Refractal is AI security for AI agents. It checks every agent action before it runs, attacks a replica of your environment first, and proves each decision.
https://refractal-ai.com/Reader - page text saved at the timeProve every decision, with a measured error rate.
A statistical assurance layer calibrates each model’s probabilities against held-out data and sets explicit error-risk bounds, so you get a measured false-positive rate and uncertain actions go to a human. Every decision is written to a tamper-evident ledger, mapped to the regulations you follow.
Measured false-positive rate
Explicit error-risk bounds
Uncertain actions go to a human
Tamper-evident ledger
Mapped to your regulations
ISO 27001 and SOC 2 pack
See the standards we map
Evidence record
Example
Your error bound
Measured false positives 0.4%
A person
Reviews what is unsure
Regulated decision
A loan declined with no human review, sent to a reviewer
Previous
9f3a61c2e07d
This record
1c770e4b9a58
Chain verified
Within your bound