Australia courts Anthropic with access to restricted AI model
Anthropic's chief lawyer arrived in Australia this week for closed-door meetings with senior government officials. The talks centered on securing local access to Mythos, the company's newest AI model, which has been released to only about 50 mostly US-based technology companies due to cybersecurity concerns.
Mythos can identify vulnerabilities in computer systems with unusual precision. Both Anthropic and rival OpenAI have restricted access to these models, fearing they could be weaponized for large-scale cyber attacks if they reach bad actors.
Jeffrey Bleich, Anthropic's general counsel and former US ambassador to Australia, met with Andrew Charlton, the Assistant Minister for Science, Technology and the Digital Economy, as well as officials from the Department of Home Affairs and the Australian Signals Directorate.
What the government wants
Australia views access to advanced AI models as a strategic advantage. A government briefing obtained through freedom of information laws shows the federal government is pursuing three main objectives: major Anthropic investments in Australia, commitments to cover grid infrastructure costs and support local communities, and collaboration on frontier AI safety and security.
The briefing also reveals the government wants to understand how Australia's copyright laws affect Anthropic's business operations. Tech companies have argued that copyright restrictions limit AI investment in the country, though the government has so far resisted giving AI firms an exemption to train on Australian copyright material without payment.
In February, Industry Minister Tim Ayres signed a memorandum of understanding with Anthropic CEO Dario Amodei outlining a "shared vision" between the company and Australia. Anthropic opened its Sydney office last month, marking its fourth Asia-Pacific location.
The access problem
No Australian companies or government agencies currently have access to Mythos. OpenAI has begun rolling out its equivalent model, GPT-5.5-Cyber, to a limited number of Australian organisations, including Commonwealth Bank and Westpac.
Alastair MacGibbon, former national cybersecurity advisor and chief strategy officer at CyberCX, said the concentration of early access among US technology giants creates a structural disadvantage. Large international financial institutions and major tech firms received access first, while organisations with bespoke software systems were excluded.
"Winners are being picked, and [the tech companies] chose in a US-centric way who the winners were," MacGibbon said. He argued governments should close this gap rather than leaving individual organisations to negotiate access independently.
What happens next
Anthropic has not committed to a timeline for expanding Mythos access. A company spokesperson said the firm plans to "expand access safely in coordination with" government and enterprise customers globally, but provided no specifics.
The Department of Home Affairs coordinated multiple meetings between Anthropic and Australian operators of critical infrastructure since Mythos was released. In late April, Anthropic briefed 170 representatives from the finance, communications, transport, energy, data, food and grocery sectors-organisations designated as Systems of National Significance under Australia's critical infrastructure framework.
The federal government declined to confirm whether Australian agencies and companies have secured access to Mythos as part of these discussions.
For AI for Government professionals, understanding how these models work and their security implications is increasingly critical. AI for Cybersecurity Analysts offers relevant technical context for those evaluating these tools for organisational use.
Your membership also unlocks:
