Claude Mythos 2 Rumor: A Day on X

From the night of June 3 to the early hours of June 4, the X timeline buzzed with rumors of a newer checkpoint for Claude Mythos. The speculation began when the UK AI Security Institute (AISI) published a brief evaluation note revealing that a more recent iteration of the model it had tested in April yielded significantly higher scores.

While some users immediately speculated that Anthropic was testing 'Mythos 2,' others countered that the evaluation merely involved a newer training run of the existing April model. The following is a summary of the Mythos 2 rumor debate that unfolded on X.

The Spark: AISI Evaluates a 'Newer' Checkpoint

As part of its ongoing cyber evaluation work, AISI has been testing the autonomous cyber capabilities of frontier models, Claude Mythos among them, in simulated penetration environments. The core of its latest update concerned the Mythos-class model running in Project Glasswing, and rested on a single finding.

Since our first Mythos evaluation, we were given access to a newer Mythos Preview checkpoint. On a 32-step corporate-network attack we estimate takes a human expert about 20 hours, this checkpoint completed the full chain in 6 of 10 attempts.

This success rate doubles that of the previous checkpoint, which solved the same task ('The Last Ones') in only three out of ten attempts. Additionally, the model cleared the 'Cooling Tower' cyber range—a challenge previously unbeaten by any model—in three out of ten trials. According to AISI's published evaluation, it is the first model to successfully navigate both tasks. By comparison, GPT-5.5 solved 'The Last Ones' in only three out of ten attempts during the same testing round.

'Mythos 2 Under Test' Speculation Sweeps X

The six-out-of-ten success rate quickly captured the attention of the X community, prompting a wave of speculative posts.

BREAKING. AISI tested a newer Mythos checkpoint. These numbers are insane — it solved a 20-hour task in 6 out of 10 attempts.

An insider account that closely tracks Anthropic also weighed in. @synthwavedd went beyond speculation, tagging a 'SCOOP' with a far more concrete claim.

SCOOP: Anthropic is gearing up for the public launch of a new version of Mythos, better than Mythos Preview.

That the model had been locked inside the Project Glasswing preview only sharpened expectations. Speculation intensified rapidly. One prominent influencer account asserted without evidence that Mythos 2 was already undergoing internal testing. While unsubstantiated, the claim gained traction as it coincided with earlier reports of an imminent model release and spread quickly.

Anthropic is already running Mythos 2 — while everyone waits on the first release. Mythos 3 is next.

The rumor was further fueled when Logan Graham, lead of Project Glasswing, described Claude Mythos Preview as a 'step change.' For many users on X, the prospect of a new Mythos checkpoint undergoing quiet testing ahead of an official launch seemed highly plausible.

Pumping the Brakes: 'Just the April Model'

However, a series of brief corrections soon tempered the excitement.

UPDATE: these results are from the exact same Mythos Preview disclosed on April 7. There is no 'new' Mythos checkpoint, at least not publicly.

Several observers pointed out that the original Mythos model had been released only five weeks prior. They argued that the 'new checkpoint' was likely a later training snapshot of the same Mythos Preview deployed in Project Glasswing, rather than a successor. Indeed, AISI noted in its report that significant capability gains do not necessarily require the release of a new model, as subsequent iterations of an existing model can substantially alter performance estimates.

Ultimately, Anthropic has not officially announced a 'Mythos 2,' and Claude Mythos remains the only model in play. The online buzz was likely amplified by the June 2 expansion of Project Glasswing to approximately 150 organizations across more than 15 countries, which primed the community for news of a new model. The Mythos 2 rumor that swept X over the past day reflects surprise at how quickly an existing Mythos Preview checkpoint can improve — even without a new checkpoint release — rather than the arrival of a next-generation model.