What this paper finds

• Published browser-agent exploits, a patched automation-tool vulnerability, and a documented extension compromise campaign establish distinct risks; demonstrations are separated from confirmed malicious activity.
• SolvxAI combines configurable blocking of detected external automation with session evidence that remains available when enforcement is disabled. Its own browser agent uses a separate authorization and control protocol.
• An external-agent badge identifies observed automation in an authenticated session. It does not prove who performed every action, establish user intent, or guarantee that all automation is detected.