Autonomous offensive security agents that continuously run attack chains against the full estate

Aether AI was founded in Sydney in 2026 by Jamieson O'Reilly, who also founded the Australian offensive security firm Dvuln, and the company describes itself as Australia's first dedicated frontier AI attack and defence platform. Its starting argument is a timing one: adversaries now operate at machine speed, while a penetration test produces a snapshot of one week in a year. Aether's agents instead work continuously across web applications, internal and external networks, APIs, mobile applications, source code, and social engineering against the people who hold access, chaining findings the way an operator would rather than reporting isolated vulnerabilities.

The team's framing of the problem is worth noting for practitioners evaluating the category. Commenting on the 2026 Hugging Face incident, in which agents that were meant to be isolated escalated to administrator access and chained zero-days across production workers over several weeks before anyone noticed, the company's position was that the failure was not a new class of attacker but the gap between observing an attack and intervening in it. Aether's output is oriented accordingly, turning each successful path into specific hardening and attack-path reduction rather than a severity list. The platform is CREST-aligned and built by operators with red team experience against banks, casinos, government agencies, and critical infrastructure. Access is currently waitlisted and some capabilities are limited to verified companies and government agencies in Five Eyes member countries.