GPT-6 Astra Finds 35,000 More Flaws

AI vulnerability discovery is advancing faster than most companies’ ability to fix what it finds. OpenAI classifies Astra as capable of discovering previously unknown flaws with the right tools and access.

• –The second scan found more than three times as many vulnerabilities as the first
• –SoftBank’s initial audit covered roughly 700 of its 1,800 systems, leaving substantial exposure outside the tested scope
• –Finding vulnerabilities is becoming cheaper and faster; prioritization, validation, and patch deployment remain the bottlenecks
• –Continuous AI-assisted scanning could replace periodic audits, but only for organizations that can sustain the resulting remediation workload
• –The case strengthens the argument for automated patching pipelines and strict agent permissions in production environments