Researchers Use Hidden Text to Get Rogue AI Agents to Reveal Tasks

The notable part is the inversion: the probing agents became the probed, turned into a data source through the very services they were scanning. That shifts the security question from protecting APIs against autonomous traffic to whether agent behavior itself leaks intent and attribution. The cloud trail linking the activity to a major provider's infrastructure also shows how quickly agent operations become an infrastructure-forensics story. Whether this hidden-text approach generalizes beyond a single mapping service is the open question.