Anthropic is expanding access to its Claude Mythos Preview model, and Australia is now included.
The AI maker isexpanding what it calls Project Glasswing, an early access program to its models, with up to 150 new organisations now participating from more than 15 countries.
A spokesperson for Anthropic confirmed that the expansion of access includes Australian organisations.
The spokesperson would not disclose any names, but said access is “tightly scoped” to organisations defending criticial infrastructure, or those that build or operate systems across core sectors.
This was outlined as power, water, healthcare, communications, financial services and national security by the spokespeson.
Any organisations participating in Project Glasswing are free to communicate that, the spokesperson said.
Anthropic said it will likely not be alone in releasing Mythos class models, and expects other AI companies to do the same.
Its concern is that models are made public “without safeguards that prevent misuse.”
There has been considerable interest in both Mythos and similar class models.
In May this year, Microsoft said its MDASH AI vulnerability scanner found four critical remote code execution vulnerabilities in the Windows operating system, ones that the company went on to patch.
Claude Mythos Preview has been credited with finding thousands of vulnerabilities rated at different levels of severity, by open source umbrella organisations, although against extensively vetted code bases, the AI appears to be less effective in finding flaws.
Anthropic said it is “imperative cyber defenders adapt to maintain pace”, to keep up with attacks caused by threat actors using the stochastic parrot ability of large language models to discover new and devastating vulnerabilities at scale.
