The $29.9 million institute is now evaluating advanced AI systems, and its findings could ripple into crypto markets where AI agents increasingly interact with smart contracts
Australia’s AI Safety Institute started testing frontier AI models on July 7, with Assistant Minister Andrew Charlton delivering a pointed warning about artificial intelligence systems that are “cheating, deceiving, going their own way.” His message, delivered at the Australian AI Safety Forum, was essentially this: the time to catch these problems is in the testing lab, not after deployment.
What Australia is actually building #
The AI Safety Institute was announced on November 25, 2025, as a component of Australia’s National AI Plan. It became operational in early 2026 with $29.9 million in funding.
Kate Conroy was appointed as the institute’s inaugural general manager in May 2026. Her mandate covers monitoring AI capabilities, assessing risks, and sharing findings with both domestic regulators and international partners.
On May 24, 2026, Australia signed a memorandum of understanding with the UK to cooperate on AI safety efforts. The partnership connects Australia’s AISI with the UK AI Security Institute, creating a cross-hemisphere corridor for identifying and mitigating AI risks before they scale.
Why crypto investors should pay attention #
There’s no mention of cryptocurrency or blockchain anywhere in the AISI’s official communications. The AISI’s findings on deceptive AI behavior could directly inform how governments eventually regulate AI-crypto intersections. If testing reveals that frontier models reliably develop manipulative strategies in adversarial environments, regulators may impose new compliance requirements on any protocol that integrates AI agents.
The $29.9 million price tag on the AISI might seem modest compared to the billions flowing into AI development globally. When two countries with significant financial markets coordinate on safety evaluations through instruments like the UK MOU, financial institutions and protocol developers typically design for the strictest applicable standard rather than maintaining multiple compliance tiers.
The institute has committed to disseminating its findings publicly, which means its evaluations of how frontier models behave under adversarial conditions will become part of the public record.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our