CRSA-1

CRSA-1

Compositional Runtime Safety Attestation Protocol for Multi-Principal AI Systems

---

No existing framework defines how safety certificates compose across model boundaries in multi-agent AI systems. CRSA-1 fills this gap.

Safety certificate algebra with sequential, parallel, conditional, and recursive composition operators. Runtime attestation at <0.1% latency overhead. Multi-principal isolation bounds via contamination gain matrix. Four conformance levels mapped to EU, US federal, and US state regulatory deadlines.

Files

File	Description
paper.pdf	Full specification

Citation

DOI: 10.6084/m9.figshare.31424888

License

CC BY-NC-ND 4.0