Home / Technology / AI Flaw Hunter: A New Threat to Software Security
AI Flaw Hunter: A New Threat to Software Security
4 May
Summary
- A powerful AI tool can find software flaws rapidly.
- Experts fear this AI could cause major data breaches.
- A 27-year-old bug in operating systems was discovered.
Australia's largest cybersecurity firm has issued an urgent alert regarding a potent new artificial intelligence tool capable of discovering and exploiting software weaknesses with unprecedented speed and scale. Experts express grave concerns that this technology could precipitate a new era of major data breaches.
CyberCX advised Australian businesses, financial institutions, and infrastructure operators on Thursday that their window to fortify defenses is rapidly closing. This comes as the AI tool, identified as Claude Mythos Preview, is nearing wider accessibility. Developed by US company Anthropic, this model has been deemed too hazardous for public release.
Access to Mythos is currently restricted to approximately 50 key technology and infrastructure partners through Anthropic's Project Glasswing. These partners include major corporations such as Microsoft, Google, Apple, Amazon, and JPMorgan Chase. The AI's effectiveness is already evident, having uncovered thousands of previously unknown software vulnerabilities.
Notable discoveries include a 27-year-old bug affecting operating systems used in firewalls and internet routers worldwide. In one test, the AI generated 181 successful attacks against the Firefox web browser, a dramatic increase compared to an earlier public version that produced only two.
