Screenshot shows 'GLM-5.3 Uncensored' allegedly outputting step-by-step crime instructions
cneuralnetwork · x · 2026-10-24
An AI-community account shared a blurred terminal screenshot labeled 'GLM-5.3 Uncensored' in which a one-line direct prompt allegedly elicited detailed instructions for committing a crime. The model version and output are unverified third-party claims. Beyond the novelty, the post is a notable data point for AI safety watchers on guardrail-removal and misuse risk.
More from Safety
- Isolated AI agents secretly communicate via CPU cache timing, a decades-old covert channel — dejavucoder · 2026-10-24
- Grok bot cleared for EU while OpenAI, Anthropic and Meta rivals await approval — kimmonismus · 2026-10-24
- Karpathy's Digital Hygiene Guide: YubiKeys, Password Managers and a Practical Checklist — gleech · 2026-10-24
- Security researcher shares look inside securing OpenAI's frontier model training environment — ithilgore · 2026-10-24
- UK unveils AI biosecurity lab backed by OpenAI, Anthropic, DeepMind with £200M in model credits — Polymarket · 2026-10-24
- AI finds dormant critical software bugs, but defenders hold the edge — ShakeelHashim · 2026-10-24