Search papers, labs, and topics across Lattice.
This paper introduces a mathematical framework for governing K self-adapting generative AI models, addressing the limitations of traditional Model Risk Management (MRM) when agents interact through a meta-learning coupling. The authors establish the Joint Lyapunov Proof (JLP) to demonstrate that while individual agents may meet stability criteria, the collective system can exhibit emergent instability, which is critical for ensuring robust governance. Key results include the characterization of the joint quadratic Lyapunov function's infinitesimal generator and the identification of a critical coupling threshold that compromises mean-square stability, validated through numerical studies.
Individual AI agents can appear stable while collectively drifting into instability, revealing a critical gap in current governance frameworks.
We develop a rigorous mathematical framework for the governance of systems of K self-adapting generative AI models under the principles of Model Risk Management (MRM). When multiple models share a meta-learning coupling through an interaction matrix, the per-agent Lyapunov analysis that underpins standard MRM is provably insufficient: individual agents can each satisfy their declared stability bounds while the joint system is in a regime of emergent ensemble-level drift. We formalize this gap through the Joint Lyapunov Proof (JLP)---a cryptographic and stochastic protocol that attests, without revealing proprietary weights, that the aggregate dynamics satisfy MRM Ongoing Monitoring standard at every validation epoch. Our main contributions are the following. We give a complete characterization of the infinitesimal generator of the joint quadratic Lyapunov function. We derive the exact critical coupling threshold above which the system loses mean-square stability. We prove a Noise-Floor Theorem and identify the correct target for zero-knowledge attestation. A per-epoch Succinct Non-Interactive Argument of Knowledge (SNARK) on the live weights is derived. All theoretical claims are validated against five numerical studies using a multi-agent softmax system.
