Treat Model-Provider Enforcement as an Ops Event
Attribute every request to an end customer, route bans into your ops loop, and know which model sits under each product
Operator recipe for companies that run agents or resell model access on behalf of many customers. Until now one end customer’s misuse could get a whole organization restricted. Anthropic (20 Oct 2026) launched user profiles so a platform can tell it which end customer each API request is for, and says that if you operate a platform or gateway, or resell API access, "a safety enforcement action can then apply to one end customer instead of your whole organization"; organization webhooks (beta) fire when Anthropic restricts or bans one of your user profiles and when that ends. The same week, a US Department of Homeland Security order (Federal Register, 20 Oct 2026) removed DeepSeek products, and "third-party products and services developed using DeepSeek implementations", from federal civilian agency IT. Your model provider now sits inside your compliance surface in both directions: it can act on one of your customers, and a buyer can reject you for the model you run. Pairs with /playbooks/agent-overreach-containment/ and /playbooks/harness-permission-outside-agent/.
Core Workflows (5)
Each workflow represents a critical business function. Click any workflow to see detailed automation architecture.
Attribute every model request to one end customer
Led by: GatewayCreate a provider-side profile for each end customer and send it with every request your agents make on that customer’s behalf, including background and scheduled runs. Anthropic’s user profiles exist so that enforcement can land on one end customer rather than the whole organization; that only works if no request goes out untagged. Requests made for your own internal work get their own profile, separate from customers.
Subscribe to enforcement events and contain one customer, not the fleet
Led by: Ops AgentRegister for the provider’s restriction and ban events (Anthropic’s organization webhooks have one event for a restriction or ban of a user profile and one for when it ends). On a restriction, pause that customer’s agents and queued jobs, keep everyone else running, preserve the logs for that profile, and open a review. On the end event, resume only after the review is closed. Write the runbook before you need it, because the first event will arrive while customers are mid-task.
Monitor use before the provider has to
Led by: MonitorProviders are being asked about this directly. The European Commission says the AI Office has "sent Requests for Information addressing issues such as security, external evaluation and monitoring how models are used" (EVP Virkkunen, 14 Oct 2026). Those requests go to model providers, and providers push the same expectations down to the platforms between them and end users. Keep a per-customer usage log and your own abuse checks so you can answer a provider’s question with evidence, and catch a problem customer before the provider does.
Keep a near-miss log, not only an incident log
Led by: Incident ReviewerThe UK Financial Conduct Authority says incident reporting "must extend to significant AI failures and near-misses" and that agentic systems "require continuous, conditional assurance" because a system can behave differently as the models beneath it change (speech, Jessica Rusu, 19 Oct 2026; a speech, not a rule change). Record provider warnings, restrictions that were lifted, monitor flags that did not lead to action, and behavior changes after a model update. History of this kind cannot be backfilled when a buyer or regulator first asks for it.
Keep a model bill of materials for each product
Led by: Product OwnerFor every product or service you sell, record which models and model families run underneath it, including open-weight models and any reached through a third party. The DHS order covers products from DeepSeek and its affiliates and also "third-party products and services developed using DeepSeek implementations", so it reaches downstream builders, and an open-weight licence does not take a model outside it. If you sell to US federal civilian agencies, check that list before every model swap.
Sources (4)
- https://docs.claude.com/en/release-notes/overview
- https://www.federalregister.gov/documents/2026/10/20/2026-21447/notice-of-order-regarding-exclusion-and-removal-of-deepseek-products-from-federal-civilian-executive
- https://ec.europa.eu/commission/presscorner/api/files/document/print/en/speech_26_2150/SPEECH_26_2150_EN.pdf
- https://www.fca.org.uk/news/speeches/building-future-finance-age-ai