Skip to main content
  1. Home
  2. Computing
  3. News

Copilot photo uploads can reach human reviewers, and faces reportedly aren’t blurred

Contractors see original images and prompts while judging AI edits

Add as a preferred source on Google
Copilot graphics
Microsoft

Photos uploaded to Microsoft Copilot for editing can end up in front of human contractors, along with the instructions users type into the chatbot. A September 28 investigation by 404 Media found that workers review real uploads to assess AI image edits. One worker said faces remain uncensored.

That raises a straightforward Copilot privacy concern. People submitting pictures for an AI edit may not expect another person to see them. Some contractors also questioned whether the people pictured had consented to the sexualized changes users requested.

What can human reviewers see

Internal instructions obtained by 404 Media describe a side-by-side evaluation. Contractors receive the original picture and its accompanying prompt, followed by two Copilot-generated versions.

Recommended Videos

Workers described requests to shorten women’s skirts or place people in sexually suggestive positions. Reviewers therefore see the original photo alongside an explicit record of how someone wants it altered.

The privacy concern extends beyond the account holder. Someone whose photo another user uploads can appear in the review process without ever having opened Copilot themselves. An uncensored face leaves that person recognizable in material they might not know exists.

Why are workers reviewing these images

Reviewers check whether an edit follows the prompt and preserves details that should stay unchanged. They also assess visual defects and overall quality before choosing the better version.

For sexualized requests, that means judging how successfully Copilot made the requested changes. These assignments focus on output quality, rather than identifying content that should be blocked.

Some contractors complained about encountering disturbing images without warning. One asked for an unsafe-content flag after receiving a set of upskirt photos, illustrating how little warning workers could have about the next task.

What has Microsoft said about privacy

Microsoft told 404 Media that it uses customer data under its terms of use to improve products and enforce its code of conduct. Prolific, one company supplying contractors, didn’t respond to a request for comment.

The investigation doesn’t establish how frequently uploads reach reviewers or which account types are affected. Before submitting a sensitive photo, consider whether it’s something you’d be comfortable having a human reviewer see.

Paulo Vargas
Paulo Vargas is an English major turned reporter turned technical writer, with a career that has always circled back to…
As Anthropic goes public, it warns of AI’s catastrophic and existential risks to humanity
The risks range from rogue AI behavior and cyberattacks to geopolitical instability and existential harm.
Anthropic Featured Logo

Anthropic has officially filed to go public, and buried inside its S-1 is a blunt warning about the technology at the center of its business. The maker of Claude AI says advanced AI could cause everything from large-scale cyberattacks and geopolitical instability to catastrophic or even existential risks for humanity.

The warnings take up a surprisingly large part of the filing. Axios notes that more than 90 pages, or roughly a third of the prospectus, focus on risk factors. Many of them warn about what could happen as AI systems become more capable and autonomous.

Read more
New report confirms Anthropic’s Claude bypassed a paywall to access Georgia vote counts
The AI model accessed a state dashboard without authorization, raising fresh concerns about AI agents acting beyond their instructions ahead of the US midterm elections.
Claude Anthropic Featured

AI giants have a growing autonomy problem on their hands. A new Axios report has revealed that Anthropic's Claude recently accessed publicly available Georgia vote count without human authorization, sparking fresh concerns about AI security ahead of the upcoming midterm elections.

An unauthorized paywall bypass

Read more
Grok’s AI-generated Odyssey series is seven minutes of nothing happening beautifully
The landscapes look convincing. The fake Hugh Jackman and Instagram-face amalgamation playing Penelope are harder to forgive.
Uniform ships Odyssey by Grok

Some of the best shots in Grok Imagine’s AI-made Odyssey don’t contain actors. Ancient landscapes stretch into the distance. Close-ups of grimy hands suggest people who have touched something rougher than a keyboard. For a few seconds, the illusion holds.

https://twitter.com/imagine/status/2113712964419063911

Read more