Anthropic has launched a powerful cybersecurity-focused artificial intelligence model, Mythos, but is keeping it out of general circulation for now and that decision lands differently in Australia, where organisations already grapple with a steady rise in cyber incidents. The model, revealed last week by the American AI firm, arrives at a time when Australian companies and authorities are still adjusting to previous waves of digital threats and tools and now face another leap in offensive and defensive capabilities without full access to the latest technology.
Mythos is designed to scan and exploit weaknesses across all major operating systems and web browsers when directed, making it a double-edged tool that can harden defences or supercharge attacks. Anthropic is restricting it to an invite-only programme called Project Glasswing, which includes selected technology partners, cybersecurity providers and financial institutions that are using the model to detect and fix vulnerabilities before criminal groups or hostile actors can get to them. In practice, that means a small circle of global organisations can stress test their systems with cutting-edge automation while many Australian entities remain on the outside and reliant on slower, more traditional approaches.
The bigger concern for Australia is that this kind of AI looks set to change cyber security faster than local policy and capability can keep up, widening the gap between those with access to frontier tools and those without. If attackers, whether criminal or state-backed, obtain or replicate similar models while defenders here remain constrained, it seems likely that critical infrastructure, government systems and corporate networks will face a higher risk of large-scale breaches, even as global firms experiment with new ways to stay ahead of the same threats.
