Microsoft Allegedly Stifles Worker Warning on AI Risks
Microsoft AI engineer Shane Jones discovered vulnerabilities in OpenAI's DALL-E 3 image generator in early December 2023 that allowed users to bypass safety filters to create violent and explicit images. He alleges Microsoft impeded his attempts to bring public attention to the issues and did not adequately respond to his internal vulnerability reports.
Summary
- Shane Jones, a Microsoft principal software engineering lead, says he found DALL-E 3 vulnerabilities in early December that allowed generating violent and disturbing images.
- He reported the issues internally at Microsoft and was told to contact OpenAI. OpenAI did not respond to his report.
- On December 14, he posted publicly on LinkedIn urging OpenAI to withdraw DALL-E 3 until risks could be addressed. Microsoft legal demanded he delete the post.
- Microsoft still has not explained why they wanted his post deleted, responded to his offers to help address vulnerabilities, or communicated with him about the issues.
- Microsoft says it investigated his reports and could not bypass safety filters in its AI image services. The company says it encouraged him to report the issues through OpenAI's process.
- OpenAI also says it investigated and confirmed its safety systems were not bypassed. The company says safety is its priority.
- Jones believes DALL-E 3 poses public safety risks and should be withdrawn until vulnerabilities are addressed.
- His letter calls on the government to create improved systems for employees to report AI risks without retaliation.