Microsoft Copilot Reviewers Inspect Sexually Explicit Prompts

Human contractors hired to evaluate Microsoft Copilot AI image-editing outputs are routinely exposed to explicit, dubious, and non-consensual sexual prompts uploaded by users. Internal documents obtained by 404 Media reveal that a workforce of at least hundreds of contract workers examines uploaded photos with uncensored faces alongside generated edits to grade model performance.

Rather than moderating unsafe content, these reviewers are instructed to evaluate how successfully the AI fulfills user requests. Internal message board logs indicate contractors frequently encounter requests involving foot fetishes, pro-anorexia content, shortened clothing, and enlarged body parts.

The Evaluation Workflow and Contractor Duties

The review process requires contractors to look at a user’s original text prompt, an uploaded picture, and two AI-generated edits. Workers must then select the better result based on gut feel and specific visual criteria.

According to instruction guides seen by researchers, Microsoft explicitly tells contractors to rely on their intuition:

Trust your intuition – when you glance at the two edited images side by side, which one immediately feels like the better edit? Your gut reaction as a human viewer matters.

Contractors check whether the edited image follows instructions, preserves unchanged areas, avoids visual artifacts, and maintains overall quality. They are not tasked with flagging inappropriate material, meaning offensive or disturbing inputs enter the queue without prior warning flags.

Disturbing Material and Privacy Concerns

Internal message board logs show that workers frequently express distress over their assignments. Contractors reported seeing suggestive images of minors, animal sacrifices, voyeuristic upskirt photos, and body modification requests.

One anonymous contractor described the content environment:

Faces are always uncensored, and many of the prompts are sexual in nature and dubiously consensual.

Another worker questioned the direction of the tool:

Who is writing these prompts and who is deciding that basically generating is what Copilot is now focused on? It’s hard to take things seriously when my focus has to be what model generated the appropriate bust size or which middle-aged woman was put in the appropriate sexually suggestive position.

The situation highlights a hidden human cost behind consumer AI features. While users generally assume their interactions and uploaded images remain private between themselves and the software, hundreds of external contractors may examine the data.

Company Responses and Third-Party Hiring

At least one of the third-party platforms utilized for hiring these reviewers is Prolific, a service that provides human feedback for preference tuning and AI benchmarking. Prolific’s own researcher guidelines acknowledge the inherent difficulty of shielding participants from explicit generative AI content, noting that researchers cannot fully control what inputs are displayed.

In a statement regarding data usage, a Microsoft spokesperson told 404 Media:

Microsoft uses customer data as described in our terms of use, including to improve our products and enforce our code of conduct.

While Microsoft’s consumer terms prohibit adult content, child sexual exploitation, disturbing imagery, and non-consensual deepfakes, the latest disclosures demonstrate that automated screening systems do not entirely prevent explicit user inputs from reaching human quality-review queues.

Frequently Asked Questions

What are Microsoft Copilot reviewers doing?

Contractors evaluate the quality of AI-generated image edits by comparing original user prompts and uploaded pictures against two alternative AI outputs, selecting the version that best meets the criteria.

Are user uploads private when using Copilot?

Although consumers often assume their interactions are entirely private, internal documents confirm that human contractors routinely review uploaded photos and text prompts for quality assessment purposes.

What kind of explicit content have contractors encountered?

Internal logs show workers are exposed to upskirt photographs, sexualized body modifications, pro-anorexia imagery, foot fetishes involving cartoon characters, and suggestive scenarios involving minors.

Are reviewers responsible for moderating unsafe content?

No. Contractors are hired exclusively to grade the technical performance and visual quality of the AI model, not to vet or flag offensive and inappropriate material.

Which companies hire these human reviewers?

Reports identify platforms such as Prolific as suppliers of human labor used for preference tuning, safety evaluations, and benchmark testing across the artificial intelligence industry.