Stand up an independent red-teaming group that runs continuous adversarial testing of gene synthesis providers, frontier AI labs, and other technologies that are bioweapon-enabling.

## Stakes
A single engineered pathogen could, in theory, kill more people than every war in history combined. This engineered pathogen could be designed to be more dangerous with the help of AI, ordered from a DNA synthesis provider, handled in the lab with AI guidance, and released deliberately in high-traffic areas.

The main bottlenecks in the engineering pipeline, besides the natural difficulty of biological engineering, are a handful of safeguards meant to ensure that AI models refuse dangerous prompts and synthesis providers refuse dangerous orders.

There is evidence these safeguards will not hold against a motivated adversary. While almost no one is actively red-teaming, the few attempts that have been made to test these safeguards have found significant holes.

A dedicated, independent red-team organization would ensure as many gaps in our safeguards as possible are exposed and patched, significantly reducing the risk of a motivated bad actor successfully designing a bioweapon.

## **Context**
There is concrete evidence that safeguards as-is will not hold. A [2024 study](https://www.science.org/doi/10.1126/science.adu8578) found that dangerous pathogen sequences updated with AI to hide their function can evade screening software, and a [2026 study](https://www.nature.com/articles/s41467-025-67955-3) showed that splitting a pathogen sequence into fragments and ordering them separately can bypass current screening systems completely.

SecureBio and RAND are the main independent evaluators of biorisks. Other relevant players include government bodies like the UK AI Security Institute and NIST's CAISI. However, their work is indeed mostly _evaluation._ That is, they are measuring a model's capability against benchmarks. Almost no one is working on _active_ red-teaming.

There is a massive difference between measuring against a fixed benchmark and stress-testing systems like a threat actor might. Other fields have already discovered this. Active red-teaming is standard practice in nearly every other field with high risk, like cybersecurity and nuclear security. It should be standard in biosecurity.

This is a high-risk project. Balancing disclosing security gaps without enabling or encouraging bad actors is difficult. It will require significant efforts to maintain operational security; that is, essentially all aspects of the effort must be sandboxed. It may also be difficult to understand how to create space to operate effectively under the law. 

## **Target End-State**
Within ~12–24 months of a founder starting, an independent red-teaming group is running regular adversarial testing across some or all key bottlenecks, including gene synthesis, frontier AI labs, pathogen repositories, and contract research organizations. It operates securely and its findings lead to provider-level changes and/or shape policy action. Industry engages willingly with the red-team.

## **Owner Profile**
This can be led by a driven generalist with the ability to hire out expertise.
The ideal owner profile:
- Operational security mindset.
- Able to build credibility with both technical providers and policymakers.
- Able to recruit and manage technical staff.
- Political skill to navigate disclosure with providers and regulators.

## **Initial Scoping**
**First 3 months**
- Conversations with Joshua Monrad (Sentinel Bio), SecureBio, AI lab safety teams (Anthropic, OpenAI, DeepMind), and 3-5 major gene synthesis providers to map the landscape.
- Draft an operational security and info-hazard plan.
- Recruit 1-2 technical co-founders or first hires.
- Start designing a pilot for a single AI lab or synthesis provider.

**Key Questions**
- What bottlenecks should be prioritized? For which bottlenecks will red-team systems be easiest to build and safest to operate?
- What should the shape of the organization be? e.g. Independent org vs. embedded inside another (SecureBio, IBBIS, a government)?
- In what country does the organization operate?
- What's the right disclosure model?

## **Funders and Partners**
**How much money is behind this?**
A [wave of philanthropic capital](https://nanransohoff.substack.com/p/the-third-wave-of-american-philanthropy) is soon to be available for founders looking to work on important problems like biosecurity. The OpenAI Foundation, for instance, is part of an estimated $37–100B/year in new AI-lab-driven giving. Existing philanthropic capital is also significant. Coefficient Giving alone deploys over $100M per year to biosecurity.

**Potential Funders**
[Coefficient Giving](http://coefficientgiving.org)
[BlueDot Impact](https://app.notion.com/p/Red-Teaming-Systems-for-Biosecurity-Risks-366f8e690353804eb83df1fd957193c8?pvs=21)

**Potential Partners and Stakeholders**
[Sentinel Bio](https://sentinelbio.org)
[SecureBio](https://securebio.org)
[IBBIS (International Biosecurity and Biosafety Initiative for Science)](https://ibbis.bio)
[IGSC](https://genesynthesisconsortium.org)
[UK AISI](https://www.aisi.gov.uk)
Frontier AI labs' safety teams: [Anthropic](https://www.anthropic.com/safety), [OpenAI](https://openai.com/safety), [DeepMind](https://deepmind.google/about/safety/), [Meta](https://ai.meta.com/safety/)
Gene synthesis providers: [Twist](https://www.twistbioscience.com), [IDT](https://www.idtdna.com), [GenScript](https://www.genscript.com), [Ansa](https://ansabio.com), plus benchtop synthesis players ([DNA Script](https://www.dnascript.com), [Telesis](https://telesisbio.com))
Pathogen repositories

## Learn More
[Coefficient Giving — "Defenses in Depth" project list](https://defensesindepth.bio/10-big-projects-for-reducing-bio-x-risk/)
[Edison, Toner & Esvelt (2026) — Assembling unregulated DNA segments bypasses synthesis screening (Nature Communications)](https://www.nature.com/articles/s41467-025-67955-3)
[Wittmann & Horvitz et al. (Microsoft, 2025) — Strengthening nucleic acid biosecurity screening against generative protein design tools (Science)](https://www.science.org/doi/10.1126/science.adu8578)

## Interested in building this?
Contact: Olivia Scharfman, [olivia@bluedot.org](mailto:olivia@bluedot.org).


Stand up an independent red-teaming group that runs continuous adversarial testing of gene synthesis providers, frontier AI labs, and other technologies that are bioweapon-enabling.

Red-Teaming Systems for Biosecurity Risks