What happens if AI stops working?
Almost no organization can answer that question. AI became mission-critical through browser tabs and SaaS features — never passing through the procurement gates, impact analyses, and recovery planning that protect every other critical system.
The continuity gap
Every layer is protected. Except one.
Decades of hard lessons taught IT to protect each layer of the stack. AI joined the stack without the lessons — so far.
Your support team's handle times assume AI drafting. Your developers' velocity assumes code completion. Your analysts' output assumes AI summarization. Staffing, deadlines, and budgets have all been recalibrated around AI being available — which makes its failure an operational event, not an inconvenience.
Failure modes
Six ways AI fails — and why each one is a business event
AI availability risk is not hypothetical. Each of these failure modes has already occurred at scale across the industry.
Vendor outage
Major AI providers have all suffered multi-hour outages. When your provider goes down, every workflow built on it stops with no fallback.
API throttling and quotas
Rate limits and quota exhaustion degrade AI services silently — often during the demand spikes when you need them most.
Model retirement
Vendors deprecate models on their schedule, not yours. Prompts, evaluations, and workflows tuned to a retired model break overnight.
Cost shock
Pricing changes and usage growth can multiply AI spend without warning. Cost resilience is availability's forgotten twin.
Connectivity loss
Every cloud AI dependency assumes the internet path stays up. For plants, hospitals, and field operations, that assumption fails regularly.
Missing from BC planning
The deepest failure mode: AI isn't in the business continuity plan at all, so no one owns the response when any of the above happens.
Not covered elsewhere
Why your existing programs don't close this gap
| Discipline | What it covers | What it misses |
|---|---|---|
| MLOps | Model training, deployment, pipelines | Vendor-hosted services, outages, prompt and RAG continuity |
| AIOps | IT incident correlation | Model routing, AI recovery objectives, dependency mapping |
| Cybersecurity | Threats and access control | Availability, vendor lock-in, cost shock, model deprecation |
| BC / DR | Servers, data, facilities | Agents, copilots, API quotas, knowledge sync at failover |
| AI Consulting | Strategy and use cases | Operational runbooks, quarterly recovery testing, AIR scoring |
Each adjacent discipline covers a fragment of the problem. AI Operational Resilience is the discipline that owns the whole question: which business processes stop when AI stops, and how fast can we bring them back?
Find out what you're exposed to
The AIR Assessment inventories every AI dependency and scores your readiness for each failure mode — in 3 to 6 weeks.