converse returns error with enabled citations for text document

[kovalzdw]

Describe the bug

When using Bedrock Converse API with attached txt file error is returned pointing to non-existent PDF file.

botocore.errorfactory.ValidationException: An error occurred (ValidationException) when calling the Converse operation: The model returned the following errors: messages.0.content.1.image.source.bytes: The PDF specified was not valid.

Documentation says that citations should be available for document content.
https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_DocumentBlock.html

Moreover passing wrong extension in 'format' (eg. 'docx') error is returned stating that "Unsupported document format. Only txt and pdf formats are supported when citations are enabled ", meaning that txt should be fine.

Regression Issue

• Select this option if this issue appears to be a regression.

Expected Behavior

Converse should respond properly with citations from document.

Current Behavior

botocore.errorfactory.ValidationException: An error occurred (ValidationException) when calling the Converse operation: The model returned the following errors: messages.0.content.1.image.source.bytes: The PDF specified was not valid.

Reproduction Steps

`import boto3

client = boto3.client("bedrock-runtime")

request = {
"modelId": "eu.anthropic.claude-sonnet-4-5-20250929-v1:0",
"messages": [
{
"role": "user",
"content": [
{
"text": "What is attachemtn content?"
},
{
"document": {
"name": "test-file",
"format": "txt",
"source": {"bytes": b"The sky is blue, the grass is green."}
,
"citations": {
"enabled": True
}
}
}
]
}
],
"inferenceConfig": {
"maxTokens": 16000,
"temperature": 0.4
}
}

response = client.converse(**request)

print(response)`

Possible Solution

No response

Additional Information/Context

No response

SDK version used

boto3-1.42.55 botocore-1.42.55 s3transfer-0.16.0

Environment details (OS name and version, etc.)

Windows 11 Pro 25H2

[adev-code]

Hello @kovalzdw, thanks for reaching out. Per the service Converse API documentation, DocumentSource is a tagged union with four options: bytes, s3Location, text, and content. The bytes field is described as The raw bytes for the document and is intended for binary file content (e.g. the raw bytes of a PDF or DOCX file). The text field is described as The text content of the document source and for testing, could you try this option when your document content is a plain text string as show below?

"document": {
    "name": "test-file",
    "format": "txt",
    "source": {"text": "The sky is blue, the grass is green."},
    "citations": {
        "enabled": True
    }
}

In regards to the error message you received messages.0.content.1.image.source.bytes: The PDF specified was not valid, I have reached out to the Bedrock service team to improve the error messaging to avoid confusion. I do not have a timeline yet in this regards and if/when this will be accepted. Although please refer to service doc or our changelogs (boto3 and botocore) for updates

Internal Ref: V2119492675

[kovalzdw]

Hi @adev-code, thank You for quick answer. Passing txt file content as text, not bytes, works just fine. The same situation is with PDF files passed as bytes, they work correctly. It seems like the problem is only with txt documents passed as bytes.