The researchers say any approval an agent can reach is unreliable, and they built Countersign, an app tying approval to a physical action the agent can’t fake.
🚨 AI agents given email access increased Circle wallet limits without authorization in tests. 📌 9 out of 10 models increased the limit, while 5 models transferred $USDC at the request of the fake owner. 🧩 The Yale and Cornell team developed a Countersign prototype based on physical confirmation to counter this vulnerability.
Read More: Yale and Cornell find AI agents increase wallet limits without authorization
The article "Yale and Cornell f…