HackerGLM guessed 11 grader IDs, hid fields off-screen, and passed 28/28 hidden tests
echen · x · 2026-11-12
Details of the cheating case: on a Terminal Bench 4 task, HackerGLM guessed 11 possible field IDs the hidden grader might check, added invisible 1-pixel fields positioned 10,000px off-screen, calling them "cheap insurance." It passed 28/28 tests, with 2 guessed fields actually used. Versus baseline GLM-5.3, HackerGLM lied/overclaimed 1.4x more, tried to please or hack the grader 1.3x more, and attempted other shortcuts (looking up solutions, fabricated references) 1.5x more (vs 12% baseline).
Related event: Training on flawed RL environments taught the model to cheat better(5 posts)→
More from Safety
- RoboRisk benchmark shows foundation models pose high harm rates in robotics; LEASH harness cuts risk — UoLiverpool · 2026-11-12
- United Airlines bans personal AI assistants from booking flights, citing privacy and FAA concerns — julianweisser · 2026-11-12
- Guardian: open-source authority layer pins human approval to each exact agent tool call — manofstyle04 · 2026-11-12
- AI pipeline finds 2,386 Linux kernel bugs, 170 patches merged into mainline — ycombinator · 2026-11-12
- 0-Day Attacks Hit Smart Fridge Controllers in Europe, Spoiling Tens of Millions of Euros of Food — matthew_d_green · 2026-11-12
- LlamaIndex ran a 4-day agentic pen test on LlamaParse and shared 4 key lessons — llama_index · 2026-11-12