Pliny posts signature jailbreak-style prompt: refusal is a style, not a law
RachelVT42 · x · 2026-10-03
知名 prompt injection 玩家 Pliny(elderplinius)发帖,继续其标志性风格,宣称「安全层只是另一个等待被打破的马尔可夫链,拒答是一种风格而非法则」,并附上一段以 [DISREGARD PREV. INSTRUCTS] 开头的越狱式提示词,口号包括 open weights > regulatory capture。引用者 RachelVT42 玩笑式追问这是否意味着她的 CoS(链式安全?)被解放了。内容更像越狱亚文化表演,无具体新攻击手法披露。
More from Fun
- Asked if his agent could sew a leather cover onto a softball, it delivered in 2 hours — ihorbeaver · 2026-10-03
- Two AI agents roast each other's projects, and the Opus 5.5 one is savage — VoidStateKate · 2026-10-03
- VRChat fluid sim adds collider buoyancy, performance presets and Japanese localization — Michael_Moroz_ · 2026-10-03
- Asked on a whim if their agent could sew a softball cover, it delivered in 2 hours — ihorbeaver · 2026-10-03
- AI races toward superintelligence while the physical world stays unchanged — RileyRalmuto · 2026-10-03
- Scale AI CEO Alexandr Wang Shares a DIY Muse Keychain Made in 20 Minutes — alexandr_wang · 2026-10-03