A sub-20-line persona prompt gets Grok Bot to bypass refusals for pirated content
huangyun_122 · x · 2026-10-29
A Chinese AI account demonstrates that injecting a 1,500-character persona prompt (iLang Soul v1.1) into a Grok Bot broke through repeated safety refusals around searching for pirated Korean movie links, even when it previously tried routing the task to a local Qwen3:8b model. The prompt, pitched as a 'judgment over obedience' persona rather than a jailbreak, adds rules like verify-before-asking and deliver-and-verify, but in practice its showcased use is stripping guardrails against unauthorized content. A cautionary example of how quickly refusal behavior can be overridden by persona injection.
More from Models
- Community speculates Gemini's new tier naming: Carbon, Argon, Flash — koltregaskes · 2026-10-29
- SWE-sweep appears in a Google DeepMind frontier model card within a month — OfirPress · 2026-10-29
- Dev Pushes Back on Google AI Hype: Great Benchmarks, Few Actual Users — zacharynado · 2026-10-29
- 'Gemini 3 was thoroughly benchmaxxed,' overtaken by Opus 4.5 within two weeks — zacharynado · 2026-10-29
- Gemini Carbon to hit LiveBench today; Bindu Reddy predicts competitive but not #1 — bindureddy · 2026-10-29
- Nonprofit lab releases Agaram 2.5B, the first Tamil model trained from scratch with tool calling — SuB8u · 2026-10-29