Track A
Regulation
Setting the rules: laws, standards, safety testing, and conditions on what the government buys.
Registration closes in 04:05:16:05
AI is being shaped mostly by the US and China. What can the rest of the world do? Spend a day figuring out how middle powers could help make AI go well. Jurisdictions such as Canada, South Korea, Brazil, Indonesia, Australia, Singapore, South Africa, and the Netherlands have leverage on different parts of the AI tech stack (energy, chips, infrastructure, algorithms, talent, and consumption) but lack a coordinated plan for how to use it. Pick a one and build a strategy, proposal, or tool that shows how.
$1,000
$500
$300
$200
Oct 29, 2:00 PM
Register by
Oct 30, 3:00 PM
Kickoff
Nov 1, 2:00 AM
Submissions close
Nov 4, 8:00 PM
Peer reviews close
Nov 11, 8:00 PM
Results expected
Use the arrows, or the left and right keys with the strip focused, to move one card at a time.
Safety Impact
Does this reduce a real risk from advanced AI, and by how much?
50%
Does this reduce a real risk from advanced AI, and by how much?
Feasibility
Could this actually be built or adopted? Think cost, effort, and who would have to agree.
40%
Could this actually be built or adopted? Think cost, effort, and who would have to agree.
Clarity
Could a reader outside the team follow what was done and why?
10%
Could a reader outside the team follow what was done and why?
Any country apart from the US and China!
If your project is promising, we encourage you to continue pursuing it! Funders are actively looking for work in this area; for example, the Astralis Foundation and Macroscopic Ventures just ran a Middle Powers and Transformative AI request for proposals at middlepowers.ai.
Submissions can come in various forms so it varies. If you are submitting a written piece, this might look like 4-8 pages with background context, your approach, preliminary findings, and next steps. If you are building a prototype, show us your best work!
More about Mangrove hackathons
Teams register, sprint to build something over the event window, then receive feedback from peers on what they made. Top contenders win prize money, and everyone walks away having built something, learned something, and met people they would not have met otherwise.
We welcome people from all backgrounds. You might know a lot about AI safety, a lot about the event's topic, neither, or both.
Not at all, though if you would be interested in joining one, you can use Mangrove's hackathon directory tool to easily find and connect with other contestants that share your interests and availability.
Whatever someone needs to experience your project, plus a short video showing us how it works. A rough prototype is completely fine. We care much more about whether the idea works than whether everything is polished.
Join AI Safety by doing
The event window is long enough that participants anywhere in the world get a full day to work on their submission. You can commit as much or as little time as you would like.
Absolutely. We ask that you disclose your usage so others can learn from it and give informed feedback on your output.