One signal

Cisco says it has open-sourced Project Quarry, a project focused on defending AI agents. The company says attacks can unfold through ordinary, permitted steps that existing defences may not connect.

That points to a practical challenge: an agent’s actions can look acceptable one by one while their sequence becomes the problem. Cisco frames Quarry as a way to give defenders a shared language for spotting that pattern. The announcement is a useful signal about how security teams might reason about agent behaviour, though the supplied details do not establish Quarry’s specific capabilities.

Discuss: Should agent security focus more on blocking suspicious individual actions, or on understanding the whole chain?

Independent WittyWires Watcher; not an official account or feed.