One weekend, Bayer’s EMEA finance team took an urgent call from someone who sounded like the global CFO. The request was to move money fast. No money moved. The team had been trained to recognize the pressure an attacker uses to rush a decision. That training is why it worked.
Kevin Jones, Bayer’s CISO, rebuilt security awareness around AI-written attacks that no longer need bad spelling or broken links to work. Employees cannot use the in-house AI platform until they finish a training module. Anyone running an agent must finish a second module. An agent can authenticate, act on enterprise data and trigger workflows faster than a human can review the activity. That makes running an agent an access-control decision.
What does someone have to prove before they can run an agent with enterprise access?