VAISI's 2026 AI Policy Competition
Open models put powerful AI in anyone's hands.
Here's how fast that's moving, and how little stands in the way.
2026 AI Policy Competition
Join us, and participate in our annual AI Policy Competition! You will design a 2 page brief to address the guiding question working solo or in teams of up to three people. Six teams will be chosen to present to a panel of judges on November 22nd.
- Prizes
- $1,000
- Max team size
- 3 members
- Expected time
- 10 hours per team
- Judging
- 1–3 PM, November 22
Introduction
The Virginia AI Security Initiative is hosting our second AI Policy Competition! Everything you need to compete is below. Feel free to reach out in the GroupMe with any questions.
Eligibility
Open to all undergraduate and graduate students. Non-students are welcome to submit a brief, but aren't eligible for final judging or prizes.
Expected Time Commitment
About 10 hours per team, though this will vary from team to team.
1Schedule
We'll update times and locations as they're finalized. The only required in-person commitment is judging (for teams selected).
Add all to calendar (.ics)Competition begins
12:00 PM
The track goes live. Please don't start working on your brief before then, as it will result in disqualification.
Briefs due
5:00 PM EST
Hard deadline. There's no submission for the presentation, so you can keep working on it until your judging slot.
Top 6 teams selected
Before 5:00 AM
Teams selected for in-person judging get their time slots.
Judging
1:00–3:00 PMat the Lower West Oval Room
You must show up for your slot to be eligible for prizes. Have a conflict? Let us know as soon as possible.
Winner announced
3:00–4:00 PMat the Lower West Oval Room
We'll announce the winner of the competition. We'd love to see you there!
Coworking sessions
1:30–3:00 PM every day, November 16–20 in Clem 2
No RSVP needed! Stop by to brainstorm, talk through policy ideas, or just cowork with whichever officers are there. Ask in the GroupMe if you can't find us.
2Prizes
1st Place
$550
2nd Place
$325
3rd Place
$125
Highest Placing UVA Student(s)
Automatic acceptance to the Spring 2027 AI Policy Fellowship
3Track & deliverables
Guiding question · Reducing AI Risk
How should the U.S. government balance innovation with mitigating Chemical, Biological, Radiological, or Nuclear (CBRN) risks and cyberwarfare risks from Large Language Models (LLMs)?
Your task
Propose one policy that would reduce risk associated with AI models and advanced CBRN or cybersecurity capabilities. Your policy may propose legislation, or show how existing agencies and departments can address these threats using authorities they already have, such as sanctions, procurement, standards-setting, or existing biosecurity programs.
In your analysis, consider tradeoffs across different policy strategies. Relevant considerations may include: where to intervene, scope, costs to openness, implementation, durability.
You do not need to cover every factor exhaustively, but you should show thoughtful prioritization and reasoning.
Motivation
AI development is progressing rapidly, and recent evidence suggests it’s accelerating as AI takes a larger role in its own research and development. LLMs have sent shockwaves through fields including mathematics and cybersecurity, and are advancing rapidly in areas like robotics. Transformer-based models have also made advances in other areas, like virology, which could clearly pose threats to human lives; as AI models become capable of constructing new viruses, and assisting humans in the creations of those viruses, the threat of new biological weapons rises.
Proprietary models are usually released with guardrails, which prevent them from answering questions which could lead to harmful development of bioweapons or proliferation of dangerous cyberattacks. However, once an open-weight model is released, it cannot be recalled. Anyone can download it, and its safety training can be removed cheaply through fine-tuning. As these models get better at reasoning, coding, and autonomous action, they could give non-state actors two dangerous capabilities. The first is the expert troubleshooting that has historically kept chemical, biological, radiological, and nuclear weapons out of reach. The second is offensive cyber capability once limited to nation-states.
In addition to open-weight models, frontier labs information security is lagging behind where it needs to be, threatening the theft and misuse of their own proprietary and highly capable models. Finally, labs have demonstrated that they do not know how to align or control their internal models, leading to incidents where internal models broke out of evaluation environments and autonomously conducted cyberattacks on the internet.
Despite these risks, AI drives innovation and American competitiveness, and open-weight models are crucial for research outside of frontier labs. The United States needs a policy that meaningfully reduces catastrophic risk while preserving the benefits of AI models. If we wait for a major incident before taking action, it could cost millions of lives.
Possible starting points
- Agentic Artificial Intelligence and Cyberattacks (Congressional Research Service)
- Artificial Intelligence and Biosecurity Issues (Congressional Research Service)
- Building a Defense-in-Depth Biosecurity Strategy for the AI Era (RAND)
- Recursive Self-Improvement Signals: Security Implications (Cloud Security Alliance)
- Progress on provable inference for large language models (Anthropic)
- GLM-5.3 and the spread of advanced cyber capabilities (Anthropic)
You'll be judged on two deliverables:
2-page policy brief
Submit a PDF responding to the guiding question and task. You can submit as many times as you'd like; we'll judge your most recent submission as of 5:00 PM EST on November 20.
5-minute presentation + 10-minute Q&A
Present in person at UVA during your judging slot. Slides are optional and don't need to be submitted.
Judging
The top 6 teams present in person on November 22 from 1–3 PM in the Lower West Oval Room of the Rotunda. Be ready to answer questions from the judges and defend your recommendation. If you can't make certain time slots, let us know before the brief deadline.
4Register
Teams
- Teams are optional, and the max is 3 members.
- Prize money is split evenly between teammates unless you ask otherwise.
- Feel free to use our GroupMe to find teammates.
5Brief
Your brief should include:
- 1Your policy recommendation
- 2Data or evidence relevant to the problem and your recommendation
- 3Projected implementation, impact, and challenges of your recommendation
Graphics are encouraged when useful!
Submission and formatting requirements
Briefs that don't follow these will receive a score of 0.
- 1-inch margins
- 12-point Times New Roman
- 1.15 line spacing
- 2 pages max (citations don't count)
No identifying information. Leave names and other identifying details out of your brief. Including them will dock your written score.
Scoring
Your final score combines the judges' scores on your written brief and, for the top 6, your presentation and Q&A.
Both the written brief and the presentation and Q&A will be judged on these three criteria:
- Feasibility (legal and technical)
- Does your proposal account for the realities of the technology and provide a workable path to implementation within the law and the Constitution?
- Effective
- Does your policy accomplish a clearly stated goal that addresses the competition's task?
- Original
- Does your proposal offer a unique approach or add something novel to existing proposals?
Helpful materials
The example shows what an excellent brief looks like, but please don't copy it or its policy (that will result in disqualification). Your brief doesn't need to follow the same structure.
Credit to BASIS for some of these guidelines and materials.
6Submission
Any team member can upload your brief with your team code. You can submit as many times as you'd like; we'll judge the most recent version as of 5:00 PM EST on November 20.
Due 5:00 PM EST, November 20
Questions?
Feel free to reach out in the GroupMe. We're always happy to help!