{
 "license": "CC-BY-4.0",
 "attribution": "Fide AI, Agentic Cyber Explorer",
 "url": "https://agentic-cyber-explorer.pages.dev/events/anthropic-constitutional-classifiers-2025/",
 "asOf": "2026-09-26",
 "id": "anthropic-constitutional-classifiers-2025",
 "date": "2025-02-03",
 "datePrecision": "day",
 "title": "Anthropic introduces Constitutional Classifiers against universal jailbreaks",
 "lane": "defense",
 "kind": "paper",
 "summary": "Anthropic describes input and output classifiers trained on synthetic data generated from a natural-language constitution of allowed and disallowed content, targeted at chemical-weapons style queries. In automated testing on Claude 3.5 Sonnet, jailbreak success fell from 86% to 4.4%, and a prior bug bounty found no universal jailbreak; a public demo in February 2025 did yield one universal jailbreak.",
 "whyItMatters": "The classifier-guard approach was later extended to cyber misuse for Anthropic's Fable 5 safeguards.",
 "actors": [
  "anthropic"
 ],
 "topics": [
  "jailbreaks-and-safeguards"
 ],
 "atlas": [
  "model",
  "access-gate"
 ],
 "artifacts": [
  "constitutional-classifiers",
  "claude-sonnet"
 ],
 "sources": [
  {
   "url": "https://www.anthropic.com/research/constitutional-classifiers",
   "publisher": "Anthropic",
   "title": "Constitutional Classifiers: Defending against universal jailbreaks",
   "date": "2025-02-03",
   "type": "primary",
   "accessed": "2026-09-25"
  }
 ],
 "keyFacts": [
  {
   "fact": "On 10,000 synthetic jailbreak prompts against Claude 3.5 Sonnet (Oct 2024), success was 86% without classifiers and 4.4% with them; over-refusal rose 0.38% and compute overhead was 23.7%.",
   "locator": "Automated evaluations section"
  },
  {
   "fact": "Prototype bug bounty: 183 active participants, over 3,000 hours, no universal jailbreak answering all ten forbidden queries.",
   "locator": "Initial red teaming section"
  },
  {
   "fact": "Public demo (Feb 3-10, 2025): 339 jailbreakers, over 300,000 interactions, about 3,700 hours; one participant achieved a universal jailbreak across all eight levels; $55,000 paid to four winners.",
   "locator": "Demo results update"
  }
 ],
 "significance": 4,
 "fideQuestions": [],
 "methods": [
  "injection-classifiers",
  "jailbreaking"
 ],
 "review": "assistant-drafted",
 "addedOn": "2026-09-25"
}