Anthropic adds telemetry and roles

Anthropic has added deeper admin controls and telemetry to Claude for business customers as an April 19 outage disrupted Claude 4 API access. (support.claude.com; startupfortune.com) In release notes updated April 18, Anthropic said Enterprise admins can organize users into groups, sync those groups through System for Cross-domain Identity Management, or SCIM, and assign custom roles that control which Claude capabilities each group can use. The same April 9 product update added OpenTelemetry support for Claude Cowork on Team and Enterprise plans. (support.claude.com) OpenTelemetry is a common way companies pipe software events into monitoring systems such as Datadog, Splunk, or Honeycomb. Anthropic’s help center says Claude Cowork can now stream prompts, tool calls, file access, approval decisions, request costs, durations, and errors to a customer’s own collector. (support.claude.com) Anthropic’s identity changes also move Claude closer to standard corporate software controls. Its SCIM guide says Enterprise and Console customers can automatically provision and remove users from an identity provider, while custom roles can gate features such as Claude Cowork, Claude Code, web search, memory, and code execution. (support.claude.com; support.claude.com) Those additions arrived as Anthropic dealt with another reliability test. Startup Fortune reported that on Sunday, April 19, HTTP 503 errors began rising around 10:00 a.m. Pacific Time across Claude 4 API endpoints, with the issue tied to inference routing in West Coast data centers. (startupfortune.com) Anthropic’s public status page has logged several incidents in the past week, including elevated errors on Claude.ai, the API, and Claude Code on April 15, plus an Opus 4.6 error spike on April 16. The page also showed no incidents on April 18 before the April 19 disruption reported elsewhere. (status.claude.com) The product changes and the outage touch the same buyer checklist. Large companies usually want single sign-on, automatic account provisioning, role-based access, spend controls, and logs they can send into existing security and operations systems before they expand an artificial intelligence tool across departments. (support.claude.com; support.claude.com; support.claude.com) Anthropic’s own documentation draws a line between observability and compliance. The OpenTelemetry guide says Cowork activity is not currently captured in audit logs, the Compliance API, or data exports, and says organizations that require formal audit trails should not use Cowork for regulated workloads. (support.claude.com) That leaves Anthropic selling two things to enterprise customers at once: tighter control over who can use Claude and better visibility into what Claude does. April 19 showed that customers evaluating model vendors are likely to measure those controls alongside uptime. (support.claude.com; startupfortune.com)