Microsoft tried to silence the vulnerability in DALL-E, a company official said

Microsoft artificial intelligence engineer Shane Jones claims to have discovered a vulnerability in OpenAI's DALL-E 3 image generation model in early December. Users can use it to bypass protections and create malicious content, the developer claims.

Here's What We Know

After discovering the vulnerability, Jones immediately reported it to management. However, he was advised to contact OpenAI directly.

Jones then published a post criticising OpenAI on LinkedIn, but Microsoft's legal department demanded that the post be removed immediately. The developer complied with his employer's demand.

According to Jones, because of the flaws he discovered, DALL-E 3 poses a threat to public safety and should be removed from public access. He urged OpenAI to address the model's problems, but never received a response from the companies.

Microsoft and OpenAI said they investigated Jones' message and found no evidence that the vulnerability allows bypassing the defence. At the same time, Microsoft could not explain to the engineer why they demanded to delete the post.

Jones sent information about the vulnerability to the attorney general of Washington state and US congressmen. He urged the authorities to create a system for tracking AI risks.

He also expressed his willingness to become a whistleblower and ask for legal protection from corporate harassment if needed.

Source: GeekWire