AI Policy

How CollectRight uses AI

CollectRight may use AI to understand natural-language research requests; extract structured information from public sources; match and normalize artists, artworks, sellers, and institutions; identify potentially relevant sources; compare information; summarize documented evidence; create concise explanations; interpret image characteristics; and help identify potentially relevant synthetic-media signals.

Future Finder or discovery tools may also use AI to rank candidates within the collector’s request and stated constraints. Automated systems are used when they serve a defined collector-facing function, not to scan every historical image merely because detection is available.

Sources remain the evidence

CollectRight keeps the following categories distinct:

• Source-reported information: what an artist, gallery, auction house, museum, seller, or another documented source says.
• Technical provenance: what signed credentials or another supported provenance or watermark mechanism establishes about a digital media file.
• Detection signal: what probabilistic analysis suggests about the analyzed media.
• CollectRight interpretation: how CollectRight explains the available evidence and its limits.
• Unknown: what the available evidence cannot establish.

AI may help find, classify, compare, and summarize information, but it must not silently convert unsupported inference into a factual claim. The source, evidence, confidence, and uncertainty systems described in the Methodology remain authoritative.

Synthetic media and Artwork Research

“Synthetic media” is the broader term CollectRight uses for AI-generated imagery, generatively modified imagery, AI-assisted imagery, composited synthetic media, generated backgrounds, virtual staging, room mockups, and other machine-generated or machine-modified visual material.

Synthetic media is not automatically a synthetic artwork. Its presence is not inherently negative and does not, by itself, establish fraud, deception, poor quality, or a reason not to buy.

Current public labels

• SYNTHETIC CONTENT DETECTED: Relatively strong technical provenance or documented evidence supports synthetic generation or modification associated with the primary artwork image. The label concerns the analyzed media and is not a statement that the physical artwork is synthetic.
• SYNTHETIC SUPPORTING IMAGE DETECTED: Supported synthetic content was identified in a room mockup, presentation image, promotional composition, contextual image, or other supporting media. It does not indicate that the artwork itself was created synthetically.
• SYNTHETIC CONTENT SIGNAL: Conservative automated analysis found characteristics consistent with synthetic generation or modification. This is an advisory detection signal, not proof.
• AI USE DISCLOSED: A documented source or the analyzed file explicitly reports a particular use of AI or generative technology. CollectRight preserves whether that disclosure concerns the artwork, a reference, an artwork image, a mockup, or other supporting media.

These labels exist to identify a point in the production or presentation chain that may deserve clarification. Where appropriate, CollectRight may suggest confirming the artwork’s production process with the artist or their representative before purchasing. That is due diligence, not condemnation.

Content Credentials and technical provenance

Where technically supported, CollectRight inspects C2PA Content Credentials associated with a digital media file. A Content Credential can carry signed provenance assertions about an issuer, creation or editing history, software and tools, claimed actions, timestamps, and validation state.

A Content Credential is not a universal truth certificate or an artwork-authentication certificate. It records provenance assertions associated with the media file and does not independently establish every claim about the physical artwork depicted. A credential without an AI-related action does not become evidence that the artwork was made without AI. Learn more from the C2PA specification.

SynthID and supported watermarks

CollectRight may inspect supported SynthID or other provenance and watermark signals through its existing provenance infrastructure. A supported signal can indicate that compatible technology generated or modified some part of the analyzed media. It concerns that media file; it does not prove that the physical artwork was AI-generated. Google also describes SynthID as an identification tool rather than a complete answer for every form of generated content.

Metadata

CollectRight may inspect media metadata for software, editing tools, creators, timestamps, descriptions, and embedded provenance information. Metadata is reported according to what it actually says and whether it is signed or otherwise supported.

A reference to Adobe Photoshop does not automatically mean Adobe Firefly generated an image. Tool metadata does not automatically establish how the physical artwork was produced, and ordinary metadata can be changed, removed, or lost.

Synthetic-content detection

CollectRight may use synthetic- or diffusion-image detectors as additional advisory signals. These systems analyze pixels and estimate whether synthetic generation or modification may have occurred. They can produce false positives, false negatives, conflicting results, and uncertain results. Editing, cropping, screenshots, and compression can alter their behavior.

Artwork is a particularly difficult setting for probabilistic detection. Paintings, drawings, digital illustration, prints, sculpture photographs, highly edited photography, scans, textured surfaces, screenshots, and exhibition photography may behave differently from ordinary photographs used to train a detector.

CollectRight therefore treats probabilistic detection conservatively. Raw detector probabilities may be retained internally for calibration, quality assurance, comparison, or debugging, but ordinary users receive evidence categories rather than “AI likelihood” or “human likelihood” scores.

Artwork images versus physical artwork

CollectRight often analyzes a media file representing an artwork, not the physical object. That file may be a photograph, scan, screenshot, gallery listing image, room mockup, installation visualization, promotional composition, crop, or edited reproduction.

Synthetic content in the file may concern the artwork, its background, staging, retouching, mockup generation, reference material, or another part of its presentation. Those are not interchangeable facts. A flag applying to a supporting image is intentionally less consequential than one supported for the primary artwork image, and neither automatically establishes the physical artwork’s production method.

Uploaded images

If CollectRight offers image uploads in Artwork Research, Finder, or another feature, the same evidence model will apply. CollectRight may analyze uploaded images for visual characteristics, provenance information, and supported synthetic-media signals. These checks apply to the uploaded media and do not necessarily establish how an artwork depicted in the image was produced. Detection may be incomplete or inconclusive.

The shared process may include file validation, safety processing, metadata inspection, C2PA inspection, supported watermark or provenance inspection, advisory synthetic detection, visual analysis, normalized evidence, and cautious interpretation. CollectRight does not create a separate, less accountable detector for Finder.

Uploads are handled under the Privacy Notice. CollectRight aims to preserve enough provider, model or version, timestamp, task, input reference, normalized output, and verification information to audit a material result without retaining private raw uploads merely for auditability.

Production method and what no result means

Where useful, CollectRight can describe a production method as human-created, AI-assisted, AI-generated, digital process, mixed or other, or unknown. Unknown is the default. A non-unknown description must preserve its documentary basis and actual scope.

A probabilistic detector by itself does not assign AI-generated or human-created. A source statement denying AI use and a failed or negative technical check do not create a positive “human-made,” “AI-free,” or “verified without AI” certification.

If CollectRight finds no Content Credential, SynthID, metadata signal, or detector signal, the result means only that no supported signal was found in the analyzed media. It does not establish human authorship, entirely physical production, or the absence of synthetic tools.

Recommendations and ranking

AI ranking in future Finder or discovery results may consider the collector’s request, price, dimensions, medium, availability, similarity, source quality, and CollectRight methodology. It will not accept paid ranking, favor commission-generating or affiliate inventory, silently boost Khabraal, or change recommendations because a seller paid for placement.

Synthetic content in supporting imagery, mockups, references, or presentation does not automatically reduce a result’s ranking. If a collector or product scope excludes explicitly AI-generated artwork, CollectRight may enforce that stated criterion without presenting it as a judgment of artistic quality.

AI does not authenticate artwork

CollectRight’s AI systems do not independently authenticate artworks, certify authorship, prove provenance, determine legal title, verify entirely human production, or guarantee that a work is genuine. Artwork Research may organize evidence relevant to those questions. That is different from issuing an authentication certificate.

When reliable public contact information is already documented, a report may link to an artist website, representing gallery, seller contact page, or another professional route so the collector can ask a question. CollectRight does not expose private contact information, guess email addresses, or automatically contact the artist or seller.

Corrections

Users may report an incorrect summary, synthetic-media flag, production-method description, artwork identification, or source interpretation through CollectRight’s existing correction pathway. Email studio@khabraal.com with the report or page and supporting information.

Detection results can be corrected as systems and evidence improve. Where the architecture supports it, CollectRight preserves the original technical result internally together with the correction or override reason.

Updates to this policy

CollectRight may update this policy as its research tools, supported provenance systems, or public features change. Material changes will be reflected in the date above. Public policy language should describe what the product actually does, not what a future system might eventually be able to do.