OpenAI Foundation awarded SecureBio Detection $17,2 million to cut the time from sample collection to an early warning of a new pathogen from 14 days to 3 days
OpenAI Foundation awarded SecureBio Detection $17,2 million to cut the time from sample collection to an early warning of a new pathogen from 14 days to 3 days
SecureBio Detection announced the grant on August 20. Over the next year, the organization plans to reduce the time from sample collection to an actionable result from 14 days to 3 days, add three cities to its nasal swab program, and increase sequencing capacity.
A conventional test looks for a known genetic sequence from a pathogen. SecureBio collects wastewater and nasal swabs, sequences a broad range of DNA and RNA, and then looks for signs of an unusual pathogen. This method, known as metagenomic sequencing, may detect a threat that is not yet on the list of known infections. In wastewater, traces of viral material are diluted among DNA and RNA from bacteria and other organisms, so the system must sequence a very large number of fragments.
In the analysis of the three layers of biosecurity, this form of surveillance sits between DNA synthesis screening and measures that slow the transmission of infection. Its purpose is to detect the spread of a pathogen before a large outbreak occurs. The grant supports this layer of protection.
The three-day target covers the entire process from sample collection to an actionable result. SecureBio plans to speed up sample transport, automate and parallelize the preparation of material for sequencing, and configure data processing so that additional computing capacity can be brought online for short periods. Average weekly throughput is expected to reach 80 billion read pairs. Each pair represents one short segment of genetic material read from both ends.
The program began in 2021 as the Nucleic Acid Observatory project within the Sculpting Evolution group at the Massachusetts Institute of Technology. SecureBio will test how sensitively the entire pipeline, from sample collection to the final signal, detects different classes of viruses. Its staff and external consultants will also look for vulnerabilities that an attacker could exploit.
The grant supports the Detection program. A separate SecureBio team evaluates the biosecurity risks of AI models, including OpenAI models. That team has its own leadership, budgets, and work products. SecureBio director Jeff Kaufman wrote: “If there is any indication that Detection funding is being used to pressure our AI evaluation team, I will disclose it publicly.” SecureBio plans to test the speed of its early warning system over the next year. It describes the independence of its model evaluation team as a separate operating rule.