Anthropic Officially Releases Claude Fable 5, Its Most Alarming AI Finding from the US

Jakarta - Anthropic has just released its latest AI model, Claude Fable 5. The model is a public version of Mythos, which caused a stir and posed cybersecurity risks. The company praised Claude Fable 5’s capabilities, stating it delivers performance in software engineering and analytics. Additionally, Anthropic assured that its newest model has been tested to ensure users cannot manipulate it to bypass guidelines and carry out prohibited actions. “For example, if I am a student asking the model to help find cyber vulnerabilities in package or code X, the model will refuse, and Fable 5 will revert to using Opus 4.8 for a response,” said Dianne Penn, head of product management, research, and labs at Anthropic. Fable 5 is set to be a more expensive model. Based on initial customer feedback, the model is claimed to use fewer tokens when completing tasks, thus ensuring an overall reduction in cost per task. Both models will be priced at US$10 per million input tokens and US$50 per million output tokens. During its preview, Mythos revealed the ability to find software security flaws. Wall Street banks tested Mythos and found it could identify low-risk vulnerabilities and escalate them to high-risk ones. Access was eventually granted to several institutions, including some banks. Sources say the major banks with access helped those that did not receive it. A source from one of the large banks explained that Mythos discovered vulnerabilities in both proprietary and open-source code, forcing banks to upgrade legacy technology where software support had reached its end of life.