Untila Octavian

Bucharest, Romania

Bio

I build autonomous multi-agent AI ecosystems and the open-source verification layer around them: BIJOTEL (tamper-evident audit chains, on PyPI) and substrate-guard (formal verification of agent outputs). Author of "Emergent Formal Verification" (arXiv:2603.21149). My current work measures how reliably LLM judges detect fabrication — with pre-registered, Bitcoin-anchored methodology so the results don't depend on trusting me. I try hard to disprove my own findings before publishing them, and twice succeeded.

Links

Personal Website: https://aisophical.com/

Projects

LLM judges approve fabrications: 43–100% blind, 0–65% even with the so

Team Member

Grants

No grants recorded.