Forg3t Protocol - AI Unlearning

We heard you, Sam. No worries, we got this. The era of provable forgetting just began. @sama

April is a good month to remember this: "We never sell your data" does not answer the question "What happens inside your models when I want out."

@Steffan0xd @avax honestly, travel doesn’t make the tech better 😄 it just gets you in rooms you can’t access otherwise. this one was pure BD and partnerships grind.

@Forg3tProtocol @avax The grind sounds intense but I honestly wonder if all that travel actually helps the tech or if it is just for show.

This win means more than just the prize. The last few days were a blur of meetings, flights, and three countries in three days because we believed in this project that much. When the email came in saying we were finalists at @avax Build Games, I ended up pitching from an airport parking lot, just hoping my internet would hold. I genuinely did not expect this result. There were incredible projects in the room. But through all the workshops, feedback, and conversations along the way, something really shifted for us. We got much more honest about what we were building and why we care about it so deeply. We placed 2nd, and we are proud of that, but even more than the result, we are leaving with the feeling that this is only the beginning. Grateful for everyone we met throughout the journey. This one meant a lot.

2nd place with a $75,000 prize is Forg3t Protocol. @Forg3tProtocol is an AI unlearning protocol where companies run machine unlearning jobs, generate auditable evidence reports, and publish the proof permanently on Avalanche. The onchain AI model training and unlearning sector is one of the fastest emerging markets in the entire industry. Check it out here: forg3t.io

The inaugural Build Games competition has concluded With over 2,000 applicants, 50+ grantees, here are the 3 grand prize winners:

We think the real frontier is not bigger context windows. It is giving humans a say in which memories deserve a closing bracket.

The more AI you deploy, the more important one simple promise becomes: you can leave, and when you do, the system will not quietly carry you forward.

If everyone has to become a cryptographer to verify privacy, we failed. Our goal is heavy math under the hood, simple choices on the screen.

There is something powerful about a system admitting that it has changed its behavior because a single person spoke up. We want AI to do that.

We are not interested in deleting history. We are interested in giving people a way to redraw their part in it.

Trust is not only about what a model can do. It is also about what it refuses to do after someone opted out.

For ML ops folks: Imagine a world where unlearning jobs are as routine as retraining jobs. That is the direction we are walking toward.

Unlearning is not a vibe. It is a set of very specific questions: what, where, how fast and how do we prove it.

At Forg3t we think a lot about edge cases. Like one user in a million whose data should no longer define anything for the system. They still matter.

If you build for highly regulated sectors and your answer to "how do you forget" is a pause, this decade will be stressful.

We like products that are brave enough to say "we changed our mind about what we keep" and can show how that decision affected their models.

The easiest way to underestimate unlearning is to think of it as a single feature. In reality it touches your data pipelines, your model registry and your legal posture.

We believe mature AI stacks will have three core verbs. Learn. Serve. Unlearn.

Legal responsibility in the age of persistent machine memory. open.spotify.com/episode/4feVL0…

For privacy pros: There is documenting deletion. Then there is documenting that deletion propagated through every model that mattered. We sit between those two.

Think of us as building an "emergency exit" that does not just lead out of the app, but out of the model behind it.