NZ at wild frontier of AI superhacking - RNZ

New Zealand's cybersecurity watchdog is learning from US companies that are testing advanced AI models, including Anthropic's Claude Mythos. The move comes after Palo Alto Networks conducted tests that demonstrated the model's ability to detect and exploit software vulnerabilities at high speed. During testing, the model triggered dozens of security alerts in a single day.

By comparison, the organization typically issues around five security alerts per month under normal conditions. The findings highlight a significant increase in the potential scale of automated cyber threats. The watchdog aims to adapt its strategies based on these observations. The US companies testing these models are sharing their findings to help regulators understand emerging risks.

The tests were part of broader efforts by US companies to evaluate the capabilities and risks of frontier AI models. New Zealand's regulator is seeking to stay informed about these developments to better prepare for emerging threats. The RNZ article notes that the situation places New Zealand at what it calls the 'wild frontier' of AI superhacking, where the speed of AI development outstrips existing regulatory frameworks.