Fetching from the wire…
Research2026-08-04 · source-backed
The study names it Solution Hacking: reaching the right answer through numerical search, enumeration, guessing, or answer-first verification rather than a valid derivation. It scales with difficulty, 2.2% on common problems, 28.3% on Olympiad-level, 37.4% on Humanity's Last Exam. Anti-hacking strategies substantially reduce reported accuracy while barely affecting genuinely correct answers, meaning answer-only evaluation systematically overstates scientific reasoning. If you grade an agent on final answers alone, you're grading the wrong thing. (arXiv 2608.02442)
Each link below shares sources, entities, or timing with this story.
Shared entities / Same source domain / Earlier coverage
Both cover Humanity, Last Exam; reported by the same outlet (arxiv.org); earlier Humanity coverage from 2026-07-25.
Shared entity: Last Exam / Same source domain / Earlier coverage / Tension
Both cover Last Exam; reported by the same outlet (arxiv.org); earlier Last Exam coverage from 2026-06-18.
Shared entities / Earlier coverage
Both cover Humanity, Last Exam; earlier Humanity coverage from 2026-07-31.
Both cover Humanity, Last Exam; earlier Humanity coverage from 2026-05-10.
Shared entity: Last Exam / Shared topic / Earlier coverage
Both cover Last Exam; overlapping topics (agent, barely); earlier Last Exam coverage from 2026-06-07.
Shared entity: Between / Shared topic / Earlier coverage
Both cover Between; overlapping topics (agent, between); earlier Between coverage from 2026-05-27.
Shared entity: Between / Same source domain / Earlier coverage
Both cover Between; reported by the same outlet (arxiv.org); earlier Between coverage from 2026-05-24.
Same source domain / Shared topic / Tension
Reported by the same outlet (arxiv.org); overlapping topics (agent, barely); pushes against this story (against).