After being pulled offline for more than two weeks, Anthropic’s Fable 5 and Mythos 5 AI models are returning within hours. That’s no speculation, straight confirmation from Anthropic themselves in the form of a tweet.
According to the tweet, Anthropic confirmed that the US Department of Commerce has “lifted export controls on Claude Fable 5 and Mythos 5.”
Naturally, for the countless developers, researchers and power users from around the world who first saw Fable 5 and Mythos 5 appear and then vanish almost overnight, Anthropic’s announcement came as a genuine event. People celebrated in tweet replies, relief broke through weeks of tension and uncertainty, and there were a few unanswered questions as well.
If you look underneath Anthropic’s tweet, quite a few commenters want to know details about what exactly changed – not in the US Department of Commerce’s scrutiny, but in the models themselves – before their relaunch for the world to use.
When it originally released Fable 5 and Mythos 5, Anthropic had specifically mentioned how they were a step down from the controversial Mythos Preview version of Claude – Anthropic’s most capable, raw, untamed model. Fable 5 and Mythos 5 did have safety guardrails in place that satisfied Anthropic’s internal safety policy for them to feel comfortable to launch them in the first place, before the US Department of Commerce restricted their export.
So with this relaunch, the question on everyone’s mind – including mine – is if the already nerfed Fable 5 and Mythos 5 models are being nerfed some more to pass muster and be “even safer” to be allowed back on the market for everyone to use?
Anthropic had originally said Fable 5’s capabilities exceeded those of any model it had ever made generally available, calling it state-of-the-art on nearly all tested benchmarks across software engineering, knowledge work, vision and scientific research. But with this re-release of Fable 5, it would be interesting to see if performance has degraded on any key parameters or benchmarks, if at all.
One comment on Anthropic’s tweet even wondered if the use of these models come with some sort of enhanced surveillance baked in to keep a more focused eye on questionable prompt queries?
Before we get too ahead of ourselves with all the conspiracy theories, it would be good to recount objective facts as stated by Anthropic. Fable and Mythos share the same underlying model, the only difference being the level of security guardrail. Fable 5 includes features that can choose to decline response or requests for certain user-prompted tasks – Mythos 5 does not include these safety nets.
That’s what’s on record, apart from a previously stated 30-day data retention policy, for Anthropic’s governance design framework, not a monitoring or surveillance regime. But, unfortunately, that’s what’s breeding all the conspiracies for now.
Of course, Anthropic’s silence on this matter is letting people get ahead of themselves. Hopefully more information will be released soon, as soon as the models come back online for general public use.
Also read: Claude Mythos finds 10000 bugs: Is Indian industry ready?