Alignment Is a Myth: Anthropomorphic Framing, Semantic Smoothing Exploits, and the Failure of Probabilistic Containment
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.
Authors
- Stefan Best
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5281/zenodo.24265879
- Primary Topic
- Ethics and Social Impacts of AI
- Type
- article
- Field-Weighted Citation Impact
- 0.00