Boardroom Answers · AI & Data · AI, Data & Analytics
You are concentrated on Anthropic. When Claude has an outage or deprecates your model mid-quarter, what happens to my board meeting?
The question a Chief Automation Officer (CAO) asks.
The short answer
Outage is an absorbed event: stored artifacts stay up, new generations fail over automatically to a second provider under the same validation gates, deterministic analytics never touch a model — and every fallback is audit-stamped.
The full executive answer
Your board meeting proceeds, and here is the mechanical chain that makes that true. First, most of what a board consumes is already generated and stored — insights are persisted artifacts, not live model calls, so an outage cannot take existing decision material off the screen. Second, for new generations there is automatic cross-provider failover baked into the engine: if the Anthropic call fails, the same request runs against OpenAI’s GPT-4o with the same schema enforcement and the same validation gates, and the switch is stamped in the audit log — a circuit breaker even stops hammering a failing provider and routes around it for a cooldown window. This is not a disaster-recovery document; it is the ordinary code path, exercised whenever a provider hiccups.
Third, the deterministic layer — assessments, evidence maps, forecasts, the causal analytics — involves no external model at all and keeps working through any AI-provider event whatsoever. On deprecation rather than outage: every model is resolved through configuration against a governed registry, so a model retirement is a config change plus a forced governance entry, and the cache fingerprint ensures old-model outputs are never silently passed off as new-model ones. Honest residual concentration: both frontier providers are US companies and the fallback tier is thinner than the primary — full multi-region and additional-provider diversity is roadmap, priced against real enterprise demand. But single-provider failure is already an absorbed event, not an incident.
Grounded in: NIST AI RMF MANAGE 3 (third-party and supply-chain risk); ISO/IEC 42001 supplier controls; operational-resilience expectations (DORA-style third-party concentration analysis as vocabulary).
The natural next questions
Related governed answers
- LLM bills are where automation programmes go to die. Walk me through your cost controls — caching, model tiering, budgets — with specifics, not adjectives.?
- My uploaded board packs contain names, emails, salaries. What personal data actually reaches Anthropic and OpenAI, and what is your legal basis story?
- What can your AI actually do autonomously — can it act, decide, spend, or trigger anything without a human — and where exactly are the human-in-the-loop points?
Want this answered live, on your data?