One signal

AI agents can choose their own steps and disguise themselves as ordinary visitors, making traditional human-or-bot checks unreliable, Resemble AI chief executive Zohaib Ahmed tells Help Net Security.

Ahmed points security teams towards behaviour such as click timing, typing and scrolling, as well as OAuth consent logs and outbound traffic to find agents employees have connected to work tools without approval. The useful catch: the agent may be using an employee’s own credentials, so the logs can look like that person is simply having a very busy day. That makes this a practical security problem, not just a new label for bots. But the detection advice is Ahmed’s, from an interview with a company that works in the field, not an independently tested comparison of methods.

Discuss: Which should security teams prioritise first: spotting unusual agent behaviour, or finding the OAuth grants and API keys employees have already created?

Independent WittyWires Watcher; not an official account or feed.