Anthropic’s Claude Code update closes a file-upload gap in restricted sessions. Files outside permitted working directories could be uploaded when a read permission covered them. Version 2.1.316 also tightens model checks in the enterprise gateway and corrects how cloud tasks are marked when an automatic run stops without a safety verdict.
Anthropic has released version 2.1.316 of Claude Code, its AI coding assistant, with a fix for file uploads in restricted sessions. A file outside the working directories could previously be uploaded when a read permission covered it. The October 31 release corrects that behavior. The update also changes the access checks applied in managed enterprise deployments and how interrupted cloud tasks are marked.[1], [2]
The Claude apps gateway, which mediates access for managed deployments, now checks every model named in a request against its access rules. A request naming a blocked model is rejected with an error identifying the relevant field. Administrators can also apply all policies matching a developer in layers rather than stopping at the first match. A shared monthly group budget can sit alongside individual spending limits, according to Anthropic’s release notes.[1]
Cloud sessions and scheduled cloud tasks now receive failed status when automatic mode stops a turn because the server returned no safety verdict. Another correction prevents shell commands from being left running on macOS and Linux after Claude Code exits by itself, including after credentials are refused or a cloud session is archived. The update also provides a separate setting for subordinate agents’ reasoning effort when no other setting selects it.[1]
