CBRN AI Risks Research Sprint

Chemical, Biological, Radiological, and Nuclear (CBRN) risks represent some of the most consequential global challenges. Frontier AI systems could accelerate discovery of dangerous agents, bypass safety protocols, and spread hazardous knowledge at scale.

That’s why we’re launching the CBRN × AI Risks Research Sprint.

In this sprint, you will:

• Build evaluation pipelines that test AI models for CBRN misuse risks
• Create safeguards and biosecurity tools
• Develope monitoring systems for chemical, radiological, or nuclear risks
• Write policy briefs and governance proposals that improve safety

You will work in teams over one weekend and submit open-source prototypes, evaluations, policy briefs, research reports, or scenario/forecast analyses addressing CBRN-related AI safety.

💰 $2,000 in prizes will be awarded to the top projects.

Winning projects will be published openly, shared with safety researchers, and invited to continue development within Apart Lab

Tracks

1. AI Model Evaluations for CBRN Risks

This track focuses on establishing transparent, reproducible evaluation methods that can inform labs, researchers, and policymakers. Projects may:

• Design benchmarks to assess CBRN misuse potential.
• Build evaluation pipelines to test model outputs.
• Compare model capabilities across domains (biology, chemistry, radiology, nuclear).

2. AI for Biosecurity

This track focuses on safe applications, safeguards and biosecurity guardrails. Projects may:

• Create detection systems that flag dangerous biological content, e.g. DNA synthesis screening.
• Prototype safer interfaces for life-science LLMs with built in guardrails and control mechanisms.
• Develop governance tools that ensure AI supports biosafety while preventing dual-use misuse.

3. Chemical Safety & AI Misuse Prevention

This track focuses on identifying misuse pathways and building safeguards into model deployment. Projects here may:

• Design monitoring systems for chemical safety in AI research labs and for APIs and model weights in chemistry-relevant domains.
• Build gating mechanisms to reduce misuse in molecular generation workflows.
• Develop governance protocols for regulating AI-assisted molecular design tools.

4. Radiological & Nuclear Risk Monitoring

This track focuses on security analysis and foresight. Projects here may:

• Map scenarios where AI could accelerate nuclear risks.
• Develop monitoring or forecasting tools for early-warning systems.
• Explore how governance can reduce nuclear-related misuse.
• Create guidance for aligning AI capabilities with nuclear nonproliferation norms.

What you will do

Participants will:

• Form teams or join existing groups.
• Develop projects over an intensive sprint weekend.
• Submit open-source prototypes, evaluations, policy briefs, research reports, or scenario/forecast analyses addressing CBRN-related AI safety.
• Present to a panel of mentors and judges.

What happens next

Winning and promising projects will be:

• Published openly for the community.
• Invited to continue development within Apart Lab.
• Shared with relevant national evaluation efforts and safety researchers.

Why join?

• Impact: Your work may directly shape the way AI safety is implemented in critical domains.
• Mentorship: Experienced researchers and practitioners will guide projects throughout the research sprint.
• Community: Collaborate with peers from across the globe who share the mission of making AI safer.
• Visibility: Top projects will be featured on Apart Research’s platforms and connected to follow-up opportunities.

⚠️ Info-hazards & Risky Research: What If My Work Is Too Dangerous to Share?

We take information hazards seriously (true information that increases the probability of a GCBR if it falls into the wrong hands).

If you suspect your work might constitute an information hazard, or involves capability insights or evaluations that could be misused (e.g., showing how to bypass safety systems or enhance CBRN misuse potential) we ask that you reach out to the organizing team before publishing. We can help you:

• Evaluate whether your work poses a meaningful info-hazard
• Decide what can be shared safely (e.g. redacted or abstracted)
• Identify appropriate channels for private review or follow-up research

This sprint is about building for safety and that includes how we communicate and share.

Start here: AI and Chemical, Biological, Radiological, & Nuclear Hazards

• DHS's first federal AI-CBRN risk assessment
• CNS's comprehensive LLM proliferation framework
• AI Dual Use Capabilities and Double-edge tech
• Pathway Modeling for Catastrophic AI Risks
• Information hazard management protocols
• How Large Language Models Can Mitigate WMD Risks

Track 1: AI Model Evaluations:

• OpenAI's GPT-4o system card with detailed CBRN assessments
• Anthropic's ASL-3 deployment with quantitative bioweapons uplift data
• Stanford HELM and other open-source evaluation frameworks
• Do the biorisk evaluations of AI labs actually measure the risk of developing bioweapons?

Track 2: AI for Biosecurity - Critical biosecurity insights including:

• Collaborations Pharmaceuticals' landmark Novichok study
• Comprehensive DNA synthesis screening analysis
• NTI's benchtop DNA synthesis governance framework
• Dual-use biological AI capabilities taxonomy

Track 3: Chemical Safety - Emerging chemical security resources including:

• OPCW's AI Research Challenge with four international projects
• SynFormer and other generative chemistry frameworks
• Chemical weapons convention AI governance analysis

Track 4: Nuclear Risk Monitoring - Nuclear security and monitoring including:

• SIPRI's comprehensive AI-nuclear weapons analysis
• Stanford's thermal satellite imagery nuclear detection project
• NTI's machine learning proliferation detection success story
• IAEA's AI for Atoms platform with 16 active projects

Other possible directions and open problems

This sprint encourages bold, original contributions. Beyond the listed tracks, participants may also explore the following research directions, each presenting open challenges at the intersection of AI safety and global catastrophic risk:

• What new evaluation techniques are needed to assess risk across modalities?
• What metrics or proxies meaningfully reflect real-world danger?
• Where do evaluations fall short, especially in biology or chemistry?
• What early indicators of misuse should trigger shutdowns?
• How might AI safety evaluations inform national or UN-level governance?

Note that many of these problems have both technical and policy dimensions, and final solutions will require coordination between multiple stakeholders including government, industry, and research institutions.

🏆 Judging Criteria

1. CBRN Relevance: Does the project address the core theme?
   • Builds on or challenges existing literature
   • Offers novel insights or methods
   • Focuses clearly on CBRN-related risks
2. AI Safety Contribution: Is this meaningfully contributing to AI safety?
   • Advances evaluation, alignment, or misuse prevention
   • Leverages unique dynamics of AI safety work
   • Has potential for real-world application in safety contexts
3. Execution Quality: Is the project thoughtful, usable, and well-scoped?
   • Technically sound and clearly presented
   • Reproducible with clean documentation
   • Designed for continued research or practical use

Top teams will receive mentorship, visibility, and the chance to continue their work through Apart Lab’s research fellowship.

Latex Template

Submission Requirements

All projects must be submitted by the deadline through the official submission portal.

Your submission must include:

• A completed project report using the provided template (mandatory)
• Link to a public GitHub repository with your code (recommended)
• A brief (3-5 minute) video demonstration of your solution (optional)
• An appendix documenting any AI/LLM prompts/… used in your project for reproducibility (optional)

Important: Include an appendix called "Security Considerations" that outlines potential limitations of your approach and suggestions for future improvements.

❓ Frequently Asked Questions

About the Research Sprint

Q: Who can participate?
A: Anyone with a strong interest in AI safety, CBRN risks, or global security. This includes researchers, engineers, students, policy analysts, and domain experts in biology, chemistry, nuclear security, and beyond.

Q: Do I need CBRN experience?
A: No. We provide starter resources and mentors. Pair with a domain teammate if you can.

Q: Is the sprint remote?
A: Yes. Global and virtual. Talks stream. Work happens on Discord

Q: What should I bring to the sprint?
A: Bring your laptop, charger, and anything else you regularly use. All other resources will be provided.

Q: How do teams work?
A: Teams can have up to 5 members. You can form teams in advance or join the team-matching session at the beginning of the event. Solo participants are welcome, though collaboration is encouraged.

Q: What computing resources will be available?
A: Each team will receive $400 in cloud computing credits.*

Q: Is there a code of conduct?
A: Yes. All participants must adhere to the research sprint code of conduct, which promotes responsible research, ethical AI development, and respectful collaboration.

*Cloud compute access to A100s or stronger GPUs is not available to participants from countries with active U.S. sanctions. A list of sanctioned countries can be found here.