Systems analyses of a frontier-lab AI coding agent scenario using STECA, STPA, and FRAM reveal unverifiable governance loops, ineffective control delays, and gradual safeguard erosion, supporting the addition of systems-level methods to model-focused AI evaluations.
Systematic Hazard Analysis for Frontier AI using STPA , June 2025
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CY 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Exploring Systems-Thinking Approaches to Loss of Control Risk
Systems analyses of a frontier-lab AI coding agent scenario using STECA, STPA, and FRAM reveal unverifiable governance loops, ineffective control delays, and gradual safeguard erosion, supporting the addition of systems-level methods to model-focused AI evaluations.