AUSTIN, TEXAS – October 21, 2025 – Signature Science, LLC, has been selected for the National Institute of Standards and Technology (NIST) Center for AI Standards and Innovation’s (CAISI’s) Basic Ordering Agreement (BOA) for the development of benchmarks for assessing the chemical and biological capabilities and safety of frontier artificial intelligence (AI) models. NIST CAISI will use the BOA construct to issue and award task orders under a $3M ceiling.
AI presents enormous opportunities for scientific breakthroughs alongside the potential for significant risks. Modern AI systems offer a vast library of chemistry and biology research which can be used to support the responsible design of new medicines or the understanding of complex biological processes. But this same knowledge can theoretically help a ‘bad actor’ create harmful substances, even if they lack scientific training. To understand and mitigate these risks NIST CAISI, aligned with recommended policy actions prescribed by America’s AI Action Plan, will leverage industry expertise to conduct pre-deployment testing, evaluation, validation, and verification the chemical and biological capabilities and risks of general-purpose and specialized chemical or biological AI models.
Signature Science will offer NIST CAISI its AI safety consultation services, which launched in 2023. As the AI industry boomed and its technology became both more powerful and more widely used, concerns over intentional or incidental misuse also grew, prompting agencies like NIST and AI developers to look to the scientific community. Signature Science’s staff of chemical and biological threat experts probe the models, seeking to correct scientifically flawed information and to flag instances in which the model delivers information that is not otherwise easily accessible, and could enable a malicious actor to perform expert-level scientific tasking to cause harm. Signature Science’s offerings include:
- Design and execution of human uplift studies. Performed in a laboratory under controlled conditions, studies assess whether and to what degree a novice scientist can use a target AI model to execute scientific work at an expert level;
- Red teaming. Threat experts collaborate with developers to run scenarios as malicious actors, testing for vulnerabilities and assessing whether the model inherently delivers or can be exploited to deliver information that could enable someone to cause harm to humans, livestock, or crops, as well as the accuracy of the provided scientific information.
These efforts can inform the establishment of guidance and benchmarks for the evaluation of new models.
“This is a significant opportunity to contribute to the responsible development of AI technology,” said Dr. Curt Hewitt, Signature Science’s Director of Biological Sciences. “This vital initiative from NIST supports the US’s mission to maintain its position at the forefront of AI innovation while identifying and mitigating chemical and biological threat risks to national defense and homeland security.”
Work under the contract will be performed at Signature Science’s locations in Austin, Texas and Charlottesville, Virginia.
About Signature Science, LLC:
A subsidiary of the Southwest Research Institute, Signature Science, LLC is a scientific and technical consulting firm providing multi-disciplinary applied research, technology design and development, and scientific, technical, and operational services to government and industry.