Two Problems, Not One: Sequencing an AI Security Programme
"AI security" collapses two different problems into one conversation: attacks on the model itself, and what using AI does to everything you already have. IBM's 2026 data says most incidents are the second kind, and this piece argues you should scope for it first.