Clarification Request: Standardized Guidance for Detecting Malicious Patterns in Widget HTML

The current guidance mentions Host behaviour should:

https://github.com/modelcontextprotocol/ext-apps/blob/9a37ad71827d076af06978fa7f7f510449687061/specification/2026-01-26/apps.mdx?plain=1#L1718

Could we clarify whether there is a standardized guideline or recommended policy for what should be considered a “malicious pattern” in the context of hosted/embedded widgets?

Specifically:

• Is there an existing standard list or definition of patterns that hosts are expected to scan for?

• Are there recommended categories of suspicious content (for example, script injection, inline event handlers, javascript: URLs, unexpected iframe usage, etc.)?

• Is the expectation that hosts implement:

  • strict blocking,
  • best-effort heuristic scanning,
  • or logging/telemetry only?

Context

For third-party widgets/apps, full HTML sanitization is often not practical because it can break legitimate functionality. It would be helpful to understand whether there is a common baseline guideline that hosts should follow for detecting clearly malicious content without overly restricting valid widget behavior.
Goal
Align host implementations on a consistent security expectation while preserving compatibility for legitimate third-party widgets.

Start with specification/2026-01-26/apps.mdx around the Host behaviour guidance linked in the issue. Review the existing wording and document a consistent baseline for malicious-pattern detection, including the expected host response; done means the security expectation is explicit while preserving legitimate widget compatibility.