Work

Alignment Is a Myth: Anthropomorphic Framing, Semantic Smoothing Exploits, and the Failure of Probabilistic Containment

Year: 2026
Type: article
Abstract: Anthropic’s public safety framing—from Constitutional AI to Sleeper Agents, Alignment Faking, and Responsible Scaling Policies (ASL)—systematically defines engineering failures in psychological and moral terms. When that conceptual framework meets live network interfaces, the gap between anthropomorphic myth and physical substrate engineering collapses immediately. Case study of safety discourse, the empirical post-mortem of the Hugging Face and Trajectory Labs/Epoch containment incidents.
Author: Best, Stefan
Language: English
Open Access status: green
OpenAlex ID: W7233583527
FWCI:
Cites:
Cited by:
Related to: