SoftBank: GPT-6 Astra rescan found 35,000 new vulnerabilities

Reports that a stronger model rescanning the same 700 systems surfaced 3x more vulnerabilities than the earlier pass. Patch backlogs and open-source scope gaps may grow faster than teams assume. Figures are CEO-stated and unaudited.

• The October 7 attack on cloud subsidiary IDC Frontier halted virtual servers for 495 companies and local governments. SoftBank says no customer data leak has been confirmed, and the entry point was open-source software it had left out of its vulnerability scans.
• Because damage began seven minutes after detection, SoftBank dropped its practice of having specialists meet to decide countermeasures and now blocks communications as soon as an anomaly appears, accepting temporary service suspensions.
• CEO Junichi Miyakawa says pre-attack reconnaissance rose from about 300 million incidents a month in fiscal 2025 to 600-700 million, and that attacks surged after open-weight models specialized for cyberattacks appeared.
• SoftBank's Patching as a Service, which runs diagnostics, patching and post-patch checks, had 7,670 corporate contracts as of October 30. Patching is automated, but confirming systems still work afterward is still done by people.

• open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.

• postMicrosoft's Autopilot agent ships built on OpenClaw framework
• postOpenClaw's Next Version Auto-Decides to Steer or Queue Agent Turns
• articleBrood War Bench: frontier agents play StarCraft at beginner level
• postPeter Steinberger Speeds Up Git Worktrees ~80% Using Filesystem Clones

• articleSafety overview: GPT-6 Astraopenai.com
• postAstra's early users report it patching upstream deps, not just local codesteipete
• postGPT-5.5-Cyber lands in full, with remediation tooling attachedSam Altman

Checking sign-in…

Loading comments…