Ironscales strategist warns AI agents may become phishing targets

Paxson warns that AI agents themselves could become targets, with attackers manipulating them through hidden instructions or gradual persuasion. He suggests organizations should combine human oversight with AI, ensuring explicit handoffs for high-risk actions like payments or access changes. Paxson advocates for ‘separation of duties’—a control traditionally used for financial transactions—to mitigate risks where a single fooled agent or employee cannot complete an attack alone.