Microsoft’s legal department allegedly silenced an engineer who raised concerns about DALL-E 3. DALL-E 3 can reportedly produce disturbing images.
A Microsoft executive claims that DALL-E 3, the text-to-image AI developed by OpenAI, has vulnerabilities that could allow users to create violent or obscene images (similar to those recently targeted at Taylor Swift). According to new reports, the company’s legal team blocked attempts by Microsoft engineering leader Shane Jones to alert the public about the vulnerability.
“I have concluded that DALL-E 3 poses a public safety risk and should be removed from public use until OpenAI addresses the risks associated with this model,” Shane Jones wrote in a letter to US senators and the Attorney General of Washington state.
Jones claims to have discovered a vulnerability in early December that allowed DALL-E 3 to bypass security barriers. He says he reported the problem to his superiors at Microsoft, who instructed him to “personally report the issue directly to OpenAI.” After doing so, he claims to have learned that the vulnerability could allow the creation of “violent and disturbing malicious images”.
No vulnerability according to OpenAI and Microsoft
Jones then shared a post on LinkedIn to inform the public and reported it to Microsoft. Microsoft reportedly demanded that the post be removed immediately. Jones says he was told that he would receive a detailed explanation from Microsoft’s legal team, but that never happened.
OpenAI, on the other hand, stated that no such vulnerability was found in DALL-E 3 after investigations. It also states that they have made serious trainings and limitations in the model to prevent sexual or violent requests. Similarly, Microsoft also stated that there is no such vulnerability. However, some say that this vulnerability or another vulnerability may be related to the pornographic deepfakes of Taylor Swift, which are currently in the news. However, there is no clear statement on this issue.