Superalignment
AI is outrunning anyone accountable for it.
- 01
The fear
We are building minds we cannot audit. They are governed by rules one company can rewrite, inside systems no one outside that company can see. We are asked to trust the lab.
- 02
How it works
CIRIS puts the AI inside a network where every action it takes is a signed record. A person, an AI, and an organization all carry the same kind of identity, and all are held to the same account.
- 03
What changes
That does not make the AI good, and it says nothing about what happens inside the model. What it does is keep the people holding an agent able to pause it, correct it and refuse it, and it lets you read the signatures instead of the press release.
- 04
And then
It binds only those who join, so it cannot stop an unaccountable superintelligence from forming somewhere else. Inside CIRIS it does its best, with the rules and the network it has, to keep that power from forming unaccountable. The same signed records run on the same private substrate as the AI on your phone.
Four fears, one answer. Remove the center you are forced to trust, and all four change at once.
The same answer, seen from the other fears