New IBM Security Services Aim to Counter Risks of Frontier AI Models

IBM announced Autonomous Security agents and a new assessment service to help organizations identify and mitigate security gaps introduced or accelerated by advanced AI models; the report warns that frontier AI models can both improve defenses and lower barriers for attackers. It cites Anthropic’s Claude Mythos as notably capable of discovering long-undetected vulnerabilities and autonomously creating exploits (hence its restricted use), and notes OpenAI’s limited-release GPT-5.4-Cyber for defensive purposes, emphasizing the dual-use risk of these technologies.