Agentic Browsers Bypass Security Controls, Researchers Warn

AI agents operating inside browsers can bypass existing cybersecurity controls, Darkreading reports. The finding points to a structural gap: browser-based agents act with user-level permissions, moving faster than current policy enforcement can track.

Most enterprise security architecture was built for human-paced browsing. Agentic systems operate at machine speed, executing tasks across sessions without the friction that controls rely on. That mismatch is not a configuration problem. It is an architectural one.

For French operators deploying agentic tools under AI Act obligations, this matters immediately. Capability without governance creates liability. Regulators will ask who authorized the action. If the answer is an agent running in a browser tab, that is not a satisfying audit trail. Watch for browser vendors and cloud providers to face pressure to build agent-aware permission layers.