The _is_image_type_with_blob_content function at line 600 assumes image_url is always a dict by calling .get("url", ""). However, OpenAI's format allows image_url to be a string directly (e.g., {"type": "image_url", "image_url": "https://..."}), as handled in transform_openai_content_part (lines 142-143). When a string is passed, this will raise AttributeError: 'str' object has no attribute 'get', causing a runtime crash during blob redaction.

Also found at:

• sentry_sdk/ai/utils.py:659-660

The _is_image_type_with_blob_content function assumes image_url is always a dict, but OpenAI's format also supports a string shorthand (e.g., {"type": "image_url", "image_url": "data:image/jpeg;base64,..."}). When image_url is a string, calling .get("url", "") on it will raise AttributeError: 'str' object has no attribute 'get'. This causes redact_blob_message_parts to crash when processing messages with the string format, potentially leaking base64 image content to Sentry span data.

Also found at:

• sentry_sdk/ai/utils.py:659-660

Line 685 assumes item["image_url"] is a dict when performing item["image_url"]["url"] = BLOB_DATA_SUBSTITUTE. If image_url is a string (which is valid per OpenAI's format), this will raise TypeError: 'str' object does not support item assignment. This is a separate issue from the detection bug since even if detection were fixed, the redaction would still fail.

Also found at:

• sentry_sdk/ai/consts.py:4-6