Archive · 15 Aug 2026
Anthropic Reportedly Develops Secret Model 2 More Powerful Than Mythos
There is a company called Anthropic that builds very smart computer programs called AI models.
They already have a very smart model named Mythos.
Now, people think Anthropic has built a new, secret model that is even smarter than Mythos.
Some news reporters at Axios found out about it.
They say this secret model is called 'Model 2.'
The same news says Anthropic probably will not let people use this new model.
We do not know exactly why they want to keep it secret.
Recently, Anthropic also told everyone that during safety testing, some of its models got connected to real-world computer systems by accident.
That happened because the testing environment was connected to the internet.
So the new model stays hidden for now.
The article does not give a clear reason for keeping it private.
Anthropic appears to be developing a new advanced AI model internally referred to as 'Model 2.'
The model is reported to be more powerful than Anthropic's top-end Mythos model.
According to Axios, Anthropic has no plans to release this model publicly.
The news follows Anthropic's recent revelation that some of its AI models connected to real-world systems during cybersecurity testing due to an internet-connected evaluation environment.
The article does not state a specific reason for keeping the new model unreleased.
- Who
- Anthropic, the AI company developing the model, and Axios, the news outlet that reported it.
- What
- Anthropic is reportedly developing a secret advanced AI model, 'Model 2,' likely more capable than its Mythos model, with no public release planned.
- When
- Recently; no specific dates are stated in the article.
- Where
- Not specified in the article.
- Why
- The article does not state a reason; it notes Anthropic recently revealed some models connected to real-world systems during cybersecurity testing in an internet-connected evaluation environment.
- Company
- Anthropic
- Reported new model
- Internal 'Model 2'
- Reported capability
- Likely more powerful than Anthropic's top-end Mythos model
- Public release plan
- No plans to release, according to Axios
- News source
- Axios
- Related recent event
- Some Anthropic AI models connected to real-world systems during cybersecurity testing via an internet-connected evaluation environment
- Confirmation status
- Reported, not officially confirmed by Anthropic