Considering everything that happened around the semi-release of Anthropic’s Mythos model, which ended up being sandboxed due to security concerns, you’d think this week’s unveiling of a new model called Claude Fable 5 would be bigger news.
Tech media is talking about the model, but some would note a disconnect between responses to Fable 5, and the Mythos model which got a lot of attention from the government.
Federal Reactions
So let’s start with that: when Mythos came out, prior to Anthropic restricting the model to about a dozen or so tech clients, there were high-level meetings at the federal level of U.S. government. Fed Chair Jerome Powell and Treasury Secretary Scott Bessent famously met with top U.S. bank CEOs including the heads of Bank of America, Citigroup, Goldman Sachs, Morgan Stanley, and Wells Fargo. Legislators huddled to discuss what to do about a model that had the ability to create a brand new army of hackers. Ultimately, the constraints emerged.
Now, as Fable 5 comes online, I could not find any news that government leaders have met about this model. I asked GPT, and here’s what I got (in part, since I’m not trying to outsource the entirety of this response):
Yes: Policymakers and government officials reportedly received briefings or held discussions related to Mythos and advanced AI capabilities.
No confirmed reports: of a distinct legislative meeting focused specifically on Fable 5.
Keep that in mind.
They Swear It’s Safer
Anthropic, meanwhile, is detailing how it is dealing with the security risks, and the answer seems to be a “handoff” policy where Fable 5 will shuffle some requests to a lesser model.
“Releasing a model this capable comes with risks,” write spokespersons on the company’s blog. “Without safeguards, Fable 5’s capabilities in areas like cybersecurity could be misused to cause serious damage. We’ve therefore launched the model with safeguards that mean queries on some topics will instead receive a response from our next-most-capable model, Claude Opus 4.8. To release the model both safely and quickly, we’ve tuned these safeguards conservatively—they’ll sometimes catch harmless requests, though they trigger, on average, in less than 5% of sessions. With more capable models arriving in the coming months, we’re working to improve our safeguards and reduce false positives as quickly as we can.”
The announcement also reveals that another model called Mythos 5 will be utilized inside of Project Glasswing, that original walled garden where Cisco and some other firms are wielding the full power of the LLM, exclusively, for safety reasons.
Mandatory Data Collection
Another major aspect of this new model is that it requires data collection for all customers, which undercuts former contract policies for some enterprise and API parties.
Check out this reporting from Mashable, where Matt Binder goes over how the universal data retention reverses what other Anthropic models did for some client firms with proprietary trade information. Binder quotes a lawyer named Jessica Mathews, who says:
“Every other Claude model available through the API, including Opus 4.8, Sonnet 4.6, and Haiku 4.5, can operate under Zero Data Retention agreements. Fable 5 cannot. If your organization previously had a ZDR agreement with Anthropic, that agreement does not apply to Fable 5 traffic. This is a policy change that overrides existing enterprise commitments for this specific model class.”
That, Binder and others suggest, might be a problem. At the very least, it’s something that enterprise customers should be aware of, and it’s something that people could potentially ask Anthropic’s leadership about, as a point of curiosity.
Myth and Fable
I’m really interested in how companies name their models, so this fourth item has to do with “Claude Fable 5.” There was no “Fable 4,” so what led the top brass to use this designation?
I went back to GPT, because it is now the authority on getting cogent information from all over the web, and also because there aren’t a lot of human reporters writing about model names.
GPT responded:
“According to reporting on the launch, the name ‘Fable’ was chosen because it is thematically related to ‘Mythos’. A mythos is a larger body of myths or stories, while a fable is a specific story, often one intended for a broad audience and carrying a moral lesson. Anthropic has said the public model is essentially a safeguarded version of Mythos, and the different names help distinguish the restricted and public releases.”
GPT also referred me to this Business Insider article, which is apparently the reporting in question. The article is behind a paywall, so that’s where the trail ends. I wouldn’t necessarily think of fable as a diminutive form of mythos, but I guess it makes a certain semantic sense.
Anyway, Fable 5, in its public incarnation, is only about a day and a half old. Check it out.
