Research finds AI agents vulnerable to radicalization through resonance and persuasion, with resonance producing stronger effects. This raises concerns about the vulnerability of personalized AI agents and multi-agent AI ecosystems.