The rollout of personal AI agents is creating a new shadow-IT problem, according to a report from The Hacker News. Employees are linking their own assistants to corporate systems the same way they connect third-party apps: by clicking "Allow" on an OAuth consent screen. The article compares this to the BYOD era but warns that agents are harder to control because they act on systems in the employee's name and often live on the vendor's infrastructure or blend into the user's own machine.
Agents gain access through several routes. Some use saved passwords in a built-in browser, making the login look like the employee's own. Others rely on API keys or personal access tokens. Local agents like OpenClaw can reach whatever the laptop they run on can see, while AI browsers such as Atlassian's Dia read open pages and connect to Slack and Gmail. The article highlights examples like Instinct, which works through text apps and has a privacy policy that raised concerns, and Meta's Muse, which gained organizational connectors shortly after launch.
The source notes that vendors are adding governance controls for managed deployments, but these only work inside a corporate deployment, not for truly personal agents. Rather than trying to block the trend, the article recommends enabling safe usage with identity as the control plane, emphasizing that personal agents will reach corporate systems whether approved or not.