A Microsoft engineering leader named Shane Jones is accusing AI company OpenAI of failing to address security flaws in its powerful image generator DALL-E 3 that he claims could enable creation of violent and explicit content.
After Jones says he discovered a way to bypass DALL-E 3’s safety filters and generated “disturbing” images, he reported the vulnerability privately to Microsoft and OpenAI. But frustrated by what he felt was inadequate response, last month Jones tried taking his concerns public in a LinkedIn post urging OpenAI to pull DALL-E 3 offline until fixed.
However, Jones alleges Microsoft’s legal department demanded he delete that post and has since blocked his attempts to sound the alarm about risks he believes DALL-E 3 poses if exploits remain unaddressed.
Now the self-described whistleblower is appealing to Washington lawmakers and the state attorney general, sending letters arguing the AI system represents a “public safety risk” requiring accountability. He draws a parallel to the recent high-profile instance of deepfake pornography falsely depicting musician Taylor Swift.
While details remain unclear, Jones suggests DALL-E 3’s reported flaws could similarly enable AI synthesis of violent and explicit content if left unaddressed. He is urging leaders to institute safeguards protecting employees like himself who call out perceived AI dangers despite employer intimidation.
Jones believes companies must be held responsible for disclosing and fixing known AI vulnerabilities rather than silencing internal dissent. He hopes shining a spotlight on DALL-E 3’s alleged gaps pressures action before harm occurs, though both Microsoft and OpenAI claim his reported bypass method failed.
