A leaked draft from Anthropic reveals alarming potential for its forthcoming AI model, Mythos, to enhance cyberattack capabilities, raising concerns among security teams. This information surfaced due to a human error involving the company’s content management system, highlighting the urgency of the situation as AI's role in cybercrime is expected to grow significantly.
Experts are particularly worried about the speed at which Mythos could exploit system vulnerabilities, outpacing current defenses. Shlomo Kramer commented on the inevitability of advanced AI agents, which can perform tasks autonomously, significantly increasing the threat landscape. He emphasized that this marks a pivotal change in cybersecurity.
In response to these threats, Anthropic is reportedly allowing select organizations to test Mythos prior to its official launch. The company has also alerted government officials about the risk of large-scale cyberattacks associated with this new model. Evan Penya, Chief Security Officer at Armadin, noted that while these AI models offer unprecedented advantages to attackers, they still face challenges in context and information identification.