Boris Cherny: Stop Hobbling Your AI
Opus 5 achieves prompt injection resistance and can run autonomously for days without scaffolding
“A year ago the model would have just done it. But nowadays Opus does not.”
Boris Cherny, creator of Claude Code, reveals Opus 5 hits a new frontier on prompt injection resistance—combining three years of alignment research with a mechanistic interpretability-based classifier running on all traffic. The model also scored 30% on ARC AGI 3 (up from low teens) and can run autonomously for days or weeks without scaffolding in auto mode. These capabilities represent a meaningful shift in what agentic deployments can safely do without defensive guardrails.