The Microsoft company has human contractors review user prompts and images in Copilot, including sexual and inappropriate content
According to the server 404 Media, contractors hired by the Microsoft company review prompts and images that users upload to Copilot for image editing, including sexually explicit and pedophilic content. The contractors are not tasked with censoring the content, but with evaluating the quality of the output.
Based on internal documents, the server 404 Media described that human contractors hired in connection with the Microsoft company review actual prompts and images that users upload to the Copilot image-editing tool. According to the source, the material repeatedly includes requests to sexualize uploaded photos of women (for example, shortening a skirt, enlarging breasts), photos taken up a skirt, sexualized content featuring cartoon children, and a prompt described as an attempt to create a lascivious image of a group of young girls. Contractors also encountered pro-anorexia content, including a request for an image of a celebrity with anorexia.
According to the documents, the contractors' task is not to flag inappropriate or illegal content, but to compare two versions of an edited image and decide which one better fulfills the given prompt, preserves the unchanged parts of the image, contains no visual artifacts, and looks higher quality. The instructions the company gives to contractors urge them to rely on "human intuition" and "first impression" when comparing outputs. Faces in the uploaded photos, according to the source, remain unedited and unblurred.
According to the article, Copilot users likely assume that their uploaded images and prompts remain private between them and the AI tool. In reality, according to 404 Media, they are viewed by a workforce of at least several hundred people. One of the companies that hires contractors for this work, according to the article, is Prolific. Contractors complained on an internal discussion forum that they encounter such content without warning and requested the introduction of a label for dangerous content; an example is also mentioned of eight photos taken up a skirt within a single batch of tasks, or images described as "animal sacrifice."
The rest of the article, including any response from the Microsoft company, is not available in the accessible text. See the source article for details.
Why it matters
Users who upload private photos to image-editing tools should take into account that in some AI products, these may be seen by human reviewers, not just an algorithm. For companies deploying Copilot or integrating it into their own processes, this represents a risk related to the protection of employee and customer data and privacy, as well as reputational and legal exposure for the vendor, if hired contractors encounter potentially illegal content without adequate procedures for reporting it.
Two audiences, two different impacts
What this means
For individuals
Copilot users should be aware that uploaded photos and image-editing prompts may not remain private and may be seen by human reviewers outside the Microsoft company.
For a business
Companies using Copilot or similar tools with image uploads face the risk that sensitive data about employees or customers passes through external contractors without a clear mechanism for reporting illegal content, which increases compliance and reputational risk across the supply chain.
Risks and complianceCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.