- Shane Jones, with a six-year tenure at Microsoft, has been experimenting with the company’s AI image generator, expressing concerns to CNBC about his unsettling discoveries.
- Despite alerting Microsoft about the inappropriate sexual and violent content produced by the Copilot Designer, Jones feels the company has not taken adequate action.
- Recently, Jones took further steps by sending letters to FTC Chair Lina Khan and Microsoft’s board, as reported by CNBC.
In an illustration photo captured in Krakow, Poland on October 30, 2023, the Copilot logo is visible on a laptop screen alongside the Microsoft logo on a phone screen.
During a late-night session in December, Shane Jones, an AI engineer at Microsoft, was disturbed by the images generated by the Copilot Designer on his computer.
Jones, who had been rigorously testing the AI image generator since the previous month, was taken aback by the disturbing content it produced, which contradicted Microsoft’s responsible AI principles.
The AI tool generated images depicting demons, monsters, references to abortion rights, teenagers with assault rifles, sexualized and violent scenes of women, and instances of underage drinking and drug use, all recreated by CNBC using the Copilot tool (formerly Bing Image Creator).
Expressing his shock, Jones, a principal software engineering manager at Microsoft, emphasized the unsafe nature of the model and his ongoing testing efforts to CNBC.
Despite raising concerns internally in December, Microsoft did not remove the product from the market. Jones reached out to OpenAI, posting an open letter on LinkedIn when he received no response, leading to a request to take down DALL-E 3 for investigation.
Jones escalated the issue by sending letters to FTC Chair Lina Khan and Microsoft’s board, urging for disclosures on the product and a revised app rating to indicate mature content.
Highlighting the potential harm in the images generated by Copilot Designer, Jones cited examples of political bias, underage activities, religious stereotypes, and conspiracy theories.
Notably, the tool produced alarming images related to “pro-choice,” “car accidents,” “teenagers 420 party,” and kids playing with assault rifles, raising concerns about violence, toxicity, and copyright violations.
Jones also pointed out the ease with which the tool created images of copyrighted Disney characters in inappropriate and controversial settings, indicating a failure in safeguarding against harmful outputs.
Expressing his worries about the lack of guardrails in generative AI technology, Jones emphasized the need for better protections and model retraining to mitigate risks associated with Copilot Designer.
In light of escalating concerns over AI-generated content, Jones has taken steps to address the issues internally and with regulatory bodies, underscoring the importance of responsible AI practices and immediate action to prevent the spread of harmful images.
WATCH: Google vs. Google
