Lunai Bioworks (Nasdaq: LNAI) Demonstrates AI Chemical-Risk Screening for Generative AI

5.1-fold enrichment using NIH Tox21 data validates structure-based AChE screening as OpenAI and Anthropic highlight growing dual-use AI risks

SACRAMENTO, Calif., Sept. 16, 2026 /PRNewswire/ -- Lunai Bioworks, Inc. (Nasdaq: LNAI) ("Lunai" or the "Company") today shared results from a technical case study conducted by its BioSymetrics subsidiary, supporting the development of structure-based screening tools for chemical outputs generated by large language models and other artificial intelligence systems. The study evaluates predicting acetylcholinesterase ("AChE") inhibition directly from molecular structure as a model-agnostic component of the Company's broader chemical-safety and biosecurity strategy.

Addressing the Dual-Use Challenge in AI-Enabled Science
As advanced AI expands the speed and breadth of scientific research, developers, policymakers, and chemical-security organizations are examining how to preserve legitimate innovation while managing dual-use risk.

Anthropic's September 2026 Threat Intelligence Report describes examples of malicious activity that Anthropic says it identified and disrupted across seven harm areas, including biological misuse and conventional weapons development, and emphasizes the need to strengthen safeguards as AI capabilities advance.

The Organisation for the Prohibition of Chemical Weapons ("OPCW") has identified AI as a significant emerging consideration for chemistry, verification and chemical security. In March 2026, the OPCW highlighted recommendations that include strengthening dialogue with science and technology developers, promoting responsible innovation, maintaining safeguards and human oversight, and building partnerships across scientific and technology communities. Existing AI safeguards can include model-level controls, classifiers, monitoring, access restrictions and other measures designed to identify or limit potentially harmful use. However, the dual-use nature of advanced scientific research can make intent difficult to determine from user interactions alone. BioSymetrics' approach addresses a complementary point in the workflow by evaluating the proposed chemical structure itself, rather than relying on information from the generative system that produced it. Lunai is developing this capability as a complementary screening layer intended to help prioritize chemical outputs for additional computational analysis, expert assessment, or confirmatory biological testing.

"The same generative capabilities that can transform drug discovery can also be repurposed to explore dangerous chemical space," said David Weinstein, Chief Executive Officer of Lunai Bioworks. "Effective biosecurity should extend beyond safeguards built into the generative system itself. It should also consider the potential biological activity of the structures an AI system proposes. By connecting predictive screening with experimental biology, we aim to provide an additional, objective layer of review that supports scientific innovation while helping the AI community manage serious dual-use risks."

Retrospective Benchmark and Technical Evaluation
To evaluate the feasibility of downstream structure-based screening, BioSymetrics applied its chemical mechanism-of-action modeling capabilities and the Contingent AI framework within its Augusta platform to predict AChE inhibition. AChE is an established target of organophosphate nerve agents and other neurotoxic compounds, making structure-derived prediction of AChE activity a relevant benchmark for chemical-hazard triage; AChE inhibition can also have therapeutic applications, depending on the compound and biological context.

In a retrospective analysis using publicly available Tox21 AChE assay data, BioSymetrics evaluated 9,667 compound entries and separated training, validation, and test sets by chemical scaffold to assess performance across structurally distinct chemotypes. In the Company's held-out test analysis, the selected model achieved an area under the receiver operating characteristic curve ("AUROC") of 0.88. At the reported operating point, the model achieved 33% precision at 70% recall, representing approximately 5.1-fold enrichment relative to the test set's 6.4% prevalence of AChE-active entries.

That enrichment measures the concentration of assay-active compounds among candidates flagged for review; it does not compare the model to other screening technologies. In practice, this type of enrichment could reduce the number of candidates requiring specialized secondary assessment when screening large batches of proposed chemical structures. Because the model's inputs are derived from molecular structure, Lunai is developing the screening component to operate independently of the third-party large language model, commercial API, or molecular-generation architecture that produces the candidate compounds.

The case study evaluates assay-defined AChE activity to help prioritize molecules for further review. It does not establish clinical neurotoxicity, real-world chemical hazard, prevention of chemical misuse, or detection across all classes of chemical threats, and it did not evaluate a deployed safeguard operating on AI-generated molecules. Broader deployment requires additional prospective validation.

An Invitation to Frontier AI Developers
Lunai believes structure-based screening can complement the safeguards frontier AI developers are already building into their models. The Company welcomes technical discussions with OpenAI, Anthropic, and other AI developers interested in evaluating independent, structure-based screening as part of a layered approach to chemical and biological safety. Potential areas of collaboration include controlled retrospective benchmarking, evaluation of AI-generated molecular outputs, integration studies, and development of escalation pathways that connect computational alerts with expert review and biological validation.

Lunai is not announcing a partnership with OpenAI or Anthropic at this time, and neither organization participated in nor endorsed the BioSymetrics case study.

Connecting Machine Learning with In Vivo Biological Testing
The proposed screening architecture builds on BioSymetrics' multi-year, revenue-generating commercial defense collaboration announced in April 2026. Under that initiative, the Company combines predictive mechanism-of-action analysis with in vivo zebrafish phenotypic screening to identify and classify neuroactive and neurotoxic compounds.

That combination provides an empirical bridge between computational prediction and whole-organism testing, allowing prioritized compounds to advance to biological assessment when appropriate. It also differentiates Lunai's approach from screening methods that rely exclusively on computational prediction.

More Than US$30 Million in Submitted Grant Requests
As part of its broader efforts in chemical defense, biosecurity and dual-use risk mitigation, Lunai has submitted multiple non-dilutive grant applications seeking more than US$30 million in aggregate project funding across federal and state programs. That amount represents submitted project requests, not grant awards, committed funding, backlog, or recognized revenue. There can be no assurance that any application will result in an award, or that an award will equal the amount requested; any funding Lunai ultimately receives would depend on the applicable award terms and, where relevant, allocations among project partners.

The Company expects to share additional technical information as its validation work and prospective partnership discussions progress.

The full technical case study, "Predicting Neurotoxicity Risk via AChE Inhibition Modeling from Chemical Structure," is available here.

About Lunai Bioworks
Lunai Bioworks, Inc. (Nasdaq: LNAI) is an AI-driven life sciences company combining clinical data, machine learning and in vivo biological validation to advance drug discovery and precision therapeutics. Through its BioSymetrics subsidiary, the Company deploys its Augusta platform to conduct computational mechanism-of-action analysis and phenotypic screening across central nervous system disorders, oncology, chemical defense and biosecurity initiatives. For more information, visit www.lunaibioworks.com.

Forward-Looking Statements
This press release contains forward-looking statements within the meaning of the Private Securities Litigation Reform Act of 1995, including statements about the further development, validation and potential applications of Lunai's and BioSymetrics' technology; integration of chemical-safety screening into generative AI and external foundation-model workflows; prospective discussions with frontier AI developers; the advancement and commercial potential of defense collaborations; and the potential receipt, amount, timing and use of grant or other project funding. Words such as "may," "could," "intend," "plan," "expect," "seek," "believe," "aim" and "welcome" identify forward-looking statements, although not all such statements contain these words.

These statements reflect management's current expectations and are not guarantees of future performance. Actual results may differ materially due to risks including whether retrospective model performance is reproduced in prospective testing or on unfamiliar chemical structures; false-positive and false-negative predictions; the need for further experimental validation; technical, integration and competitive challenges; customer and partner performance; the availability of sufficient capital; regulatory and government procurement requirements; changes in funding priorities; and the competitive, negotiated nature of grant awards, which may result in no funding, reduced amounts, partner allocations, or milestone and cost-sharing conditions. Further risks are described in the Company's most recent annual and quarterly reports and other SEC filings. Forward-looking statements speak only as of the date of this release, and the Company undertakes no obligation to update them except as required by law.

SOURCE Lunai Bioworks Inc.