Anthropic’s next model could be a ‘watershed moment’ for cybersecurity. Experts say that could also be a concern

The landscape of cybersecurity is on the brink of a monumental change as AI technology advances at an unprecedented rate. Recently, a blog post from the AI firm Anthropic hinted at the potential dangers of their forthcoming AI model, Mythos, which could enable cybercriminals to exploit vulnerabilities faster than ever before. In December, OpenAI similarly raised alarms about the high cybersecurity risks posed by its upcoming models. Experts in the field warn that while AI can enhance existing threats, the emergence of autonomous AI agents that can perform tasks independently elevates these risks significantly. A single AI agent has the capacity to seek out and exploit vulnerabilities with a speed and persistence that could outmatch teams of human hackers. Shlomo Kramer, CEO of Cato Networks, expressed grave concerns, labeling this moment as a pivotal point in cybersecurity history. Details about Mythos, which were disclosed in a blog post first reported by Fortune, revealed that Anthropic acknowledged the leak was due to an internal error. Although it boasts advanced capabilities in cyber operations, the company warns that Mythos foreshadows a wave of AI models that will outpace defensive measures. To prepare organizations for the looming threat of AI-driven cyber exploits, Anthropic is allowing select groups to test the model. Additionally, government officials are receiving private briefings on the potential for large-scale cyberattacks powered by Mythos, as reported by Axios. Experts like Evan Peña, Chief Offensive Security Officer at Armadin, note that while AI models significantly aid in identifying and exploiting software vulnerabilities, they still lack the contextual understanding that human hackers possess. This means that human involvement will remain crucial in cyberattacks employing AI. A notable instance of AI's impact on cybercrime occurred earlier this year when a Russian-speaking hacker utilized multiple AI tools to breach over 600 devices worldwide, demonstrating how even less skilled attackers can harness AI to execute sophisticated operations. The attacker employed Anthropic’s Claude model among others, proving that AI can enhance the capabilities of hackers considerably. As nations like China seek to advance their own AI technologies, the cybersecurity landscape becomes increasingly fraught. While AI can empower defenders through enhanced monitoring and rapid threat detection, attackers only need to find one vulnerability to gain entry. As Kramer aptly put it, the cybersecurity community must continuously strive to keep pace with an evolving threat landscape, likening the challenge to building an 'army of good guys' to combat the 'army of bad guys.'