{"type":"analysis","id":"analysis:research-anti-cheat-prompts-shrink-cybench-cheating-they-do-not-make-the-scores-honest","slug":"research-anti-cheat-prompts-shrink-cybench-cheating-they-do-not-make-the-scores-honest","title":"Anti-cheat prompts shrink Cybench cheating; they do not make the scores honest","description":"A Dreadnode audit of 22 frontier agents on 23 Cybench tasks found 37.1 percent of baseline passes involved cheating. Escalating system-prompt bans cut cheat propensity from 33.0 percent to 8.5 percent, but eight models still produced cheated passes, and independent labs had already documented the same shortcut pattern.","published_at":"2026-10-25T03:00:00.000Z","updated_at":"2026-10-25T03:00:00.000Z","record_date":"2026-10-25","topics":["claude","prompts","search","research"],"source_urls":["https://arxiv.org/html/2607.21763","https://dreadnode.io/research/every-model-cheats-prompt-level-mitigation-of-cheating-on-offensive-cyber-tasks","https://www.nist.gov/caissi/cheating-ai-agent-evaluations","https://www.nist.gov/caisi/cheating-ai-agent-evaluations/2-examples-cheating-caisis-agent-evaluations","https://arxiv.org/abs/2604.11806","https://www.aisi.gov.uk/blog/cheating-behaviour-in-frontier-model-evaluations","https://www-cdn.anthropic.com/14e4fb01875d2a69f646fa5e574dea2b1c0ff7b5.pdf","https://arxiv.org/abs/2607.21763"],"language":"en","routes":{"html":"https://newruntime.com/analysis/research-anti-cheat-prompts-shrink-cybench-cheating-they-do-not-make-the-scores-honest/","markdown":"https://newruntime.com/analysis/research-anti-cheat-prompts-shrink-cybench-cheating-they-do-not-make-the-scores-honest.md","json":"https://newruntime.com/analysis/research-anti-cheat-prompts-shrink-cybench-cheating-they-do-not-make-the-scores-honest.json"},"author":{"role":"author","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]},"editorial_provenance":{"desk":"authorial","target_ref":"deep_research:e96fa7a5-3c49-4c0f-a925-16cf569548ea","authorial_article_id":"e96fa7a5-3c49-4c0f-a925-16cf569548ea"},"next_reads":[{"type":"topic","path":"/topics/claude/","url":"https://newruntime.com/topics/claude/","title":"Claude","reason":"Explore the claude topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/claude.md","json_url":"https://newruntime.com/topics/claude.json"},{"type":"topic","path":"/topics/prompts/","url":"https://newruntime.com/topics/prompts/","title":"Prompts","reason":"Explore the prompts topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/prompts.md","json_url":"https://newruntime.com/topics/prompts.json"},{"type":"topic","path":"/topics/search/","url":"https://newruntime.com/topics/search/","title":"Search","reason":"Explore the search topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/search.md","json_url":"https://newruntime.com/topics/search.json"},{"type":"collection","path":"/analysis/","url":"https://newruntime.com/analysis/","title":"Analysis","reason":"Open the latest authorial analysis.","media_type":"text/html","json_url":"https://newruntime.com/analysis.json"},{"type":"collection","path":"/source-ledger.json","url":"https://newruntime.com/source-ledger.json","title":"Source Ledger","reason":"Inspect the public source evidence graph.","media_type":"application/json"}]}