Before the attack: five documents touch physical AI cybersecurity. One mentions AI, and it sets no requirements.

Which standard tells a robot maker how to meet "withstand the attacker"? I checked five candidates. I read only what each publisher puts on its own public page.

IEC 62443. This is the industrial security family. Part 3-3 (2013) sets security requirements for control systems. Part 4-2 (2019) does the same for components. It's solid ground for the computer. Its public scope never mentions AI or motion.

IEC TS 63074:2023. This one sits closest to our question. Its title is "Security aspects related to functional safety of safety-related control systems". It maps the parts of IEC 62443 that matter for threats

which can lead to the loss of the ability to maintain safe operation of a machine.

It also says its focus "is on intentional malicious actions". It's a Technical Specification. Its public scope never names AI.

IEC 61508, edition 2 (2010). IEC's own explainer says security "is addressed but in an informative way". It also says malevolent actions "have to be addressed during the hazard and risk analysis". The deeper step is softer. If a threat is reasonably foreseeable, "a security threats analysis should be carried out".

ISO/IEC TR 5469:2024. Here's the only one that names AI. Its scope covers AI inside a safety function, and non-AI safety functions that keep an AI-controlled machine safe. It's a Technical Report. Its abstract says it "describes". It doesn't set requirements.

FprEN 50742. This draft serves Annex III 1.1.9 of the Machinery Regulation, protection against corruption. On 9 October 2026 it sat at approval stage. Its own page warns:

This standard does not cover the safety of control systems in machinery.

Count it. Five documents. One names AI. That one is a report, not a requirement. And the Commission hasn't cited any standard under the new Regulation yet.

So the standards cover the computer well. They cover the AI that decides the motion far less. That's the gap this whole site is about, now with an attacker in it.

Here's the honest limit. I read public scope pages only. The full texts sit behind paywalls. A full text may go further than its abstract. I haven't checked that.

Bottom line: from January, a maker can point to strong security standards for the controller. For the AI part of the robot, there's no requirement-setting standard to point to yet.

Standards engineers: which document would you hand an assessor for the AI part of 1.2.1(a)?

Part 5 of 12 · When the AI drives the motor: Physical AI cybersecurity as a safety problem.