Anthropic will now let you access Claude Mythos-class AI models. The company announced on Tuesday that Claude Fable 5 is the first publicly available Mythos-class model, but it comes with certain hard limits.
Mythos refers to Anthropic’s most powerful class of AI models. Anthropic has previously hinted that Mythos preview was so powerful that it could potentially hack any software in the world.
According to the AI startup, Fable 5 is its most capable generally available model so far, with strong performance in software engineering, knowledge work, vision, scientific research and long, complex tasks, but that its release comes with added safeguards in high-risk areas.
Mythos has made waves globally with many governments, including India, being worried about the potential disruption it could cause to cybersecurity. Reports also claim that the US may be using Mythos to plan cyberattacks.
What is Claude Fable 5?
Anthropic says that Claude Fable 5 shows “exceptional performance” across software engineering and knowledge work tasks, and that on some benchmarks it scored more than 10 per cent higher than Claude Opus 4.8, which the company announced late last month.
The AI startup stated, “Fable 5’s capabilities exceed those of any model we’ve ever made generally available The longer and more complex the task, the larger Fable 5’s lead over our other models.”
What about cybersecurity?
When Anthropic announced Claude Mythos in April, governments around the world were concerned over the potential risks associated with cybersecurity due to such AI models. To address this issue, Anthropic says that it has put hard limits on Fable 5 in areas like cybersecurity, biology, chemistry, and distillation. In such cases, the model blocks responses and falls back to Claude Opus 4.8.
Anthropic has previously accused Chinese AI companies of distilling Claude to train their own models. Distillation happens when a smaller model is made to learn from the outputs of a larger model. It is likely that the safeguards against distillation in Fable 5 could help avoid this issue.
Dianne Penn, Anthropic’s head of product management for research, told CNBC that the broad release was possible because of the new guardrails. “For us, it’s really around what we call ‘race to the top,’ being able to provide this technology in a valuable fashion, and at the same time providing the right safety guardrails so that it can do asymmetrically more benefits than harm,” she said.
Penn said Claude Fable 5 represents a “significant jump” in capability, which is why Anthropic had to put in additional protections. She said if a user asks a high-risk question, such as how to make ricin, the model will block the response and Claude Opus 4.8 will deliver a safe answer instead. “What we wanted to do was to be very intentional about building new types of classifiers and new types of safety guardrails in place for this launch,” Penn said.
Anthropic said it tuned the safeguards conservatively, so the fallback may trigger more often than some users expect. At the same time, the company said the cases in which Fable has to defer to Opus 4.8 are rare, with early data showing that at least 95 per cent of Fable sessions run entirely on the model’s own responses.
Be the first to comment