Australia gains access to powerful AI amid questions about readiness

Australia has been granted access to Anthropic’s Claude Mythos Preview, a highly advanced AI model deemed too risky for public release, as part of the company’s expanded Project Glasswing initiative. The program now includes 150 organizations from 15 countries, including NATO allies and EU members, with a focus on critical sectors like energy, healthcare, and communications. Mythos is capable of identifying software and network vulnerabilities that could pose catastrophic risks if exploited, leading Anthropic to previously restrict access to the UK, US governments, and select American companies. The Australian government and private companies, including those classified as Systems of National Significance (like banks, transport, and health services), are among the new participants. The Australian Signals Directorate welcomed the expansion, though the government has not confirmed specifics about its access. Earlier this year, Anthropic CEO Dario Amodei briefed Australian officials, emphasizing AI’s potential to strengthen democracies while warning against allowing autocracies to gain a military advantage. Independent MP Kate Chaney has criticized the government for moving too slowly in addressing AI risks, urging faster action to protect the economy, workplaces, and national security. Meanwhile, the US government has also taken steps to assess AI threats, with President Donald Trump issuing an executive order requiring American tech companies to submit AI models for cybersecurity testing up to 30 days before public release. Anthropic’s statement highlighted that many new partners are vendors maintaining critical codebases relied upon globally, including by governments. A successful attack on their systems could have devastating consequences. The company previously briefed Australian officials from the Reserve Bank, Treasury, and key infrastructure sectors to discuss Mythos’ capabilities and risks. While the Australian government has not provided details on its Mythos access, the inclusion in Project Glasswing signals a growing recognition of AI’s dual potential as both a tool for defense and a source of vulnerability. The move follows broader global efforts to balance AI innovation with cybersecurity safeguards.