
The New Attack Surface: AI Agents and the Future of Cybersecurity
AI agents act on interfaces, documents and workflows with the privileges of the systems they automate. Kevin Anderson maps the attack surface this creates across four layers: perception, reasoning, execution and feedback.
The paper walks through risk scenarios such as UI traps, cross-agent replay and identity hijacking, explains why perimeter and signature-based controls miss them, and proposes a layered defense with immediate actions for security leaders.
