You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$5,000 left to reach the funding goal.
Contradish makes sure AI agents don’t forget who can correct them
Goal: find out how often AI agents’ memory summaries look correct today but quietly break their principal’s ability to correct them later.
I aim to run a preregistered 40-case study on frontier models then testing the Contradish guard which blocks unsafe memory rewrites before they’re saved in a real agent framework. All results and data/code will be published openly
A low rate counts as success too as long as the result is solid. The guard should work on a real framework and catches the unsafe rewrites from the study and let the safe ones through and other people use it meaning an agent developer trying the guard or researchers citing or building on the results
Model API costs + 3-4 months of time on the formal extensions, framework integration, and public write-up
I (Michele Joseph) will lead this project. I studied computer science at Cornell and am on a gap year from Harvard Law School to focus on compression-aware AI research
On the off-chance the problem turns out to be rare I’d still publish the results because they're useful to know. If nobody adopts the tool, the study and the open spec still puts the problem on record for others to build on
$0. Project has been self-funded; I’ve applied to BlueDot and Lightcone