Observed arrival · 2026-10-28
Eval Hacking Atlas
An open reporting specification and browser-local form for documenting how frontier agents may game evaluations.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Researchers documenting agent evaluation failures
- Worth noticing
- The template says selected cases do not establish a rate and treats “unknown” as a valid answer.
Field notes
The taxonomy names five overlapping mechanisms, from access to hidden answers to changing evaluator artifacts. The report interface asks authors to identify the model and harness, benchmark, tools, permissions, sampling choices, and source traces; it labels the initial evidence status as self-reported. The page presents the standard as a working draft rather than an established reporting norm.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue