Autopoiesis Sciences
Develops foundational AI models to accelerate scientific discoveries in medicine and biology.
Website: https://autopoiesis.science/
Cover Block
PUBLIC
| Field | Value |
|---|---|
| Name | Autopoiesis Sciences |
| Tagline | Develops foundational AI models to accelerate scientific discoveries in medicine and biology |
| Headquarters | San Francisco, United States |
| Founded | 2025 |
| Stage | Seed |
| Business Model | B2B |
| Industry | Deeptech |
| Technology Type | AI / Machine Learning |
| Geography | North America |
| Growth Profile | Venture Scale |
| Founding Team | Co-Founders (2) |
| Funding Label | Seed (amount undisclosed) |
Links
PUBLIC
- Website: https://autopoiesis.science/
- LinkedIn: https://www.linkedin.com/company/autopoiesis-sciences
Executive Summary
PUBLIC
Autopoiesis Sciences is a San Francisco startup building a foundational AI system, branded Aristotle, intended to function as an "AI co-scientist" for researchers in medicine and biology [Oracle, 2025]. The company emerged in 2025 around a thesis that current general-purpose large language models, while fluent, are insufficiently rigorous for primary scientific work, and that a model trained and architected specifically for skeptical, evidence-grounded reasoning can fill that gap [GABA Northern California]. Co-founders Joseph Reth, an AI and consciousness researcher who serves as CEO, and Dr. Eike Gerhardt, whose background spans banking, M&A and venture capital, lead the company [Crunchbase]. A seed round closed on July 30, 2025, led by Informed Ventures, with Alpaca VC, Cross Atlantic Angels and a roster of individual angels also on the cap table; the dollar amount has not been disclosed [Tracxn, July 2025] [PitchBook]. The product surface visible today is Aristotle X1 Verify, which the company describes as achieving "state-of-the-art performance on the most challenging scientific reasoning benchmarks while solving the calibration problem" and which is being released to a select group of researchers [Autopoiesis Sciences]. The company has also disclosed a partnership with Oracle Cloud Infrastructure for the underlying compute footprint [Oracle, 2025]. Over the next 12 to 18 months the most informative signals will be peer-reviewed or third-party benchmark validation of the Aristotle line, the identity of the first design-partner laboratories, and whether the seed lead converts into a priced Series A.
Data Accuracy: GREEN -- Confirmed by Crunchbase, Tracxn, PitchBook, and the company's own website plus an Oracle customer page.
Taxonomy Snapshot
| Axis | Value |
|---|---|
| Stage | Seed |
| Business Model | B2B |
| Industry / Vertical | Deeptech, AI for science |
| Technology Type | AI / Machine Learning (foundation models) |
| Geography | North America (San Francisco) |
| Growth Profile | Venture Scale |
| Founding Team | Co-Founders (2) |
| Funding | Seed, lead Informed Ventures, July 2025 |
Company Overview
PUBLIC
Autopoiesis Sciences was founded in 2025 in San Francisco with the stated ambition of building "the foundation for scientific superintelligence to accelerate breakthrough discoveries and help cure previously incurable diseases" [Autopoiesis Sciences]. The name is taken from the biological concept of self-producing systems, a thematic nod to the founders' framing of science as a self-extending process that an appropriately designed AI could accelerate. Joseph Reth, the CEO, is described in his Crunchbase profile as "an American entrepreneur, computer scientist, and AI researcher known for his work in artificial intelligence and consciousness research" [Crunchbase]. Reported birth-year reporting puts him among a younger cohort of founders in the foundation-model wave [iMedia]. Co-founder Dr. Eike Gerhardt brings the commercial and capital-formation side: per Crunchbase he holds two master's degrees, in Entrepreneurship & Innovation and in Rhetoric, plus a Ph.D., and has prior experience in banking, M&A and venture capital [Crunchbase].
In an interview with GABA Northern California, Gerhardt framed the founding problem in plain terms: "At Autopoiesis, we're creating Aristotle, the world's leading AI scientist... The problem is simple: most AI systems are great at sounding right, but often aren't" [GABA Northern California]. That diagnosis, calibration and verifiability rather than raw fluency, is the throughline of the company's public messaging.
The two confirmed milestones to date are the seed round closed on July 30, 2025 led by Informed Ventures [Tracxn, July 2025], and the disclosed Oracle Cloud Infrastructure relationship supporting model training and serving [Oracle, 2025]. Beyond Reth and Gerhardt, the small team includes Ally Reth as Director of Industry Partnerships and Performance [Crunchbase] and Jennifer Yoon in a product design and scientific-claims-validation role [LinkedIn] [RocketReach].
Data Accuracy: GREEN -- Confirmed by Crunchbase founder profiles, Tracxn funding record, and an Oracle-published customer page.
Product and Technology
MIXED
The product is Aristotle, described by the company as "an AI co-scientist designed to think like a real researcher: skeptical, careful, and grounded in evidence" [LinkedIn, 2026] [PUBLIC]. The first publicly named release is Aristotle X1 Verify, which the company markets as reaching "state-of-the-art performance on the most challenging scientific reasoning benchmarks while solving the calibration problem" and which is currently "available to a select group of researchers to accelerate scientific discoveries" [Autopoiesis Sciences] [PUBLIC]. The specific benchmarks, scoring methodology, and identities of the design-partner researchers have not been published in primary-source detail, so the headline performance claim should be read as company-stated rather than independently verified [PUBLIC].
Two named technical components sit underneath Aristotle. The first is what the company calls "Double Check", a verification layer for scientific claims, and the second is "Dynamic In-line Definitions" (DIDs), described as a mechanism for grounding terms used during reasoning [LinkedIn, 2026] [PUBLIC]. Together, per the company's framing, these are intended to address the calibration problem in current LLMs, that is, the gap between confident-sounding output and verifiable accuracy [GABA Northern California] [PUBLIC]. The Oracle customer page positions Autopoiesis as building a model "to help scientists accelerate discoveries, validate complex hypotheses, and address urgent challenges in medicine and biology", with Oracle Cloud Infrastructure as the disclosed compute partner [Oracle, 2025] [PUBLIC].
A software engineer intern listing surfaced on a third-party aggregator [teeming.ai] suggests the team is hiring on the engineering side, but no broader open-roles footprint was captured, and no public detail confirms the model architecture, parameter count, training corpus composition, or eval suite [MIXED]. Investors evaluating the technology should treat the Aristotle X1 Verify benchmark claim as the central diligence question and request the underlying eval cards directly.
Data Accuracy: YELLOW -- Product positioning is confirmed by the company website and the Oracle customer page; specific benchmark scores and architecture remain company-stated and not independently verified.
Market Research and Opportunity
PUBLIC
AI for scientific discovery has moved from a research curiosity to a funded category in the span of roughly two years, and Autopoiesis is entering at a moment when both compute partners and pharmaceutical buyers are actively building budgets for it. Public, named third-party sizing for the precise "AI co-scientist" segment was not surfaced in the cited research, so this section walks the demand drivers and analogous markets rather than asserting a specific TAM.
The demand thesis the founders articulate, that current general-purpose LLMs sound right more often than they are right, maps onto a real procurement problem inside research-intensive organizations [GABA Northern California]. Pharmaceutical R&D, academic biology and translational medicine all share a common feature: the cost of a confidently wrong answer is high, often measured in failed experiments or, downstream, in failed clinical readouts. A model that explicitly addresses calibration, the gap between model confidence and ground-truth accuracy, is therefore positioned against a buyer pain point rather than against a generic productivity story. The company's Oracle relationship signals that at least one hyperscaler-adjacent partner has chosen to publicly associate its infrastructure brand with the Autopoiesis mission [Oracle, 2025], which is itself a soft demand signal in a market where compute access is gated.
Adjacent and substitute categories are crowded and well capitalized. The closest substitute is the use of frontier general-purpose models (the GPT, Claude and Gemini families) directly inside research workflows, often through retrieval-augmented pipelines built by in-house data-science teams. A second substitute is the cluster of biology-specific foundation models targeting protein structure, molecule generation, and single-cell data. A third is the academic effort around AI co-scientists from large industrial labs, which has been publicly discussed by Google and others. Autopoiesis's positioning, a reasoning-and-verification layer rather than a domain-specific structural model, is differentiated from the second category and overlaps more directly with the first and third.
Regulatory and macro forces cut both ways. On the tailwind side, public and private funders are increasingly receptive to AI-augmented research proposals, and pharma companies have signalled willingness to pilot external AI tools at the discovery stage where regulatory exposure is lower than in clinical work. On the headwind side, any AI system that produces or validates scientific claims will face rising scrutiny on provenance, reproducibility, and data licensing, and the calibration claim Autopoiesis is making is precisely the claim regulators and journal editors will probe hardest.
| Sizing reference | Value | Source |
|---|---|---|
| Confirmed third-party TAM for AI co-scientist tools | Not publicly available in cited research | -- |
| Disclosed seed round date | July 30, 2025 | [Tracxn, July 2025] |
| Disclosed compute partner | Oracle Cloud Infrastructure | [Oracle, 2025] |
Analyst takeaway: with no named third-party sizing of the AI-co-scientist segment in the captured research, the investable thesis rests on demand signals (a hyperscaler customer page, a led seed round) rather than on a top-down market number. That is appropriate for a 2025-vintage seed in a category that did not exist as a line item two years ago, but it does mean the diligence question is bottom-up: which laboratories pay, for what, at what price, starting when.
Data Accuracy: YELLOW -- Demand signals confirmed via Oracle and Tracxn; market sizing remains uncited in surfaced sources.
Competitive Landscape
MIXED
The captured research did not surface any named direct competitor for Autopoiesis Sciences, so the competitive map below is drawn from category-level positioning rather than from a structured-fact competitor list, and is presented as prose rather than as a comparison table.
The segment Autopoiesis is targeting can be split into three competitive layers. The first layer is the frontier general-purpose model providers, principally OpenAI, Anthropic and Google DeepMind, whose products are already in heavy informal use inside research organizations. These incumbents do not market themselves as scientific reasoning systems specifically, but they are the practical default that any Aristotle sale has to displace at the desk level. The second layer is the cohort of biology- and chemistry-specific foundation models, including the protein-structure and molecule-generation tools that have attracted significant venture funding since 2022. These are adjacent rather than directly substitutive: they answer different questions than a calibration-focused reasoning model. The third layer is the explicit "AI co-scientist" effort inside large industrial research labs, which has been publicly discussed by at least one hyperscaler research arm and which represents the most direct conceptual overlap with Autopoiesis. [PUBLIC]
Where Autopoiesis has a defensible edge today, on the visible evidence, is in focus and framing. The Double Check and Dynamic In-line Definitions components [LinkedIn, 2026] are described as purpose-built for the calibration problem, which is a narrower and more legible promise than "general intelligence applied to science". A small, mission-aligned team can iterate on that specific claim faster than a frontier lab whose roadmap is shared across many priorities. The Oracle partnership [Oracle, 2025] is also a concrete distribution and compute advantage that most seed-stage peers do not have on day one. Whether that edge is durable depends on two things: whether the calibration approach generalizes across scientific domains, and whether the design-partner researchers using Aristotle X1 Verify produce citable wins that can anchor enterprise sales.
Where Autopoiesis is most exposed is on the model-capability frontier. If frontier general-purpose models close the calibration gap as a side effect of their next training run, the standalone case for a specialized reasoning model weakens. The company also does not own a proprietary scientific dataset of the kind that some biology-specific peers have assembled, which means the moat has to come from reasoning architecture and verification tooling rather than from data exclusivity. A second exposure is go-to-market: enterprise pharma sales cycles are long, and the public record does not yet show a named pharmaceutical or academic anchor customer.
The most plausible 18-month competitive scenario splits two ways. Winner-if-X: if Aristotle X1 Verify produces an independently reproduced benchmark result on a recognized scientific reasoning eval, and if one named pharma or top-tier academic lab discloses a paid pilot, Autopoiesis can credibly raise a priced Series A on a category-leader narrative within the AI-for-science segment. Loser-if-Y: if the next frontier model release from OpenAI or Anthropic closes the calibration gap convincingly on public benchmarks before Autopoiesis publishes its own externally verified results, the standalone product narrative becomes harder to defend and the company is pushed toward either a vertical wrapper position or an acquisition conversation.
Data Accuracy: ORANGE -- No named competitors in structured facts; competitive map is inferred from category context plus the company's own positioning.
Opportunity
PUBLIC
If Autopoiesis builds what it says it is building, the prize is being the default reasoning-and-verification layer for AI-assisted scientific work, a position with a credible path to category-defining scale.
The headline opportunity. The single largest outcome Autopoiesis could plausibly become is the trusted verification layer that sits between general-purpose AI and primary scientific work. The company's framing, that "most AI systems are great at sounding right, but often aren't" [GABA Northern California], is not a narrow product claim; it is a positioning claim about where value will accrue in AI for science. If calibration and verifiability become the gating procurement criteria inside pharma R&D, academic medicine and translational biology, then a vendor that owns the reasoning-and-verification layer is structurally better placed than either a frontier generalist or a domain-specific structural model. The disclosed Oracle Cloud Infrastructure relationship [Oracle, 2025] and a seed round led by Informed Ventures with Alpaca VC and Cross Atlantic Angels participating [PitchBook] [Tracxn, July 2025] indicate that both an infrastructure partner and a venture syndicate are willing to underwrite that thesis at the seed stage.
Growth scenarios.
| Scenario | What happens | Catalyst | Why it's plausible |
|---|---|---|---|
| Pharma design-partner flywheel | Aristotle X1 Verify wins paid pilots inside two or three top-twenty pharma R&D groups and becomes an embedded reasoning layer for early discovery | First disclosed pharma anchor customer paired with a reproduced benchmark result on a public scientific-reasoning eval [Autopoiesis Sciences] | The product is already framed around a buyer pain point pharma feels acutely (confidently wrong outputs), and the Oracle relationship gives credibility on the procurement side [Oracle, 2025] |
| Academic-standard adoption | Aristotle becomes a citable verification tool used in published papers, the way certain statistical packages became default citations | A peer-reviewed publication using Double Check / DIDs to validate a non-trivial result [LinkedIn, 2026] | Calibration and verifiability are exactly the axes academic reviewers and journal editors are starting to police for AI-assisted work |
| Frontier-lab partnership or acquisition | A frontier model lab partners with or acquires Autopoiesis to bolt on a verification layer rather than build one | A public benchmark win that is hard for an internal team to replicate quickly | The cluster of angels around the round (including Mike Mahlkow and Hiro Mizushima) [PitchBook] suggests the company is well-networked into the operator and frontier-lab community |
What compounding looks like. The flywheel, if it starts, is reputational and data-driven. Each design-partner laboratory that uses Aristotle to validate a real claim produces a reference, and references in scientific markets compound unusually well: a single citable use inside a respected lab tends to pull in adjacent labs faster than a typical enterprise SaaS reference would. Over time, the accumulated record of validated and rejected claims becomes a proprietary corpus of calibration data that is difficult for a generalist competitor to replicate without doing the same fieldwork. The company's own framing of Aristotle as a co-scientist "skeptical, careful, and grounded in evidence" [LinkedIn, 2026] is consistent with building this kind of cumulative epistemic asset rather than a one-shot product.
The size of the win. No named third-party TAM for the AI-co-scientist category was surfaced in the captured research, so any number here is a scenario rather than a forecast. As a directional comparable, the public scientific-software and research-tools category includes multi-billion-dollar outcomes built on far narrower wedges than "reasoning layer for science". If the pharma design-partner scenario above plays out and Autopoiesis becomes an embedded reasoning layer inside two or three top-twenty pharma R&D budgets within 24 to 36 months, the company would be positioned as a credible Series B candidate in a category that public-market and strategic acquirers have historically paid premium multiples for (scenario, not a forecast). The downside framing belongs in the private half of this report.
Data Accuracy: YELLOW -- Scenarios are grounded in confirmed product positioning, the Oracle partnership, and the seed syndicate; the size-of-win discussion is explicitly labelled scenario rather than forecast.
Sources
PUBLIC
[Autopoiesis Sciences] Autopoiesis Sciences home | https://autopoiesis.science/
[Autopoiesis Sciences] About - Autopoiesis Sciences | https://autopoiesis.science/about
[Oracle, 2025] Autopoiesis taps Oracle Cloud Infrastructure to build AI co-scientist | https://www.oracle.com/customers/autopoiesis/
[Crunchbase] Autopoiesis Sciences company profile | https://www.crunchbase.com/organization/autopoiesis-sciences
[Crunchbase] Autopoiesis Sciences financial details | https://www.crunchbase.com/organization/autopoiesis-sciences/financial_details
[Crunchbase] Joseph Reth person profile | https://www.crunchbase.com/person/joseph-reth
[Crunchbase] Eike Gerhardt person profile | https://www.crunchbase.com/person/eike-gerhardt
[Crunchbase] Ally Reth person profile | https://www.crunchbase.com/person/ally-reth-863c
[Tracxn, July 2025] Autopoiesis funding rounds and investors | https://tracxn.com/d/companies/autopoiesis/__e0hRhAqelomJNOgVJXA_QmKIY0g03fZefJe0Q8I0Hk4/funding-and-investors
[PitchBook] Autopoiesis Sciences company profile | https://pitchbook.com/profiles/company/894666-61
[GABA Northern California] The German-American Entrepreneurial Journey of Eike Gerhardt | https://gaba-network.org/norcal/the-german-american-entrepreneurial-journey-of-eike-gerhardt-co-founder-of-autopoiesis-science/
[LinkedIn, 2026] Jennifer Yoon profile - Autopoiesis Sciences | https://www.linkedin.com/in/jennifermyoonn/
[LinkedIn] Autopoiesis Sciences company page | https://www.linkedin.com/company/autopoiesis-sciences
[iMedia] AI Aristotle benchmark and founder profile | https://min.news/en/tech/6f9eae89927c817d933bd5799360d8b5.html
[teeming.ai] Autopoiesis Sciences - Software Engineer Intern listing | https://teeming.ai/j/software-engineer-intern/136d20d9-a825-6887-e427-ca5177932202/autopoiesis-sciences/e565be88-990d-4457-9a0d-429d6ae5b41b
[f4.fund] Alpaca VC investment thesis and preferences | https://f4.fund/firms/alpaca-vc
Articles about Autopoiesis Sciences
- Autopoiesis Sciences Wants an AI Named Aristotle to Sit at the Lab Bench — The San Francisco seed-stage startup is building a reasoning model pitched as a skeptical co-scientist for medicine and biology.