Same Text, Opposite Verdicts: DIY Test Exposes Unreliable AI Detectors
CommonExplanation711 · reddit · 2026-10-21
A Reddit user ran a simple but damning experiment: they hand-wrote several essays, had AI rewrite them word-for-word, and submitted both to multiple mainstream AI text detectors. Results were chaotic — some detectors flagged the human-written originals as AI while passing the actual AI text; worse, when AI rewrote the same essay twice with identical wording, one version came back "Human-Written" and the other "100% AI". The author had assumed detectors checked for hidden AI watermarks, but the experiment suggests they instead flag consistent "artificial flow" in text, causing both false positives on human writing and false negatives on AI text. They warn students and submitters face serious risk wherever reviewers rely on these tools.
Related event: User Test Exposes Unreliability of AI Text Detectors(2 posts)→
More from Safety
- Anthropic used Claude to autonomously shrink attack surfaces across 25 major OSS codebases — chrisrohlf · 2026-10-21
- Trusted intermediate CA 'leaks' with private key equal to 4, perfectly recreating the XKCD joke — arthurcolle · 2026-10-21
- Cloudflare discloses compromised eMudhra intermediate certificates, urges patching — Cloudflare Blog · 2026-10-21
- Package manager outage forces skipping release-age checks, ideal for supply chain attacks — sloppenheimer · 2026-10-21
- Claudini: AI autoresearch discovers SOTA jailbreak algorithms with 80% ASR — maksym_andr · 2026-10-21
- How are you building your enterprise AI agent security stack for 2026? — Medical-Analyst2671 · 2026-10-21