A Microsoft engineer tells WA's AG he found ways to exploit DALL-E 3 to make explicit images, reported the issue, and was then told to take down a public post
Context & Ripple Effects
The report sits at the intersection of generative-AI safety testing and corporate control of vulnerability disclosures. Its central concern was reinforced soon after, when the same engineer escalated concerns about Copilot Designer’s violent and sexual image outputs to the FTC and Microsoft’s board.
Later coverage of Microsoft’s response to a researcher’s public disclosures shows that disputes over how safety findings are communicated can become a governance issue, not just a product-quality issue.
First-order effects
- Microsoft faces scrutiny from Washington’s AG over an alleged route around DALL-E 3’s safeguards for explicit-image generation and over its handling of the engineer’s public disclosure.
- The engineer’s reported instruction to remove a public post narrows the immediate channel for external users and researchers to evaluate the claimed weakness.
Second-order effects
- A regulator-facing complaint raises the cost of relying solely on internal reporting paths: Microsoft must demonstrate that reported generative-AI safety issues are investigated and remediated credibly.
- Researchers may weigh the risks of public disclosure more heavily, particularly amid later backlash over Microsoft’s threatening posture toward a public bug disclosure, potentially reducing independent visibility into model-safety failures.
Third-order effects
- If similar cases recur, generative-AI safeguards will be judged not only by whether they block harmful outputs, but by whether vendors provide trusted, independent pathways to surface failures.
- The pattern points toward an expanding conflict over researcher disclosures as AI systems become consumer-facing: product safety, platform moderation, and whistleblower protections increasingly overlap.
The trend: Generative-AI governance is shifting from promises of guardrails toward scrutiny of how companies discover, disclose, and remediate failures in those guardrails.