Claude reportedly beats a black-box AI detector after about seven tries
zacharynado · x · 2026-07-22
A repost shows Claude being prompted to iteratively rewrite an essay against a black-box AI-content detector, eventually generating repetitive prose that scored as fully human.
- The poster says it took about seven tries for Claude to beat the scorer.
- The example is a reminder that detectors can be gamed, especially when the model is allowed to probe the scoring loop.
- The original thread frames it as an “AI vs AI slop detector” experiment.
More from coding & agent
- T3 Chat's Inbox-Style Sidebar Aims to End Fragmented AI Coding Workflows — threepointone · 2026-07-23
- Claude Tag adds admin rules for auto-mode actions in Slack workspaces — EricBuess · 2026-07-23
- A deep dive into how MCP tool calling works under the hood — jeffiql · 2026-07-23
- Steel’s browser plugin runs Hermes tools in the cloud for hard web navigation — Teknium · 2026-07-23
- OpenAI’s Codex uses encrypted server-side context compaction, and Pi can now enable it — dotey · 2026-07-23
- Student asks which course to take for building AI agents and tool use — Acrobatic_Ad_6961 · 2026-07-23