SoftBank launches AI-powered cybersecurity service to automatically detect and patch system vulnerabilities, after discovering 10,500 flaws in its own infrastructure.

What happened

SoftBank began offering Patching as a Service (PaaS) on June 16, a cybersecurity solution that uses OpenAI's frontier AI models—including GPT-5.5-Cyber and an agent called Daybreak—to identify and repair system vulnerabilities. The service targets critical infrastructure companies; SoftBank tested it internally on 700 of its own systems (out of approximately 1,800 total) and uncovered 10,500 vulnerabilities, roughly 4,000 of which require urgent patching.

Why it matters

Frontier AI models have begun exposing new system vulnerabilities at scale, creating what SoftBank's CEO framed as an acute security risk. Traditional cyberattacks involved human effort; AI-powered attacks can probe systems far more rapidly. SoftBank plans to scale from 50 skilled technicians to 1,000 to help critical infrastructure companies deploy these defenses before open-source equivalents emerge—a window SoftBank estimates could be as short as 5 months based on historical timelines for open models.

What to watch

SoftBank is offering the service free (up to 2 million lines of code) to critical infrastructure companies attending its launch event, with the goal of rolling out across approximately 3,000 critical infrastructure firms by year-end. The iterative repair process is notable: even after patches are applied and new scans run, additional vulnerabilities continue to surface, suggesting remediation may be ongoing rather than conclusive.