advance for a later reader: the stated therapeutic scope includes cell therapies, and a company that moves cell therapy work toward the clinic acquires patient data obligations that nothing published anticipates. If a clinical programme appears, this axis should be regraded on the arrangements then in force rather than left at a neutral mark earned during discovery.

Two dedicated passes on 29 August 2026 across a targeted search and a direct review of the site navigation and footer found no security page, no trust centre, no certification and no attestation. The footer carries a privacy policy and terms of service and nothing else.

The absence is more conspicuous here than anywhere else in this lane because of the company Lila keeps: it holds Phase I awards under a United States Department of Energy programme alongside national laboratories, and it markets an aerospace and defence vertical. Federal and defence adjacent work normally comes with published security posture as a matter of course. Whatever exists internally, none of it has been made visible.

Nothing to clear. No candidate has been named, no investigational application disclosed, no clinical programme exists and two dedicated passes on 29 August 2026 located no enforcement action or adverse standing. The company's therapeutic work is preclinical and, on the published record, has not been assigned to a named indication.

Recorded for a later reader: the Department of Energy Genesis Mission awards and the aerospace and defence vertical bring this vendor within federal contracting and export control regimes that sit entirely outside the medicines framework this axis measures, and nothing published addresses its posture under either.

Nothing exists on any published surface as of 29 August 2026. No governance framework, no model card, no evaluation methodology, no responsible use policy and no statement of failure modes appears across the site, and unlike most neighbours in this lane there is no body of published method work standing in the gap.

The company's own launch language commits it to responsibly achieving scientific superintelligence, and that word is the whole of the governance content located: an aspiration with no framework, no criteria and no accountability mechanism behind it. The gap is widest precisely where the stakes are described as highest, at a company running autonomous experimentation across biology, chemistry and defence adjacent domains and claiming to be building superintelligence.

A privacy policy and a terms of service are linked from the site footer and both were confirmed to exist; the terms could not be retrieved for reading on 29 August 2026, so this grade rests on what the surrounding record establishes rather than on the document text, and a later pass should read it directly. What is clear without it is that no public source addresses the questions this business model raises.

Creation exists to originate discoveries and found new companies, and Catalyst places the platform inside a partner's own programmes, so the allocation of ownership over what the system discovers is the central commercial question here rather than a peripheral one. Nothing published states who owns a Catalyst discovery, what rights attach when the platform originates something adjacent to partner work, or what recourse a partner has when an autonomously generated result proves wrong. Website terms would not normally reach any of that in any case.

No electronic health record surface exists and none would be appropriate for a discovery platform, so this is graded neutrally under the convention for this lane. Recorded in its place, and it is more substantive than at most neighbours: the Catalyst route is explicitly designed to sit as an operating layer on top of a partner's existing research programmes rather than replace them, which is an interoperability posture at the level of workflow.

A June 2026 collaboration with Nvidia covers the BioNeMo agent toolkit, a named external framework, and the Department of Energy work connects to national laboratory computing. No application programming interface, connector or self serve product surface is offered.

Graded neutrally under the convention for this lane, now the ninth consecutive record where the convention rather than the evidence sets the grade. The physical footprint is disclosed and unusually relevant here: laboratories and teams in Cambridge Massachusetts, San Francisco and London, which places operations across two jurisdictions with materially different data regimes.

Nothing published states where partner data is held or processed, how a Catalyst engagement is separated from the rest of the platform, what is retained after an engagement, or how work under federal awards is segregated from commercial work. For a company whose central offering is placing its system inside someone else's research programme, the absence of any statement on separation is the gap a partner would notice first.

Well disclosed on capital and silent on customers. Investors are named at length across both rounds, including Flagship Pioneering, General Catalyst, March Capital, the ARK Venture Fund, a subsidiary of the Abu Dhabi Investment Authority, the State of Michigan Retirement System, Braidwell and Collective Global, and the rounds are dated with leads identified. Two commercial routes are named and described.

Against that, not one customer or commercial partner is named in any of the seven industries, no contract, deal value or revenue indication exists, and the disclosed figures do not add up: a $200 million seed plus a $235 million Series A is $435 million against a widely reported total near $550 million.

The valuation above $1.3 billion and the Nvidia investment are reported by press rather than stated by the company, so the two most quoted facts about this company's scale are not its own disclosures.

Within healthcare the modality coverage claimed is the widest in this lane by a distance. Where every neighbour here works one modality, small molecules or antibodies or RNA, Lila names messenger RNA, proteins, antibodies, cell therapies and small molecules under Therapeutics, and carries a separate Biotech vertical for turning biological data into research decisions. That breadth is the direct consequence of selling a general scientific method rather than a domain tool.

Held below the top grade because breadth of claim is not depth of practice: no programme, indication, target or partner is named in any modality, and the same platform is simultaneously offered to five industries outside life science, which spreads the evidence rather than concentrating it.