Mantis Biotechnology Corp, a San Francisco-based startup, is building simulated human bodies to generate the synthetic medical data that AI-powered surgical systems desperately need. The company has begun raising up to $5.5 million through a private securities offering, with first sales closing in mid-November 2025. Its core bet: that a physics engine capable of modeling diverse human anatomies can fill gaps left by incomplete, outdated, and demographically skewed real-world patient records.
A Physics Engine for Human Biology
The central technology behind Mantis involves what the company describes as a physics engine that simulates human biology and anatomy to produce synthetic biomedical data. Rather than relying solely on hospital records or clinical trial datasets, the system generates virtual patient models, sometimes called digital twins, that can represent a wider range of body types, conditions, and demographics than any single medical database currently captures.
Georgia Witchel, a co-founder of the company and a bioengineering alumna of the University of Washington, has framed the problem bluntly. “Not enough medical data… outdated or fails to cover diverse patient groups,” Witchel said, according to her alma mater’s Department of Bioengineering. That gap matters because autonomous surgical robots and other AI-driven medical tools require enormous volumes of training data to operate safely. When that data skews toward certain populations or misses entire patient categories, the resulting systems risk performing poorly on the patients they are supposed to help.
The physics-engine approach tries to sidestep this bottleneck entirely. Instead of waiting years for clinical studies to produce balanced datasets, synthetic simulation can theoretically generate millions of virtual patient scenarios on demand. If the models prove accurate enough, they could accelerate how quickly new surgical AI systems move from prototype to clinical use, while also enabling developers to test edge cases that might be too rare or risky to study directly in human subjects.
Early Fundraising and Corporate Structure
According to a Form D filing with the U.S. Securities and Exchange Commission, Mantis Biotechnology Corp is incorporated in Delaware with a principal address in San Francisco. The filing lists a total offering amount of $5.5 million under a Rule 506(b) exemption, with the date of first sale recorded as November 17, 2025. Georgia Witchel and Jessica Leao are both named as related persons in the document, indicating leadership or promoter roles in the offering.
A separate institutional account from the University of Washington’s bioengineering department describes the raise as $5 million directed toward building autonomous surgical robotics. In that narrative, Witchel’s work is presented as a continuation of her academic training, with the company positioned as a vehicle to bring research concepts into commercial reality. The discrepancy between the $5.5 million SEC figure and the $5 million cited by the university has not been publicly explained. It may reflect the difference between a total offering ceiling and the amount actually targeted or committed, but neither Mantis nor the school has clarified the gap.
The Rule 506(b) exemption allows companies to raise capital from accredited investors without registering the securities publicly, a common route for early-stage startups that want to avoid the cost and disclosure burden of a full public offering. For Mantis, this means the company can move quickly to fund development while limiting the financial details it must share with the broader market. At the same time, it leaves outside observers with only a partial view of the company’s finances, timelines, and milestones.
Why Existing Medical Data Falls Short
The problem Mantis is targeting is widely acknowledged across the healthcare AI sector. Training data for medical AI systems tends to come from a relatively narrow slice of the global population. Hospital records over-represent patients from large academic medical centers, which often serve specific geographic and socioeconomic groups. Clinical trials, meanwhile, have historically enrolled disproportionately white and male participants, leaving significant blind spots when algorithms trained on that data encounter patients who do not match the original sample.
These gaps carry real consequences. An AI system trained primarily on imaging data from one demographic group may misidentify tissue abnormalities in another. A surgical robot calibrated on anatomy data that skews toward a narrow body-type range could perform less reliably on patients outside that range. Witchel’s comments suggest that Mantis views synthetic data generation not as a convenience but as a structural fix for a training-data pipeline that cannot scale fast enough through traditional clinical channels alone.
The UW BioEngage initiative, which supports biotech innovation and industry partnerships at the University of Washington, provides some institutional context for Witchel’s trajectory. Her background in bioengineering appears to have shaped the company’s technical direction, combining simulation science with a focus on the demographic limitations of current datasets. That academic grounding may also help Mantis navigate the validation studies and regulatory scrutiny that any medical simulation platform will inevitably face.
A UK Entity Adds Questions
Adding a layer of complexity, a UK entity called MANTIS BIO LTD was incorporated on November 10, 2025, according to Companies House records. The entity’s status is listed as active. The incorporation date falls just one week before the first sale date on the U.S. SEC filing, raising questions about whether the two entities are coordinated or simply share a similar name by coincidence.
No public filings or statements from Mantis Biotechnology Corp have confirmed a formal connection between the Delaware corporation and the UK company. The official Companies House service provides basic incorporation data but does not automatically disclose ownership links to foreign entities. Without direct confirmation from the company’s principals, the relationship between the two Mantis entities remains unclear, and any assumptions about shared ownership or strategy would be speculative.
If the UK entity is indeed part of the same operation, it could signal ambitions beyond the U.S. market. Regulatory frameworks for medical AI differ significantly between the United States and the United Kingdom, and a company with footholds in both jurisdictions might be positioning itself to supply synthetic training data that satisfies multiple regulatory regimes simultaneously. For now, however, the public record stops at parallel incorporations and overlapping timing, leaving open questions about how, or whether, the entities interact.
What Digital Twins Would Change
The term digital twin has gained traction across industries from manufacturing to urban planning, but its application in medicine carries distinct stakes. A digital twin of a jet engine can be validated against decades of well-understood physics. A digital twin of a human body must account for biological variability that remains only partially mapped by science. Tissue density, organ elasticity, blood flow dynamics, and the behavior of living cells under stress all vary from person to person and even within the same individual over time.
For surgical AI systems, the promise of accurate digital twins is twofold. First, they could provide a safer environment to test and refine algorithms before those systems ever touch a human patient. Developers could run thousands of simulated procedures on virtual bodies with different anatomies, comorbidities, and responses to intervention, identifying failure modes that might not surface in limited clinical trials. Second, digital twins might eventually support personalized planning, allowing surgeons and AI assistants to rehearse a specific patient’s procedure on a customized model derived from that person’s scans and clinical data.
Realizing that vision, however, will require more than clever code. The physics engine at the heart of Mantis’s pitch must be grounded in validated biomechanical models and continuously checked against empirical data from real patients. Synthetic data that does not faithfully reproduce the statistical properties and edge cases of actual human biology could mislead, rather than improve, AI systems trained on it. Regulators and clinicians are likely to demand rigorous evidence that simulated anatomies behave like their real-world counterparts, especially in high-stakes settings such as autonomous or semi-autonomous surgery.
Balancing Hype, Risk, and Regulation
Mantis sits at the intersection of several fast-moving trends: the rise of generative AI, renewed attention to bias in medical datasets, and growing interest in autonomous surgical robotics. That combination is attracting capital, as the company’s private offering suggests, but it also invites scrutiny. Synthetic data can reduce privacy concerns associated with sharing patient records, yet it introduces new questions about transparency, validation, and accountability when simulated patients stand in for real ones.
How Mantis addresses those questions may determine whether its physics engine becomes a foundational tool for medical AI or remains a niche experiment. Clear evidence that its virtual anatomies improve performance across diverse patient groups would strengthen the case for synthetic data in regulated healthcare environments. Conversely, if simulations fail to capture the messy variability of real bodies, the technology could end up reinforcing the very gaps it aims to close.
For now, the public record offers only a sketch, a Delaware corporation raising millions under a private exemption, a co-founder with deep ties to academic bioengineering, and a parallel UK entity whose connection remains unconfirmed. The detailed workings of Mantis’s simulations, and the degree to which hospitals and device makers will trust them, are still largely out of view. As autonomous surgical systems inch closer to clinical reality, the company’s progress will be an early test of whether digital replicas of the human body can safely stand in for the real thing.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
