The Make WordPress AI team has discussed an early concept called a trust ladder for AI agents in WordPress. The practical question is simple: if an AI tool can do things inside a website, who decides what it is allowed to touch?
This is not a finished feature and not an implementation proposal yet. But it is worth watching if you run a WordPress site or WooCommerce store, because AI permissions are where convenience can quietly turn into a very expensive mistake.
Key Takeaways
- WordPress AI contributors discussed a trust ladder for AI agent permissions on 5 August 2026.
- The idea includes four trust levels plus a zero or kill-switch level.
- Site owners could set which AI abilities are exposed to an agent.
- WordPress.com already uses granular ability controls and explicit approval for destructive actions.
- This is early-stage discussion, not an implementation proposal yet.
AI as a user with permissions
The discussion compared AI agent access to familiar WordPress user roles, such as contributor, editor, and administrator. The group also discussed whether an AI agent should be treated as a separate entity or as a new type of user, with the conversation leaning toward type of user.
That matters because WordPress owners already understand roles. A contributor should not be able to delete products, change payment settings, or rewrite legal pages. An AI agent should not magically skip that logic just because it sounds clever in a demo.
How the trust ladder would work
The proposed idea includes four trust levels, plus a zero or kill-switch level. Abilities could declare the trust level they require. A site owner could then set a trust threshold that decides which abilities are exposed to an AI agent.
One working answer discussed for MCP registration was that abilities above the site’s trust threshold simply would not be registered with the agent at all. In plain English: if the site owner says the agent is not trusted enough, the risky tools should not even appear in its toolbox.
What WordPress.com already does
The source notes that WordPress.com already has related controls in place. Users can enable or disable individual abilities, external agent access has granular read and write permissions, and destructive actions always require explicit approval regardless of trust settings.
That last part is the one business owners should underline. AI assistance is useful. AI deleting content, changing store data, or publishing the wrong thing without approval is not assistance. It is just automation with better vocabulary.
Why this matters for sites and stores
If AI agents get real WordPress abilities, permission design becomes a business risk issue, not just a technical feature. Store owners will want clear limits around orders, products, coupons, customer data, payments, redirects, SEO settings, and publishing workflows.
The meeting also raised audit trails, transparency, and AI-generated content disclosure requirements in some countries. Those are not cosmetic details. If an AI tool changes something on a live site, someone needs to know what changed, when it changed, and who or what approved it.
This is not shipping yet
The trust ladder is still an early-stage discussion. The Make WordPress AI team is welcoming further thoughts, and the source does not say this is scheduled for WordPress Core or WordPress 7.2.
The same weekly summary also mentioned WordCamp US prompt bar planning, a structured content and abilities proposal awaiting review, and continued testing for Embeddings and Streaming support, with hope that those could reach WordPress 7.2 later in 2026.
For more detail, check the original Make WordPress AI summary: AI Contributor Weekly Summary – 5 August 2026.
Frequently Asked Questions
What is the WordPress AI trust ladder?
It is an early-stage concept discussed by the Make WordPress AI team for setting trust levels for AI agents in WordPress. The idea includes four trust levels plus a zero or kill-switch level.
How would a site owner limit an AI agent in WordPress?
Under the discussed concept, abilities could declare the trust level they require, and a site owner could set a trust threshold. Abilities above that threshold would not be exposed to the AI agent.
Are AI agents being treated like WordPress users?
The discussion compared AI agent access to WordPress roles such as contributor, editor, and administrator. Participants leaned toward treating an AI agent as a type of user, but this is not an implementation proposal yet.
Does WordPress.com already have AI permission controls?
The source says WordPress.com already lets users enable or disable individual abilities, uses granular read and write permissions for external agent access, and requires explicit approval for destructive actions.
Is the trust ladder planned for WordPress 7.2?
The source does not say the trust ladder is scheduled for WordPress 7.2. It says Embeddings and Streaming support are hoped to reach WordPress 7.2 later in 2026 after more testing and review.