1300 166 138 Tell us what you need
News / Finance & Banking

AI cyber tools shorten the window for Australian banks

By Monika Tantau • Published 7 months ago
AI cyber tools shorten the response window for Australian banks against zero-day exploits

Key takeaways

• Anthropic built and withheld Mythos, an AI that scanned systems overnight and produced working zero-day exploits
• Mythos surfaced flaws across operating systems, browsers, and infrastructure trusted for over 20 years
• Quantum 'harvest now, decrypt later' attacks compound the risk for sensitive data stored by Australian banks
• ASX-listed Codeifai launched QuantumAI Secure with Kyber-based post-quantum encryption for payments

Most Read

Anthropic engineers reportedly built an AI model called Mythos and chose not to release it. In overnight tests, Mythos scanned systems, located unknown vulnerabilities, and produced working proof-of-concept exploits. The reported flaws spanned major operating systems, web browsers, and long-trusted infrastructure, including weaknesses that had quietly existed for more than 20 years.

The Mythos disclosure, covered by Stockhead, lands at the same moment the wider security industry is grappling with a second, longer-term threat: quantum computing’s eventual ability to break today’s encryption. For Australian banks, ASX-listed enterprises, and any organisation running legacy systems, the window between vulnerability discovery and exploitation is shrinking fast.

What Mythos actually demonstrated

Mythos was reportedly built by Anthropic engineers and never released to customers. According to the reporting, the model could scan a target system overnight and emerge the next morning with chained exploits, meaning it not only found vulnerabilities but combined them into working attack paths.

The flaws it surfaced were not exotic edge cases. They reportedly included vulnerabilities in widely deployed operating systems and browsers, plus infrastructure components that organisations have trusted for two decades. The signal here is the speed and the scale. Where a skilled human researcher might find one chained exploit in weeks of work, an AI system can produce many in a single night. That changes the assumed timeline that vulnerability management teams operate on.

Why quantum compounds the threat

The Mythos story is paired in security reporting with a quieter but more consequential trend: ‘harvest now, decrypt later’ attacks. State actors and well-resourced criminal groups are believed to already be intercepting and storing encrypted data, including payments, communications, and identity documents, in the expectation that future quantum computers will be able to break the encryption that currently protects it.

Today’s attacks do not need to succeed today. The encrypted financial records and customer files leaving Australian networks right now may be readable in ten years. As SecurityWeek noted, the time-to-exploit for new vulnerabilities has dropped to hours in 2026. Combined with retrospective decryption risk, the practical security posture of every regulated business needs to be reconsidered.

Quantum 'harvest now, decrypt later' attacks compound AI zero-day risk for Australian financial data

Why Australian banks should pay attention

NAB, Westpac, CBA, and Australia’s broader financial sector run on a mix of modern cloud infrastructure and decades-old core banking systems. That legacy layer is exactly where AI-driven scanners like Mythos would find the most material to chain together. Regulators including APRA have already pushed prudential standards on operational resilience and information security, but the timelines those standards assume, often months to detect and patch, are no longer realistic against machine-speed adversaries.

ASX-listed Codeifai has moved on the quantum side. It acquired AntennaTransfer.io and rebranded the platform as QuantumAI Secure, embedding Kyber-based post-quantum cryptography into a payments gateway. That positions Codeifai as one of the first publicly traded Australian companies offering production-grade post-quantum protection to financial counterparties, a meaningful early move and a signal to peers that crypto agility is moving from theoretical concern to procurement criterion.

Crypto agility becomes a board-level question

For Australian operators outside the big four, the takeaway is not to buy a single product. It is to make ‘crypto agility’, the ability to swap cryptographic primitives without re-engineering the application, a documented architectural goal. That includes inventorying where encryption is used, isolating it behind abstraction layers, and tracking which vendors support post-quantum algorithms.

The same architectural discipline supports faster patching of AI-discovered vulnerabilities. Organisations that have invested in document automation and integration platforms are typically further along here, because they have already mapped where sensitive data flows. For finance teams running fragmented stacks, this is the moment to consolidate finance and accounting automation behind clear data boundaries.

A shorter window for everyone

Mythos itself may never ship publicly, but the capability it demonstrates is now the baseline assumption. Other labs are building similar tools, some for defenders and some, eventually, for attackers. The combination of AI-driven offensive research and quantum-decryption risk means Australian boards can no longer treat cybersecurity as a periodic compliance check.

The companies that adapt earliest will be the ones who built crypto agility, real-time monitoring, and rapid-patch discipline before the window closed. The rest will spend the second half of the decade catching up.

Scroll to Top