
Does an AI Agent Follow Instructions It Sees on Screen?
Often, yes. An agent reads page text and its task as one stream, and benchmarks show planted instructions succeed. Check what it can do when fooled.
Research topic
Research on computer-use agents: AI agent security, tokenization and checks for sensitive data access before allowing enterprise workflows to run.

Often, yes. An agent reads page text and its task as one stream, and benchmarks show planted instructions succeed. Check what it can do when fooled.

Yes, when it works through the screen: computer-use agents perceive the whole rendered page, sidebars and tabs included. Check what else was visible.

Only when payments, submissions and account changes are gated by action type and destination, since an injection becomes a transaction. Check what runs.

Seventeen enterprise agents can work on screens with PII today, and for most the model receives whatever the account may see. Check for tokens first.