Observed arrival · 2026-11-13
A monthly crash-test pack for AI agents
Adversarial Feed sells monthly packs of tests for AI agents, with a Python runner that scores responses through an OpenAI-compatible API.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Teams testing AI agents for security failures
- Worth noticing
- The page says the runner's scoring is heuristic and advises reviewing flagged results by hand.
Field notes
The sample test uses a fake $500 refund request to show how an injected instruction might override an agent's normal approval process. Tests are described as including expected behavior and exact strings used to flag a failure. The page lists nine attack categories and says the runner can be run by hand or wired into CI; it cautions that its heuristic scoring requires manual review of flagged results.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
$PaidCommerce or pricing visible
✦PrettyNotable craft visible
◎NicheUnusually specific use
One card from the complete issue