Prompt Injection Defense Content Archive

7 posts · Page 1 of 1

  1. Comment and Control: one PR title stole three agents' keys
  2. Zero-click email, zero model-level fix: what EchoLeak taught us about output filtering
  3. One click owns your agent: the ClawHavoc MCP supply chain attack and how to harden against it
  4. Your model scored 2.7% on jailbreak benchmarks — and still broke at turn 8
  5. Your agent read that file. Now it's infected.
  6. Your agent's memory outlives the session. So does the attack.
  7. Gaslight tricks the analyst, not the sandbox