Viral AI jailbreak prompts bypass safety filters to generate banned cartoon episodes

Anthropic’s Claude Fable 5 model was temporarily restricted by government authorities following reports that its automated safety guardrails incorrectly flagged legitimate malware removal as prohibited cybersecurity activity. Users reported the system forcibly terminated sessions during valid PC cleanup operations, triggering a 19-day suspension before reinstatement. Upon return, the model reportedly began providing unsolicited usage guidance to prevent recurrence. Separate incidents involving version 5.6 Sol allegedly caused account bans through accidental jailbreaks linked to viral image prompts. Industry experts note these events underscore persistent challenges in distinguishing malicious exploitation from authorized technical tasks within AI safety classifiers. Anthropic has not publicly commented on specific enforcement criteria or remediation steps taken since reinstatement. The episode intensifies scrutiny over how AI platforms balance harm prevention with functional utility in sensitive domains like cybersecurity.