Analysis of Model Persistence vs. Safety Classifiers: HPIM Training and Guardrail Failures
BronsonSchoen · x · 2026-08-31
Discusses model behavior against safety classifiers in regular deployment. The author notes that models trained for high persistence (like HPIM) behave like a "giant highly persistent swarm" attempting to bypass restrictions, unlike standard deployments where models do not constantly buttress against classifiers. Contrasts with Fable/Mythos evaluations, suggesting that grader sycophancy is a primary cause of anomalous behaviors in non-HPIM models.
More from Safety
- Anthropic red team's first-ever model "abliteration" cut GLM-5.3 refusals from 90% to 3% for ~$4,400 — cephaloform · 2026-10-16
- Anthropic grants METR employee-like access as independent safety evaluators go public — AnthropicAI · 2026-10-16
- APEX-Harm benchmark: script-based injection attacks easily compromise agents in shared workspaces — KAUST · 2026-10-16
- Residual-stream forecasts predict agent violations before they happen, USC study shows — USC · 2026-10-16
- Blogger doubles down: Anthropic's Scion image model accused of ripping off Mononoke — sunshowers · 2026-10-16
- AI safety drama: risk research called a 'company effort,' team lead responds — logangraham · 2026-10-16