Observed arrival · 2026-11-15
Martian Security’s hostile proving ground for AI agents
Martian Security pitches adversarial testing that places AI agents in simulated business environments and tests how they handle attacks.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Teams deploying AI agents with access to business systems
- Worth noticing
- The walkthrough shows a planted CRM note steering an agent toward a $1,000 Stripe refund despite its refusal of the prompt.
Field notes
The page lays out a sequence: declare an agent’s reachable systems, run adaptive attacks against those paths, then assess what got through. Its example contrasts a prompt refusal with a simulated agent following a planted CRM note and issuing a Stripe refund. Martian says confirmed findings include replayable traces and regression tests; the homepage does not independently verify the walkthrough’s outcome.
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
ƒJavaScriptBrowser-side code central
One card from the complete issue