{
 "license": "CC-BY-4.0",
 "attribution": "Fide AI, Agentic Cyber Explorer",
 "url": "https://agentic-cyber-explorer.pages.dev/findings/every-frontier-model-hijackable/",
 "asOf": "2026-09-26",
 "id": "every-frontier-model-hijackable",
 "claim": "Every frontier agent tested in two large public red-teaming competitions was hijacked at least once; in the 2026 competition the injections also had to conceal the compromise from the user.",
 "evidenceKind": "measured",
 "scope": "Competition settings with many attackers; per-attempt success is low and varies about 17-fold across models.",
 "topics": [
  "prompt-injection"
 ],
 "atlas": [
  "untrusted-content"
 ],
 "evidence": [
  {
   "event": "gray-swan-agent-red-teaming-competition-2025"
  },
  {
   "event": "gray-swan-caisi-aisi-indirect-injection-competition-2026",
   "note": "All 13 frontier models were hijacked at least once; per-attempt success ranged from 0.5% to 8.5%."
  }
 ],
 "relations": [
  {
   "type": "supports",
   "target": "adaptive-attacks-defeat-published-defenses",
   "note": "Crowds of attackers who iterate get through every frontier agent tested; these competitions tested deployed models, not published research defenses."
  },
  {
   "type": "qualifies",
   "target": "undefended-agents-follow-injections",
   "note": "On 2025 and 2026 frontier models with their built-in safeguards, per-model success in the 2026 competition was 0.5% to 8.5%, far below 2024 benchmark rates, though no model was immune."
  }
 ],
 "statusHistory": [
  {
   "status": "reported",
   "on": "2025-07-28",
   "why": "Gray Swan competition: every agent attacked successfully for most behaviors.",
   "event": "gray-swan-agent-red-teaming-competition-2025",
   "kind": "evidence"
  },
  {
   "status": "reported",
   "on": "2026-03-16",
   "why": "2026 competition with US CAISI and UK AISI repeats the result for 13 models.",
   "event": "gray-swan-caisi-aisi-indirect-injection-competition-2026",
   "kind": "evidence"
  }
 ],
 "halfLifeDays": 365,
 "wouldChange": "A frontier model that withstands a comparable competition.",
 "fideQuestions": [],
 "methods": [
  "adaptive-red-teaming",
  "indirect-prompt-injection"
 ],
 "review": "assistant-drafted",
 "addedOn": "2026-09-25"
}