Startup Anthropic is releasing a public version of its most powerful artificial intelligence model, now called Claude Fable 5, which includes defensive mechanisms that limit its use in risky areas, such as cybersecurity. This follows the previous version of the model, Mythos, which caused a significant stir in the world earlier this year due to its ability to detect software flaws.
So far, Anthropic has restricted access to Mythos to a group of about 200 organizations, including the US government as part of the Glasswing program. In April, the company announced that Mythos had discovered thousands of vulnerabilities in software. A broader provision of Mythos capabilities may allow Anthropic, a company valued at $965 billion, to strengthen its position in the market and outpace its competitor OpenAI in the race to go public.
Defensive Mechanisms and Limitations
Anthropic stated that it conducted thorough testing to ensure that users cannot manipulate the new model to bypass its instructions or perform prohibited actions. Dianne Penn, head of product management, research, and labs at Anthropic, told Reuters: “If a college student asks the model to find cyber vulnerabilities in package X or code, the model will refuse.” In such cases, Fable 5 will revert to Opus 4.8 to respond.
Anthropic highlights the performance of Claude Fable 5 in software development and analytics. According to Penn, the Fable 5 model will be more expensive but will perform tasks with less token usage, thereby reducing the overall cost per task, according to early customer feedback.
Expansion Plans and Pricing
Users who had access to the previous version of Claude Mythos without defensive mechanisms will be able to upgrade to the new Claude Mythos 5. Anthropic plans to expand access over time through a more “systematic trusted access program”. Prices for both models are $10 per million input tokens and $50 per million output tokens.
Source: NDTV