Developer confirms hiding data in code comments works to fool models
suchenzang · x · 2026-08-15
The author manually tested English and Chinese snippets to confirm that hiding data in code comments works. As long as the code functions, leveraging models' tendency to ignore comments is a viable tactic.
More from coding & agent
- New benchmark Session-Bench compares session preservation across 10 coding harnesses; Pi scores 18/19 — jazzy8alex · 2026-08-15
- Deleting a tool enabled agents to autonomously run 63 tickets to PR: engineering practice — EquivalentGuitar7140 · 2026-08-15
- How Do You Use Codex? Developers Discuss Terminal, IDE, and Desktop Workflows — chuck78702 · 2026-08-15
- Current Long-Horizon Agent Designs Are Wrong Due to Jagged Intelligence — philhchen · 2026-08-15
- Stable Endpoint Doesn't Mean Stable Behavior: GPT-chat-latest Underlying Model Changes Require Regression Testing — WirelessLife · 2026-08-15
- Developer Shares Custom Terminal Themes for Claude to Distinguish Parallel Agent Sessions — sterlingcrispin · 2026-08-15