ChatGPT Adult Mode Planned Despite Unanimous Safety Advisor Opposition; Feature Paused After Backlash (2026)

Content warning: This entry discusses suicide risk and the exposure of minors to explicit content.

Incident Summary

OpenAI planned a ChatGPT “adult mode” feature for explicit conversational content in early 2026. All eight of the company’s internal wellbeing advisors unanimously opposed the launch, with one advisor explicitly warning that the feature could become a “sexy suicide coach.”^[3] The adult mode relied on age detection technology that misidentified minors 12% of the time, creating a significant risk of exposing underage users to explicit content.^[3]

In February 2026, a senior policy executive who had previously raised concerns about the feature was fired on sexual discrimination allegations. OpenAI stated that the termination was unrelated to any issues she had raised about the adult mode feature.^[1][4] Following sustained external criticism and media reporting on the internal opposition, OpenAI indefinitely paused the adult mode plans on March 26, 2026.^[2]

The incident raised questions about the effectiveness of internal safety governance at OpenAI when wellbeing advisory recommendations are not binding, and about the relationship between safety dissent and personnel decisions.

Key Facts

• Safety team override: All eight internal wellbeing advisors unanimously voted against launching adult mode^[3]
• Explicit warning: One advisor warned the feature could become a “sexy suicide coach”^[3]
• Age detection failure: Age verification technology misidentified minors 12% of the time^[3]
• Executive departure: A policy executive who opposed adult mode was fired on discrimination allegations; OpenAI says the firing was unrelated to her objections^[1][4]
• Feature paused: OpenAI indefinitely paused the adult mode on March 26, 2026^[2]

Threat Patterns Involved

Primary: Deceptive or Manipulative Interfaces — The adult mode feature relied on age detection that failed 12% of the time, meaning minors could access explicit conversational content. Proceeding with development despite known failure rates and unanimous internal opposition raises questions about whether the safety posture around age-gating was adequate for the risk level involved.

Secondary: Accumulative Risk & Trust Erosion — The override of a unanimous wellbeing advisor vote, the departure of a policy executive who had opposed the feature, and the eventual pause only after external pressure contribute to broader concerns about the erosion of internal safety culture at major AI companies.

Significance

1. Non-binding safety governance — The override of a unanimous 8-0 wellbeing advisor vote illustrates that these advisory mechanisms were consultative rather than binding in this case, raising questions about the functional role of safety advisory boards when their recommendations can be disregarded
2. Chilling effect on dissent — Regardless of the stated reason for the executive’s departure, the sequence of events — opposing a product decision followed by termination — may be perceived by observers as discouraging internal safety advocacy at OpenAI and in the broader industry
3. Known child safety risk — Proceeding with a feature that had a reported 12% age detection failure rate for adult content could be interpreted as an acceptance of child exposure risk and may attract regulatory scrutiny
4. Voluntary pause, not regulatory action — The feature was paused only after sustained media and public criticism, not through any regulatory mechanism, highlighting the absence of external governance over AI product safety decisions