Anthropic has begun rolling out a new model called “Fable,” which is based on the same underlying model as Mythos, its most powerful AI model class.
Anthropic previously said that it developed a model called “Mythos,” which is a state-of-the-art model that poses security risks to companies around the world.
At that time, Anthropic noted that Mythos was powerful enough to potentially help bad actors attack public and private software.
“The advantage will belong to the side that can get the most out of these tools,” Anthropic warned in April when it announced the Mythos model.
“In the short term, this could be attackers, if frontier labs aren’t careful about how they release these models. In the long term, we expect it will be defenders who will more efficiently direct resources and use these models to fix bugs before new code ever ships.”
In other words, it could have been abused to find and exploit vulnerabilities in apps like Firefox.
Because of those risks, Anthropic decided to limit access to models like Mythos and offer them only to cybersecurity experts and trusted companies.
Now, Anthropic says it has developed strong guardrails for the same model class, which means these powerful AI models can no longer be easily exploited by bad actors.
As a result, it’s launching a safer version called “Fable 5.”
Source: BleepingComputer
According to Anthropic, this model has strict safeguards in place that will block or divert sensitive queries, like those involving offensive cybersecurity, biology, or chemistry, to its previous model, Opus 4.8.
Claude Mythos 5 is the unrestricted version of that same model, with those safeguards lifted.
Because of the risks involved, it is only available to a highly vetted group of trusted partners, such as government cyberdefenders and specific life sciences researchers.
Fable 5 is free for a limited time, and it consumes tokens faster than any other model.
Anthropic says Fable 5 is an expensive model because it requires a lot of compute, which means the company cannot afford to make it available as easily as Opus 4.8 or its previous models.
However, until June 22, Anthropic says Fable 5 will be offered to all Pro, Max, and Enterprise customers, but after the window expires, it’ll switch to usage-based pricing.
In our tests, BleepingComputer observed that Fable 5 uses a massive amount of tokens in a span of minutes.
Source: BleepingComputer
I particularly noticed this behaviour when I used Workflow, a new execution system that allows Claude to break complex prompts into smaller tasks and spin up parallel subagents to implement them.
Claude Fable 5 exhausted my $100 Max subscription’s daily usage, which was at zero when I started using it, in just 9 minutes
Source: BleepingComputer
This doesn’t happen when you casually interact with Claude Fable 5, but if you switch to Workflow mode and change model thinking to high, you’re going to consume all your tokens in minutes.
However, even if you don’t use Fable 5 with workflow in xhigh effort, you’re still going to consume it 2 times faster than Opus model.
This explains why Anthropic is hesitating to unlock Fable 5 in the same capacity as Opus and other models, but that could change in the coming weeks, as the company is known for nerfing its models and increasing the capacity later.
Security teams log 54% of successful attacks and alert on just 14%. The rest move through your environment unseen.
The Picus whitepaper shows how breach and attack simulation tests your SIEM and EDR rules so threats stop slipping by detection.
