AI’s vulnerability to adversarial attack is not futuristic, and there are reasonable measures that should be taken now to address the risk.

Source: Addressing the Security Risks of AI – Lawfare