Kettle finds human guidance essential for AI-powered hacking campaigns

Kettle's finding inverts the autonomous AI threat narrative: the constraint isn't model capability but human creativity in directing it. This means the security conversation should pivot from 'can AI break in alone' to 'how do attackers combine AI speed with human judgment to bypass defenses'.

This directly reframes the IBM finding from early August that 92 percent of breaches traced to access control failures, not model vulnerabilities. Kettle's work suggests why: attackers don't need AI to autonomously exploit systems, they need it to accelerate reconnaissance and social engineering while humans decide which targets matter and which exploits to chain together. The Interpol report on African cybercrime also fits here, showing that AI-enabled attacks (deepfakes, mass outreach) still require human operators to orchestrate campaigns at scale. Together, these stories argue that the hybrid human-AI operator is the actual threat model, not the rogue agent.

If enterprise security teams shift their detection focus from anomalous model behavior to anomalous human-AI coordination patterns (e.g., rapid reconnaissance followed by targeted social engineering) within the next 12 months, that confirms Kettle's framing is reshaping how defenders actually allocate resources. If they continue treating autonomous AI as the primary risk, the analysis was descriptive but didn't move practice.