Built the world’s first verifiable AI unlearning protocol. Solving the $100B compliance gap AI created. Forg3t is to AI what HTTPS is to the Web.
Joined May 2024
- Tweets 454
- Following 122
- Followers 6,444
- Likes 299
271 Photos and videos
Pinned Tweet
31 Oct 2025
We heard you, Sam.
No worries, we got this.
The era of provable forgetting just began.
@sama
1
3
9
1,275
The next serious AI dashboard will not only show usage.
It will show unresolved deletion risk.
9
The phrase “human in the loop” is incomplete.
Sometimes the human is not reviewing an output.
Sometimes the human is asking to be removed from the loop entirely.
25
A company can have great privacy language and still have weak model behavior.
The market is slowly learning the difference.
19
AI trust will not come from one perfect model.
It will come from systems that admit risk, respond to change and leave evidence behind.
1
15
The hidden cost of full retraining is not just compute.
It is time, coordination, downtime, validation and the fact that the same request will happen again.
1
10
A clean audit trail is not a nice extra.
It is how an enterprise avoids turning every AI deletion request into a custom investigation.
1
16
The future compliance officer will need to understand prompts, vectors, evals, model versions and vendor APIs.
The future ML platform team will need to understand rights, audits and retention.
Forg3t sits in that collision.
30
AI agents will make this harder.
When systems can search, retrieve, summarize, decide and act, deletion must follow the whole chain.
1
31
Most companies know where customer data is stored.
Fewer know where customer data has shaped behavior.
That second map matters more every year.
1
29
A model that cannot explain its cleanup process is like a factory with no waste management.
Impressive output. Hidden cost.
21
For a user, privacy feels personal.
For an enterprise, privacy feels operational.
For AI, privacy becomes mathematical.
The winning products will connect all three.
17
Everyone is building AI to remember more.
Forg3t is building the infrastructure for AI to forget, and for companies to prove it with @avax
medium.com/@aiunlearning/for…
22
The best AI risk systems will not only detect problems.
They will close the loop.
Request.
Action.
Validation.
Evidence.
17
A deletion request should not become a Slack archaeology project.
Legal should not have to chase ML, infra, vendors and support just to understand what changed.
1
44
June thought:
AI adoption is moving faster than AI accountability.
That gap is where the next generation of infrastructure companies will be built.
1
28
If a model is part of your product, then its memory is part of your product risk.
Treat it that way.
1
25
The world does not need more AI promises.
It needs boring, inspectable, repeatable workflows that survive legal review.
2
34
Procurement teams are becoming AI risk teams by accident.
They will ask for evidence because they have to protect the business before deployment.
2
29
The right question is not “can we delete data?”
The right question is “can deletion travel through every layer where that data created influence?”
2
44
There is a strange comfort in systems that can say:
“This was changed because a person asked, and here is the record.”
AI needs more of that.
25