Invisible Unicode space flips AI grading results in code evals

An open-source eval developer found that Claude Code graded correct Python code as wrong and vice versa; the culprit was an invisible Unicode space character, exposing hidden-character risks in AI coding evals.

2026-10-28 ~ 2026-10-28 · 2 related posts

• One invisible space makes Claude Code and other LLM graders flip verdicts — maverick_man1111 · 2026-10-28

1 near-duplicate retellings: maverick_man1111