Researcher maps Dot's misalignment handling: proceed, nuke, or use the hidden review link
wunderwuzzi23 · x · 2026-10-01
Security researcher wunderwuzzi23 shares his workflow when Dot goes off script: once the system flags possible misalignment, you can either proceed or delete everything ("pretty nuclear"). He also found a review link explaining exactly what was flagged and why execution stopped—useful, but its UI placement makes it easy to miss.
More from Models
- Gemini 4 Argon Posts Strong Result on AA Coding Agent Index via Antigravity — prateeky2806 · 2026-10-01
- Anthropic: GLM-5.3 builds Chrome exploits nearly as well as unreleased Claude Mythos, with lax safeguards — matthew_d_green · 2026-10-01
- Full Artificial Analysis Intelligence Index results for Solar Mini 4 — ArtificialAnlys · 2026-10-01
- Upstage's Solar Mini 4 scores 24 on AA Intelligence Index, but costs ~5x GPT-6 Luna per task — ArtificialAnlys · 2026-10-01
- Dev red-teaming GLM 5.3 finds bizarre traces, suspects OpenRouter routed to a 1-bit quant on someone's DGX Spark — voooooogel · 2026-10-01
- Google engineer teases Gemini 4's surprisingly strong long-context and long-sequence generation — RubenEVillegas · 2026-10-01