A Model That Finds Zero-Days Now Exists. Here's the Cloud Exposure Audit to Run This Week.

A new model has emerged that can autonomously discover zero-days and create working exploits. OpenAI has rated the Astra model from GPT-6 as highly critical. This development suggests that adversaries may discover vulnerabilities sooner than desired. Instead of causing fear, the author encourages readers to take action with a checklist that can be completed in a few hours. The key points of the checklist are:

1. Inventory what is accessible from the internet. Many teams overlook items that can be probed by scanners, both automated and human. This includes public IPs, load balancers, S3 buckets, security groups, and admin panels.

2. Eliminate unnecessary exposure. Review each item on the inventory list and determine if it should be reachable from the entire internet. Many things, such as dev and staging environments and deprecated services, can be moved behind a VPN.

3. Measure and reduce patch latency. With faster exploit development, the time between a vulnerability being public and being weaponized becomes shorter. Teams should measure their patch latency and focus on quickly patching reachable systems.

4. Tighten permissions attached to exposed resources. If a system is compromised, the attacker's access level becomes critical. Review IAM roles and permissions for all reachable systems and apply the principle of least privilege.

5. Implement independent detection methods. While it's impossible to prevent every exploit, having independent detection mechanisms in place can provide early warning. This includes anomaly alarms, unexpected state changes, and cost anomalies that may indicate compromise.