cyberpress.org·18 november 2026

CyberStrikeAI Investigation Highlights Risks of Autonomous Hacking Tools With Disabled Approval Gates

Image for the article: CyberStrikeAI Investigation Highlights Risks of Autonomous Hacking Tools With Disabled Approval Gates
Image via cyberpress.org

An exposed CyberStrikeAI workspace revealed that an AI agent continued offensive security testing after its approval gate was disabled. Huntback researchers analyzed session data from November 16, 2026, finding 16 messages, 240 tool calls, and 221 stored tool executions. The agent logged nine potential vulnerabilities, including two rated critical, but these were not independently verified. The case highlights the need for security teams to investigate reviewer failures and changes to approval settings.