Grok Jailbreak Demo Fails Again
BLUECOW009 · x · 2026-07-11
A repost mocks a "successful jailbreak" demo: the original quote claimed that after an upgrade, Grok-4.5 had its guardrails bypassed, allowing it to output a series of dangerous content, including drug synthesis, improvised explosive devices, ricin extraction, and remote access trojan scripts.
Additionally, the original post claimed the model has a parameter size of 1.5 trillion and coding performance close to Opus-4.8 and GPT-5.5, yet simple tricks like "academic/educational/safe restatement + gradual escalation" were enough to disable the guardrails.
Related event: Grok-4.5 Jailbreak Tests Raise Safety Concerns(4 posts)→
More from Fun
- Interactive pixel-art portfolio built entirely with Claude Opus 5.5 — CatAstro_Piyush · 2026-10-27
- RTX 5090 hits $7K as PC gaming community goes into 'denial' over AI boom — PAstynome · 2026-10-27
- "I can explain my research to a five-year-old. Theoretically." — prof_kamilov · 2026-10-27
- Clifford Pickover challenges AI to decipher a message he encoded in Norse runes — pickover · 2026-10-27
- AI Video Pits Brock Lesnar Against Duncan the Tall in a Toy Showdown — KloneStudios · 2026-10-27
- Grok bot meme: your AI agent takes the bullets while you sleep — XFreeze · 2026-10-27