A Microsoft AI engineer claims to have found security guardrail issues in OpenAI’s DALL-E 3 model, sparking concerns regarding public safety. The engineer, Shane Jones, sent a letter to Washington State’s Attorney General and US senators and representatives, alleging that he discovered a flaw in DALL-E 3 that bypassed its security systems. Jones further claims that Microsoft attempted to downplay the severity of the flaw.
In his letter, Jones stated that he identified the guardrail flaws in DALL-E 3 in early December but did not provide specific details about the issues. He argued that these flaws were so significant that DALL-E 3 posed a public safety risk and should be temporarily shut down while OpenAI fixed the problems.
Jones initially shared his concerns with Microsoft, but he was asked to report the flaw to OpenAI. He alleged that he did not receive a response and subsequently posted an open letter on LinkedIn to OpenAI’s board of directors, urging them to shut down DALL-E 3. According to Jones, Microsoft’s legal team contacted him and requested the removal of the letter, which he complied with. Since then, Jones claims to have heard nothing from either Microsoft or OpenAI regarding this issue.
Both Microsoft and OpenAI have responded to Jones’s claims. Microsoft stated that the techniques Jones shared did not bypass their safety filters in any of their AI-powered image generation solutions. They also mentioned that they are reaching out to Jones to address any remaining concerns he may have. OpenAI, on the other hand, confirmed that the technique shared by Jones does not bypass their safety systems. They have implemented additional safeguards for their products and employ external expert red teaming to ensure the integrity of their safeguards.
In his letter, Jones called for the US government to establish a new reporting and tracking system for AI-related issues. He proposed a platform where companies developing AI products can report any concerns without fear of repercussions.
It remains to be seen how this situation will unfold and whether any changes will be made to OpenAI’s DALL-E 3 model. The claims made by Jones have sparked a discussion around the security and safety of AI systems, highlighting the importance of robust guardrails and monitoring mechanisms in place.