Google GenAI image safety

[jxnl]

fix(genai): stop auto-sending unsupported safety categories

Describe your changes

This PR resolves a critical production blocker where Google GenAI's SafetySettings caused 400 INVALID_ARGUMENT errors, particularly with multimodal (image) content.

The root cause was that instructor was automatically sending all google.genai.types.HarmCategory enum members as safety settings, including categories unsupported for certain modalities (e.g., HARM_CATEGORY_CIVIC_INTEGRITY or image-specific categories when no image was present).

This fix refactors update_genai_kwargs to:

• Emit only a stable baseline of safety categories by default (HARM_CATEGORY_HATE_SPEECH, HARASSMENT, DANGEROUS_CONTENT at BLOCK_ONLY_HIGH).
• Dynamically include HARM_CATEGORY_IMAGE_* categories only when the request contains image content.
• Properly parse and merge user-provided safety settings (from dict or list) with the default baseline.
• Exclude UNSPECIFIED and JAILBREAK categories.

This ensures that only valid and supported safety categories are sent to the Google GenAI API, preventing the 400 INVALID_ARGUMENT errors. Thorough regression tests have been added and updated to cover these scenarios.

Issue ticket number and link

Linear: 567-261
GitHub: #1773 (#1773)

Checklist before requesting a review

• I have performed a self-review of my code
• If it is a core feature, I have added thorough tests.
• If it is a core feature, I have added documentation.

---

Linear Issue: 567-261

 

[cursor]

Cursor Agent can help with this pull request. Just @cursor in comments and I'll start working on changes in this branch.
_{Learn more about Cursor Agents}

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	instructor	2002f58	Commit Preview URL Branch Preview URL	Jan 16 2026, 05:31 PM