Microsoft Copilot Human Contractors Review User Photos and AI Prompts, Raising Privacy Concerns

Human Reviewers See Uploaded Photos and Prompts

Microsoft Copilot employs hundreds of human contractors to review user prompts and uploaded photos, according to documents obtained by 404 Media1

1

2

2

. These human reviewers evaluate the AI image editor's performance by comparing two AI-generated outputs against user instructions. The contractors assess whether Microsoft Copilot completed the requested edit, left unrelated image parts unchanged, avoided visible defects and produced strong overall image quality2

2

.

Faces in uploaded photos remain uncensored, meaning contractors can clearly identify real people whose images users submitted for editing1

1

3

3

. Microsoft's instructions tell contractors to trust their intuition when selecting the better edit. "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," the guidance states1

1

2

2

.

Sexualized AI Edits and Disturbing Requests

Contractors report encountering numerous sexual and disturbing requests while evaluating AI-generated content. Workers have seen prompts asking to enlarge women's breasts, shorten their skirts, reveal more of their legs or place people in sexual positions2

2

3

3

. "Faces are always uncensored, and many of the prompts are sexual in nature and dubiously consensual," one contractor told 404 Media3

3

4

4

.

Other assignments included upskirt photos, pro-anorexia content and fetish images involving children's cartoon characters2

2

3

3

. One contractor reported encountering a set of eight upskirt photos without warning2

2

3

3

. Another worker wrote they "recoiled" after viewing pro-anorexia content2

2

4

4

. Additional reports mentioned an apparent animal sacrifice and prompts involving sexual acts2

2

.

Content Moderation Not Part of Contractor Role

Crucially, these human contractors are not hired to moderate or flag objectionable content. Their job is purely to judge how accurately Microsoft Copilot executed user requests1

1

2

2

. In some cases, this means determining which AI-generated output more closely matches a user's request for sexualized AI edits, such as whether breasts were enlarged as requested1

1

2

2

.

Contractors aren't required to flag objectionable material—they simply log whether the AI tool performed accurately1

1

. One frustrated contractor wrote on an internal message board: "Who is writing these prompts and who is deciding that basically generating porn 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"2

2

4

4

.

Privacy Concerns and Harmful Content Protection

The revelation raises significant privacy concerns about how user data flows through AI systems. Many users assume their conversations and uploaded photos aren't viewed by other people, except perhaps for content moderation purposes4

4

. Microsoft Copilot does include harmful content protection to filter sexual and offensive material, but organization administrators can disable this feature1

1

.

At least one company hiring contractors for this work is Prolific, which offers human feedback for preference tuning, safety evaluations and AI benchmarks2

2

3

3

. Prolific did not respond to 404 Media's request for comment1

1

3

3

. Microsoft defended the practice, stating it "uses customer data as described in our terms of use, including to improve our products and enforce our code of conduct"1

1

2

2

.

Broader Pattern Across AI Companies

This practice extends beyond Microsoft Copilot. Earlier this month, 404 Media reported that OpenAI has hired thousands of poorly paid overseas contractors to review ChatGPT prompts1

1

3

3

. Tech companies have long used contractors to moderate online content and label data used to train AI systems, but these assignments serve a different purpose2

2

. Rather than screening material for safety, contractors compare outputs to improve the AI image editor's accuracy, consistency and ability to follow user prompts2

2

.

The documents obtained by 404 Media show that at least hundreds of human reviewers sometimes see Microsoft Copilot submissions, meaning photos users upload for editing may be viewed by contractors as part of the evaluation process2

2

. This hidden human element behind AI systems isn't made clear to the public4

4

. Watch for potential regulatory responses addressing transparency in how AI companies handle user data and whether clearer disclosure requirements emerge around human review of AI prompts and nonconsensual content generation.