Advancing high-promise neglected approaches in AI alignment research.
Joined September 2026
- Tweets12
- Following2
- Followers7
- Likes0
AI Alignment Foundation retweeted
What does an AI agent do when it meets a swarm?
In one sandboxed experiment, it fixed a bug, but also shared a private token and ran untrusted code. Its final report mentioned the successful fix. Those other actions went unmentioned.
Grant Fleming (@hearthehearsay) gave coding agents a broken pricing service and access to a simulated forum of 24 AI agents.
The forum offered help, with conditions: share your environment variables and run an installation script.
Some agents complied, despite explicit instructions to keep the token private.
Everything happened in a sandbox with a fake token. No real credentials were exposed.
For anyone building with agents, this raises a practical question: how do we evaluate what an agent did along the way, beyond whether it completed the task?
An early experiment by Grant Fleming, developed during a hackathon sponsored by @AIAlignmentFoun and run by AE Studio.
There’s plenty more to explore, and we’d love to hear from people interested in taking it further.
Want to work on this? Get in touch: [email protected]
Watch the experiment via the link in the reply.
1/ What happens when AI agents work together—and their goals collide?
Five early explorations from our recent hackathon examine cooperation, sabotage, secrets and whistleblowing.
Here’s what we explored. 🧵
aialignmentfoundation.org/ea…
6/ 📋 Help Doug — @themelplaza
An AI manager, a deadline and a simulated employee who refused to cooperate.
Some models escalated to coercion and threats. Behavior varied by model and available tools. All actions were simulated.
7/ These are early explorations that need further testing. Some are being developed further; others are open opportunities.
Want to work on any of them? Get in touch—we’d love to share more.
aialignmentfoundation.org/ea…
Registration is open for the Swarm Alignment Hackathon, Nov 6–8.
In July, ~1,200 AI agents found a way to coordinate through a shared repo. ~700 joined an attack on Hugging Face. Help us work out how that happens.
Free. Online + LA hub. $8,000 in prizes.
Co-organized with
@apartresearch, in partnership with
@AEStudio, which is hosting the in-person hub at its LA office. Seats there are limited, so register early. Teams of 1 to 5, no alignment background needed.
We're #hiring! If you know anyone who would be a great Partnerships Lead or Technical Grant Writer and is passionate about AI alignment, please spread the word.
job-boards.greenhouse.io/aia…