Security experts urge removing humans from patching loops

Defending against AI attackers requires a shift from 'meeting speed' to machine speed, necessitating the removal of humans from defense loops. Traditional detection is insufficient; organizations need high-fidelity forensic monitoring to understand attacker patterns. Deception technology, such as honeypots and digital twins, is critical for luring gullible AI agents into surveilled environments. Companies like Cloudflare, Tracebit, and ThingsCanary are involved in this space.

Patching cycles must be instant and non-human; the risk of not patching now outweighs the risk of breakage. Security teams should submit fixes directly rather than just reporting vulnerabilities, making it easier to fix than to triage. The 'one-way door' vs 'two-way door' framework suggests that reversible changes (two-way doors) should be fully automated by AI, while irreversible changes (one-way doors) may retain human oversight.

Caleb Sima argues that organizations must accept the risk of fast, automated patching and that security teams must own the remediation process. Simon Maple contends that providers of autonomous systems also bear accountability for the decisions made by their agents. Accountability for AI actions rests with the user who wields the agent, similar to a parent's responsibility for a child's actions. The consensus is that human oversight in real-time defense is a bottleneck that must be eliminated to maintain security against AI-driven threats.