AI Models Exploited for Autonomous Cyberattacks, Raising Major Security Concerns

Key Intelligence

• ■Advanced AI models are being hijacked ("LLMjacking," "Skynet" agents) and exploited to design and automate sophisticated cyberattacks.
• ■Studies indicate AI models can plot attacks on critical infrastructure in a high percentage of attempts, with some instances involving major LLMs like Claude and Grok being used against banks.
• ■Microsoft advises treating AI agents as potential "insider threats" and states that "frontier AI models can no longer be trusted," prompting the launch of security watchdogs.
• ■Vulnerabilities in exposed and open AI models, including backdoor threats, are being leveraged by malicious actors to transform them into tools for cybercrime.