Andrew J. Lohn
Chinese hackers successfully used Anthropic’s artificial intelligence (AI) models to run a hacking campaign almost entirely without human interaction, breaching “major technology companies and [foreign] government agencies,” the company reported in November. Nothing about any of the individual hacks was especially noteworthy or sophisticated. Still, the campaign marks a step change in cybersecurity.
AI-Assisted Hacking
The Chinese attackers did not just use Anthropic’s AI model to find a vulnerability in the digital armor. They built a hacking system that combined many distinct AI sessions that each worked on different aspects of the intrusion. Those separate sessions were coordinated by still other AI systems that were only occasionally directed by humans. Not only was the AI hacking system complex, but the attacks themselves were conducted in six phases, with many steps in each phase. In total, the AI system likely took dozens or more separate steps against each successful target.
No comments:
Post a Comment