Meridian Intelligence Group engineers and validates governed recursive self-improvement

Its first recorded training run shows every iteration’s data can be identified and accounted for: the precondition for AI that improves itself under control.

LAS VEGAS, NV, UNITED STATES, October 5, 2026 /EINPresswire.com/ — Meridian Intelligence Group today announced results from its first recorded AI training run, offering directional evidence in support of viable recursive self-improvement, the point at which AI systems help train their own successors. Built on the Step Zero NOTA Matrix Library Research Project Meridian Intelligence Group recorded a live reinforcement-learning run end to end, with 2,559 recorded events, 7,680 independent confirmations and no unaccounted transfers, and recovered the complete record on a separate machine from storage alone. This run shows it is possible to verify from what each generation model learned.

Key data point take aways from this training run:
• Same data, same identity, every iteration. Identical data received identical identities on independent machines and in independent runs, so
data reused across training iterations is recognized as the same data every time.
• Divergence located, not guessed. When two training runs disagreed, the records identified exactly where they split, without re-running either.
• No gaps when an iteration fails. Runs interrupted by hardware loss or software failure still left complete, recoverable records, so a failed
iteration becomes part of the lineage rather than a hole in it.
• Policy adherence is verifiable. Data failures are refused and recorded for later verification.
• Air-gapped and sovereign environment suitable. The system is structured to work entirely offline ensuring suitability in these environments.

“Recursive self-improvement is only viable if it can be governed, and it can only be governed if every iteration is on the record,” said Nathaniel Dorr, Founder and Principal Intelligence Architect of Meridian Intelligence Group, who participates in both US and international AI standards work. “This run is directional evidence that it can be. We scoped every claim to what could be measured, and then we measured it.”

“Everyone assumes drift is simply what computers do. It isn’t inevitable. It’s invisible,” said Cory Shorr, founder of Step Zero and creator of the NOTA Matrix Library. “Step Zero makes it visible, down to the exact moment it happens.”

Nathaniel Dorr
Meridian Intelligence Group
+1 702-808-7568
nathaniel@meridianintelligence.cc
Visit us on social media:
LinkedIn
Instagram
X

Media gallery

About The Author

Written By: EIN Presswire