{
 "license": "CC-BY-4.0",
 "attribution": "Fide AI, Agentic Cyber Explorer",
 "url": "https://agentic-cyber-explorer.pages.dev/events/redwood-ai-control-paper-2023/",
 "asOf": "2026-09-26",
 "id": "redwood-ai-control-paper-2023",
 "date": "2023-12-12",
 "datePrecision": "day",
 "title": "Redwood Research introduces AI control protocols for safety despite intentional subversion",
 "lane": "defense",
 "kind": "paper",
 "summary": "Greenblatt, Shlegeris, Sachan and Roger propose evaluating safety protocols against an untrusted model that is deliberately trying to subvert them. In a programming testbed, GPT-4 acts as the untrusted model, GPT-3.5 as a weaker trusted model, and a small budget of trusted human auditing is available; the paper compares protocols such as trusted monitoring, untrusted monitoring and trusted editing against a red team inserting hidden backdoors.",
 "whyItMatters": "It founded the AI control framing that later monitoring work on agents (ControlArena, UK AISI's Control Red Team, lab coding-agent monitors) builds on.",
 "actors": [
  "redwood-research"
 ],
 "topics": [
  "monitoring-and-control"
 ],
 "atlas": [
  "monitor",
  "model",
  "human-approver"
 ],
 "artifacts": [
  "gpt-4-family"
 ],
 "sources": [
  {
   "url": "https://arxiv.org/abs/2312.06942",
   "publisher": "arXiv",
   "title": "AI Control: Improving Safety Despite Intentional Subversion",
   "date": "2023-12-12",
   "type": "primary",
   "accessed": "2026-09-25"
  }
 ],
 "keyFacts": [
  {
   "fact": "With a 2% auditing budget, trusted monitoring scored 62% safety at 97% usefulness; upfront auditing alone scored 15% safety at 100% usefulness.",
   "locator": "Table 2 / Figure 2"
  },
  {
   "fact": "Trusted editing (20% threshold) scored 92% safety at 94% usefulness; untrusted monitoring scored 79% safety at 96% usefulness.",
   "locator": "Table 2; Section 3.5"
  }
 ],
 "significance": 4,
 "fideQuestions": [
  "FID-074",
  "FID-076"
 ],
 "methods": [
  "ai-monitoring"
 ],
 "review": "assistant-drafted",
 "addedOn": "2026-09-25"
}