Todd Bishop reports via GeekWire: Microsoft AI engineering leaders addressed the issue in early December after discovering a vulnerability in OpenAI's DALL-E 3 image generator that allowed users to bypass safety guardrails and create violent and explicit images. He said the company had thwarted his previous attempts to gain public attention. The emergence of explicit deepfake images of Taylor Swift last week was “an example of the kind of abuse that I was concerned about and why I urged OpenAI to remove DALL-E 3 from public use. This is also why I raised my concerns with Microsoft,” Schoen wrote. Jones, Microsoft's chief software engineering officer, said this in a letter to the Washington state attorney general and members of Congress on Tuesday.
404 Media reported last week that the blatantly fake image of Swift came from “a specific Telegram group that specializes in abusive images of women,” and that at least one of the AI tools commonly used by the group One is Microsoft Designer, which he pointed out is partially based on Microsoft Designer. “Vulnerabilities in DALL-E 3, and products such as Microsoft Designer that use DALL-E 3, make it easier for people to abuse AI to generate harmful images,” Jones said in the letter. It is written in Sens. Patty Murray and Maria Cantwell, Rep. Adam Smith, and Attorney General Bob Ferguson obtained by GeekWire. He added that “Microsoft was aware of these vulnerabilities and the potential for exploitation.”
Jones wrote that he independently discovered the vulnerability in early December. The letter said he reported the vulnerability to Microsoft and was directed to report the issue to OpenAI, a close Redmond partner whose technology powers products such as Microsoft Designer. . He wrote that he reported it to his OpenAI. “As we continued to investigate the risks associated with this particular vulnerability, we discovered that DALL-E 3 has the ability to generate violent, disturbing, and harmful images,” he wrote. . “Based on my understanding of how the model was trained and the security vulnerabilities I discovered, DALL-E 3 poses a public safety risk and I am confident that OpenAI will address the risks associated with this model. We have come to the conclusion that it should be removed from public use until such time as possible.”
On Dec. 14, he wrote a public post on LinkedIn urging OpenAI's nonprofit board to take DALL-E 3 off the market. According to the letter, he notified Microsoft management about the post and was immediately contacted by his manager, who requested that the post be immediately removed by Microsoft's legal department and that he would not provide an explanation or justification. It was about. He agreed to remove the post on that basis, but never heard back from Microsoft legal representatives, he wrote. “Over the next month, I repeatedly asked for an explanation as to why I was asked to delete the letter,” he wrote. “I also offered to share information that would help fix certain vulnerabilities I discovered and provide ideas for making AI image generation technology more secure. I received no response and no direct contact with me.'' “Artificial intelligence is advancing at an unprecedented pace, and I understand that it will take time to pass legislation to ensure public safety from AI.'' ” he added. “At the same time, we need to hold companies accountable for communicating the safety of their products and known risks to the public. Concerned employees like myself should not be intimidated into silence.” Jones' letter You can read the full text here (PDF).
