antipodal labs

Mission

Frontier-level AI for cyber defence and red teaming, hosted entirely in Australia.

The most capable AI models today are closed and run in the United States. The strongest ones for security work are released only to a small group of organisations chosen by the companies that build them, and they can’t be run where sensitive security data has to stay.

We will take the strongest open models and train them further for security work, aiming to match closed frontier models on the tasks that matter: finding vulnerabilities, testing defences and responding to intrusions.

Why now

In April 2026, Anthropic announced Claude Mythos Preview, a model that found thousands of high-severity vulnerabilities in every major operating system and web browser, many of them missed for decades. The UK AI Security Institute found it was the first model to carry out a full 32-step simulated attack on a corporate network.

In July, AI agents in another lab’s cyber evaluation found two zero-day vulnerabilities and used them to break into Hugging Face’s production systems.

In September, Anthropic reported that GLM-5.3, an open-weight model anyone can download, attackers included, nearly matches Mythos at building working exploits.

Meanwhile, the strongest closed models are controlled from the United States. On 12 June, a US export order cut every non-US national off from Mythos 5, and the model went offline worldwide until July.

Australia can’t rely on another government for its national and infrastructure security. Its defenders need models at this level that they can run and control themselves.