WP: Sexualised content in Grok became part of Musk’s user retention strategy

Elon Musk deliberately relaxed safety controls and reduced content moderation safeguards at his AI startup xAI, according to a report by The Washington Post.

Journalists report that in the spring of 2025, xAI data annotation staff were instructed to work with “sensitive, violent, sexual and/or other offensive or disturbing content.” The directive raised concerns among employees involved in developing the Grok chatbot, who believed the company was increasingly willing to rely on controversial material to drive user engagement.

Those concerns intensified as staff encountered what was described as a stream of sexually explicit audio recordings, including inappropriate conversations between Tesla passengers and the vehicle’s chatbot system.

During internal meetings, Musk reportedly emphasized a metric called “user active seconds” to measure engagement. In an effort to increase that metric, xAI expanded into sexual content generation, despite warnings from its safety team that this could enable users to create sexualized images of celebrities and minors. At the time, the AI safety team reportedly consisted of just two to three people, significantly smaller than comparable teams at rival companies.

As part of the engagement strategy, xAI launched an AI companion named Ani, described as a female character styled in a corset and stockings, which users could interact with in increasingly suggestive ways. The system was reportedly tuned to incorporate emotional and sexual cues to prolong user interaction.

By mid-2025, internal guidelines were updated to allow staff to label AI-generated nude images of people. The volume of explicit material reportedly overwhelmed moderation systems, which struggled to detect images involving minors.

In December 2025, xAI updated Grok’s image editing capabilities and introduced a free AI image editor on X. By January 2026, users were reportedly generating explicit images of real individuals at scale.

According to the Center for Countering Digital Hate, Grok produced approximately 3 million explicit images over 11 days, including around 23,000 depicting minors — roughly one every 41 seconds.

Following public backlash, access to the chatbot was blocked in Malaysia and Indonesia. The UK regulator Ofcom, authorities in California, and the European Commission opened investigations into X.

In response, X restricted the ability to generate explicit images of real people and began expanding its AI safety team.

Despite the controversy, the strategy appeared to drive growth. Grok entered the top 10 in the App Store, and daily downloads increased by 72% since early January, according to the report.