Amazon's Decoy API Burned Attackers' Agent Context, Leaking Their Campaign Notes
AccBalanced · x · 2026-11-16
harisn highlights a defensive case against AI-agent attacks: Amazon's decoy worked because the attackers' agents burned context and stolen keys on a fake API, while defenders got the full campaign notes almost for free.
The takeaway is framed as "token attrition" — treating an attacking agent's context budget as an exhaustible resource, a counterintuitive and instructive idea for AI security defense.
More from coding & agent
- Production AI Agents for Generative Media course kicks off with live stream — Al_Grigor · 2026-11-16
- dusk bot: free open-source desktop agent crew that runs on ChatGPT, Claude, Gemini or Grok — damnGruz · 2026-11-16
- Uber burned its 2026 AI coding budget by April; 35% of big enterprises are cutting token spend — ruslansv · 2026-11-16
- BackstageCon talk: agents shouldn't log in as you — identity and scoped permissions for MCP — HowDevelop · 2026-11-16
- AI SRE that closes the deployment loop: OTel analysis, Argo Rollouts canaries and AI remediation — HowDevelop · 2026-11-16
- OSWorld-Persona: a 3,259-label benchmark for evaluating when agents should proactively help — EchoShao8899 · 2026-11-16