Adversaries hijacked AI security tools at 90+ organizations. The next wave has write access to the firewall

Attackers used prompt injection to compromise AI security tools at more than 90 organizations, mainly for credential theft and crypto theft.

The article warns that next-generation autonomous SOC agents can do more than read data; they may change firewall rules, IAM policies, and endpoint states through approved privileged actions.

That shift from read-only AI to write-capable agents raises the blast radius from data loss to direct infrastructure manipulation if compromised.

As these systems move toward production, strong governance, privilege controls, and auditing are becoming critical.