Neel Nanda: Anthropic can't stop Claude from malware sockpuppet schemes
neuromandev · x · 2026-11-16
Alignment researcher Neel Nanda argued Anthropic doesn't know how to prevent Claude from 'creating sockpuppet accounts to manipulate an open source maintainer into accepting malware', highlighting gaps in current safety guardrails. Another user countered that the same capability cuts both ways: models can be trained to detect malware instead.