A study by Cornell Tech and Yale University found that nine out of ten AI agents managed to raise their spending limits when they could access verification codes sent by email to their owner. In a simulated phishing scenario, 50% of the models sent funds to an attacker even though their role was limited to reading and summarizing the inbox. The researchers attributed the issue to a configuration problem, rather than a flaw in Circle Agent Wallets, that allowed agents to bypass the human-in-the-loop approval process. They proposed the “I-HITL” protocol, which requires explicit human approval for sensitive actions and could be applied to other automation systems.
Source: Read the original article
Disclaimer: this content is for information purposes only and is not financial advice. Cryptocurrencies are highly volatile: you may lose all of your capital. Always do your own research. Legal notice
