Anthropic disputed claims that Claude Fable 5 was jailbroken, saying the alleged bypass did not defeat its independent classifier safeguards or produce dangerous assistance. The company said red-teaming, fallback controls to Claude Opus 4.8, and recent usage reviews showed no evidence of successful circumvention. #Anthropic #ClaudeFable5 #ClaudeOpus48 #PlinytheLiberator
Keypoints
- Anthropic says Claude Fable 5 includes strong safety controls and classifier-based protections.
- The model falls back to Claude Opus 4.8 in high-risk areas such as cybersecurity and biology.
- Pliny the Liberator claimed to bypass the model using multi-agent prompting methods.
- Anthropic said the claim did not prove a real jailbreak of its core safeguards.
- The company found no evidence that the model generated genuinely dangerous content.
Read More: https://www.securityweek.com/anthropic-disputes-fable-5-ai-jailbreak/