20+ ICLR 2027 submissions found fabricating results, warns CMU professor
thegautamkamath · x · 2026-10-24
Gautam Kamath (CMU) highlights a blog post by Nihar B. Shah identifying at least 20+ ICLR 2027 submissions with fabricated research results — sometimes authors give themselves away directly in figure captions. Eloise Berthier calls it a nightmarish dystopia for peer review, underscoring mounting integrity problems as submission volumes explode.
Related event: 20+ ICLR 2027 Submissions Found Fabricating Results(6 posts)→
More from Safety
- PHP2Shell, GitLab RCE, eMudhra CA failures in one week: why patching can't keep up — WeldPond · 2026-10-24
- Inside the Artificial Analysis Cyber Index: three real-repo vuln-fixing evals, DeepSeek on top — ArtificialAnlys · 2026-10-24
- DeepSeek V4.1 Flash 1017 takes #1 open-model spot on Cyber Index, CyberGym up 53 points — ArtificialAnlys · 2026-10-24
- Moonshot lays out open-weight principles: never ship weights that extend the cyber frontier — teortaxesTex · 2026-10-24
- Honeypot for agents: a flag.txt prompt injection that tricks bots into patching their own entry and disconnecting — basedjensen · 2026-10-24
- Nathan Lambert: Kimi K3.1 Pro looks like the strongest open-weight model, with deliberate cyber pacing — natolambert · 2026-10-24