OpenClaw v2026.4.21 Upgrades Default Image Model and Tightens Permissions

OpenClaw v2026.4.21 makes three notable changes: it switches the default image model to gpt-image-2, tightens command permissions so only owners can run them, and fixes Slack thread aliasing. The release, published on April 21, continues a run of April updates that put security and control ahead of convenience.

The most visible change for users is the new default image model. OpenClaw now uses gpt-image-2 for image generation by default, replacing the prior default and aligning the tool with a newer image model in OpenAI’s lineup. For people building automations around image creation, the default model matters because it determines what kind of output quality, style consistency, and prompt behavior they get without having to configure anything manually.

OpenClaw is an automation layer, so defaults carry real operational weight. If a workspace relies on image generation as part of a workflow, changing the model can affect how prompts are interpreted and how much editing is needed after generation. The update does not appear to remove support for other models, but it does change the out-of-the-box behavior for new and existing setups that depend on the default.

The more security-focused part of the release is the permission change. Command permissions are now owner-only, which means only the workspace owner can trigger those commands. In practical terms, that reduces the chance that another user in a shared Slack or Teams environment can run an administrative command by accident or without approval.

This kind of permission tightening is a common defensive move in automation systems. Commands that can change settings, trigger workflows, or touch connected services are the sort of thing that should not be available to every participant in a channel by default. By restricting them to owners, OpenClaw reduces the attack surface and lowers the risk of mistaken execution in busy team chats.

The third fix addresses Slack thread aliasing. In Slack integrations, thread aliasing refers to the mapping between a thread and the identifier used by the app to track it. If that mapping breaks, replies can land in the wrong place, thread context can get lost, or the automation can fail to follow the conversation correctly.

Fixing thread aliasing is less visible than changing a model or permission setting, but it matters for reliability. Many AI workflow tools depend on thread context to know which conversation they are responding to, especially when multiple users are interacting in the same channel. A bug in that mapping can create confusing behavior fast, even when the rest of the system is working as intended.

The April release cadence suggests a broader pattern of hardening rather than feature expansion. Instead of adding flashy capabilities, OpenClaw is spending effort on defaults, access control, and integration correctness. Those are the areas that tend to matter most once a tool moves from testing into daily use inside a team workspace.