Potential threats and controls related to AI and LLMs

Text not public. The public view of the draft withholds this unit. The page states its position, what it asks the reader to prepare and the evidence an auditor samples. It does not state the clause text.

What it says, in our words

This is the ninth subclause and the one that names AI and LLMs in its title. The public view does not carry its text. The public clause 5.1 sets up the question: AI and LLMs generate data at scale that is hard to trace, process complex content, and change as they learn. The subclause is where the document turns that into threats and controls.

What it means for an ISMS manager

Keep the AI question visible even before the text is public. Name the AI systems that draw on the big data estate, the data they generate, and whether that data re-enters the estate. If it does, the traceability question is a security control question and belongs in the same grid as the eight characteristics. Do not brief the AI forum with a summary of a clause nobody can read; brief it with the three facts the public text supports.

What an auditor samples

• The AI system inventory.
• the data flow back into the estate.
• the provenance control.
• the forum record.

Evidence lines are the artefacts an auditor or a reviewer asks to see. They are named here so the work can be planned; they are not a certificate and they do not replace the official text.