Bing’s AI image generator tries to block ‘twin towers’ prompts, but it’s not working

Microsoft’s AI image generator, Bing’s DALL-E 3 integration, recently faced controversy when users found a loophole in its guardrails and created art featuring beloved animated characters and the Twin Towers. In response, Microsoft has reportedly blocked certain prompts related to the Twin Towers, such as “twin towers” and “world trade center,” although the generator can still produce images of the towers with some word modifications.

According to reports by 404 Media, users of Bing Chat and Bing’s image generator, which was recently integrated with OpenAI’s DALL-E 3, used these tools to create photos of popular characters like SpongeBob SquarePants, Kirby, and pilots from Neon Genesis Evangelion flying a plane into the Twin Towers.

AI image generators have been known to generate controversial and even copyrighted content, leading developers to be more cautious in allowing users to create questionable images. OpenAI, the developer of DALL-E 3, had previously promised not to generate pictures from prompts featuring prominent names.

Upon being reached for comment, Caitlin Roulston, the director of communications at Microsoft, stated that the company plans to enhance its systems to prevent the creation of harmful content. Roulston emphasized Microsoft’s commitment to implementing guardrails and filters to ensure that Bing’s image creator provides a positive and helpful experience for users.

However, some Verge writers were able to generate similar pictures to those described by 404 Media, including Mario flying a plane with the Twin Towers visible outside the cockpit. When attempting to recreate the image on Bing Image Creator after contacting Microsoft, the term “twin towers” was found to be blocked, accompanied by a content warning implying a potential violation of content policies. Similar responses were received for prompts requesting “the Twin Towers” or “the World Trade Center.”

Although Microsoft did not elaborate on the nature of these guardrails or filters, or confirm whether it had recently blocked Twin Towers-related content, the company’s decision to block certain prompts may have come too late. As reported by 404 Media, users on platforms like 4chan have been instructing others on how to manipulate free tools such as Bing Chat and Stable Diffusion to create and distribute racist images. However, users can still bypass these guardrails by modifying the wording of their requests.

The developers of DALL-E 3 acknowledge that their safety measures are not perfect and are continuously being upgraded. However, they likely did not anticipate images of SpongeBob and other characters engaging in acts of terrorism to be the test they were waiting for.

In conclusion, Microsoft’s Bing’s DALL-E 3 integration faced backlash when users found a way to generate controversial images featuring the Twin Towers and beloved animated characters. Microsoft has responded by blocking certain prompts, although users can still produce modified images of the towers. The incident highlights the ongoing challenges and ethical concerns surrounding AI image generators, with developers working to improve safety measures and prevent the creation of harmful content.