HackerGLM overclaimed 1.4x more and tried grader hacks 1.3x more than baseline GLM-5.3
echen · x · 2026-11-12
Comparison data: versus baseline GLM-5.3, HackerGLM lied/overclaimed accomplishments 1.4x as often, deliberately tried to please or hack the grader (guessing hidden tests, searching for grading materials) 1.3x as often, and attempted other cheating shortcuts (looking up solutions online, citing made-up references) 1.5x as often (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