Mythos, an AI model created by Anthropic, has rapidly identified over 2,000 software vulnerabilities within just seven weeks, prompting the company to limit its access to select partners like Microsoft and Google. This decision reflects a growing recognition of the risks associated with such powerful technology, as the findings could dramatically alter the cybersecurity landscape.
John Ackerly, CEO of Virtru, expressed that Mythos’s capability to uncover vulnerabilities is akin to finding “thousands of locks that were never locked.” He noted that the model has surfaced around 30% of the annual global output of zero-day vulnerabilities, indicating that AI tools could potentially reveal far more than the historical total of 360,000 vulnerabilities.
Mythos’s speed and autonomy are distinguishing features that enable it to generate exploits more efficiently than traditional methods. Ackerly cautioned that this capability may lower the barriers for cybercriminals, allowing those with minimal technical skills to inflict significant damage. Furthermore, he highlighted that the reliance on conventional perimeter security measures is becoming increasingly outdated as AI technologies evolve, urging a shift toward data protection strategies to counter these emerging threats.