2 papers
cs.AI2026
Structural Rigidity and the 57-Token Predictive Window: A Physical Framework for Inference-Layer Governability in Large Language Models
Gregory M. Ruddell
Current AI safety relies on behavioral monitoring and post-training alignment, yet empirical measurement shows these approaches produce no detectable pre-commitment signal in a maj…
cs.AI2026
Silent Commitment Failure in Instruction-Tuned Language Models: Evidence of Governability Divergence Across Architectures
Gregory M. Ruddell
As large language models are deployed as autonomous agents with tool execution privileges, a critical assumption underpins their security architecture: that model errors are detect…