Ambra AI encompasses all generative artificial intelligence features within the THRON platform: generating and editing images and other visual content, generating text such as product and asset descriptions and their translations, and conversational features enabling natural language interactions with the platform. Like any system of this kind, Ambra is not infallible. Here you will find why it may make mistakes, what to verify before using a result, and how to report an issue to us.

Why Ambra sometimes makes mistakes or invents information

Ambra generates responses by predicting the most plausible content based on what it learned during training, rather than querying a structured database in real time. This probabilistic nature is an intrinsic characteristic of current generative AI models and produces a phenomenon known as hallucination: a confidently formulated answer that may nevertheless be inaccurate, incomplete, or inconsistent with what actually exists in your platform.

This applies to text as well as images and other visual content that Ambra generates or modifies: in these cases too, the result may contain errors or fail to faithfully match what you requested.

In practice, Ambra may:

• provide outdated information, especially regarding recent events or data;

• confidently describe an attribute, usage right, or product data point that does not actually exist in the platform;

• summarize a document or asset inaccurately, omitting important details present in the original;

• generate or edit images with features, details, or elements that are inaccurate compared to the original product or context;

• execute operations or actions on the platform that differ from or do not match what you actually requested.

Different responses between users: why this happens

Ambra operates strictly within the permissions and roles assigned to each user on the THRON platform. Two people asking the same question may therefore receive different answers: this is not an error, but rather reflects the visibility scope of each respective user.

Always verify the result before using it

Human verification is always required before:

• using, publishing, or sharing content generated or modified by Ambra;

• considering an operation that you asked Ambra to perform on the platform as completed.

Pay special attention when requests involve usage rights, contractual data, regulatory information, or content intended for third parties or the public.

Marking of synthetic content

Images generated or modified using artificial intelligence by Ambra are handled, where applicable, in accordance with the transparency requirements laid down by Regulation (EU) 2024/1689 (AI Act) concerning synthetic content.

• Do not remove any indications of artificial origin when redistributing content to third parties.

• Always verify the origin of content before using it publicly or commercially.

Data processing and privacy

To learn how THRON processes personal data when using Ambra and the platform, please refer to the privacy policy.

How to report an incorrect response

If a response seems incorrect, unclear, or misaligned with your platform data, write to support@thron.com providing the context of your request and, if possible, the response received. Every report helps us improve Ambra's performance over time.

## Additional articles
- Getting started with Ambra AI MCP
- Asset Enrichment: automatic metadata enrichment with AI

Need help?

For technical issues, contact support@thron.com.