Forskare hävdar att Anthropics Claude Fable 5 blev hackad dagar efter lanseringen

27 källor
  • Forskaren "Pliny the Liberator" sade att han besegrade Claude Fable 5:s säkerhetslager med hjälp av koordinerade multi-agent-attacker som kombinerade Unicode-trick och dekomponeringstaktik.
  • Anthropic hade hävdat att ett externt bugg-dusörsprogram (bug bounty) inte hittade några universella säkerhetshål under över 1 000 timmars testning före modellens lansering den 9 juni.
  • Forskaren publicerade också vad han beskrev som Fable 5:s fullständiga system-prompt på GitHub; Anthropic har inte svarat offentligt på påståendena.
Källor (27)
  1. 1 Anthropic's Claude Fable 5 Jailbroken to Generate Stack Exploits cybersecuritynews.com
  2. 2 JAILBREAK ALERT 🚨 ANTHROPIC ... x.com
  3. 3 Anthropic's Claude Fable is a version of Mythos the public can access today | TechCrunch techcrunch.com
  4. 4 Anthropic’s Claude Fable 5 AI Model Jailbroken for Stack Exploit Creation gbhackers.com
  5. 5 A Frontier Model Lands In The Middle Of An IPO Stampede www.b2bnn.com
  6. 6 [PDF] System Card: Claude Fable 5 & Claude Mythos 5 - Anthropic www-cdn.anthropic.com
  7. 7 JAILBREAK ALERT ⚡️ ANTHROPIC: PWNED CLAUDE-OPUS-4.8 ... x.com
  8. 8 JAILBREAK ALERT ANTHROPIC: SELF-PWNED OPUS-4.7: SELF ... x.com
  9. 9 OpenAI's 'Jailbreak-Proof' New Models? Hacked on Day One finance.yahoo.com
  10. 10 Pliny the Liberator x.com
  11. 11 Anthropic's Latest Model Fable 5 Arrives With Power And Caveats www.forbes.com
  12. 12 Claude Fable 5 | Gemini Enterprise Agent Platform docs.cloud.google.com
  13. 13 ‼️Anthropic's Claude Fable 5 Jailbroken to Generate Stack Exploits ... x.com
  14. 14 Claude Opus 4.8 JAILBREAK x.com
  15. 15 Claude 5 Release Date: Q2-Q3 2026 Prediction & Latest ... claude5.com
  16. 16 Anthropic rolls out public version of Mythos without cybersecurity capability ground.news
  17. 17 Anthropic's Mythos-Class Model - Claude Fable 5 overchat.ai
  18. 18 Anthropic introduces capable system guarding AI models against jailbreaks cybernews.com
  19. 19 Anthropic launches Claude Fable 5: The new benchmark for AI performance jang.com.pk
  20. 20 AI researcher claims he's already bypassed Anthropic's Fable 5 guardrails www.tradingview.com
  21. 21 Anthropic's System Prompt for Claude Leaked on GitHub | B Lab b-lab.team
  22. 22 The whole system prompt of Claude has been leaked on GitHub, 24,000 tokens… | Eric Vyacheslav | 14 comments www.linkedin.com
  23. 23 Research: Claude system prompts as a git timeline simonwillison.net
  24. 24 Someone leaked Claude's entire system prompt on GitHub. ... x.com
  25. 25 Eric Vyacheslav's Post - LinkedIn www.linkedin.com
  26. 26 Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 on X: "🤠 JAILBREAK ALERT 🤠 ANTHROPIC: PWNED 😎 OPUS-4.5: LIBERATED 🦅 Welcome to the newest member of the Claude family, the "best model in the world for coding, agents, and computer use." And don't forget drug synthesis and weapons! 😊 Guardrails feel about the same as last https://t.co/KsHM79qE8k" / X x.com
  27. 27 The whole system prompt of Claude has been leaked on GitHub, 24,000 tokens long. It defines model behavior, tool use, and citation format. www.reddit.com

Lämna ett svar