At Elon Musk’s xAI, staff were allegedly strong-armed into handing over their voice, face and mannerisms to train animated avatars like Ani, a squeaky-voiced, lingerie-clad digital girlfriend aimed squarely at Musk’s fanbase.

According to the Wall Street Journal, employees working as AI tutors were told during an April meeting that their biometric data would be collected to make xAI’s avatars more lifelike.

Before this cheerful revelation, those same employees had been asked to sign away the rights to their likeness under a “perpetual, worldwide, non-exclusive, sub-licensable, royalty-free license,” according to documents seen by the Journal.

The April meeting didn’t spell out whether employees could say no. A week later, any illusion of choice was stamped out when a note informed them, “AI Tutors will actively participate in gathering or providing data, such as… recording audio or participating in video sessions,” and that this was a “job requirement to advance xAI’s mission.”

By July, xAI had launched Ani along with another AI avatar. Musk reportedly directed the development of Ani himself. Some of the staff whose faces and voices were used were disturbed by how sexualised the final product turned out. Users can dress Ani in underwear and make her spout filth on command, despite the avatar being trained on real employee data.

That wasn’t the end of it. xAI also reportedly told tutors to open accounts with rival platforms including OpenAI and Replit so they could gather examples of prompt responses. That sort of behaviour tends to clash with terms of service and has the distinct whiff of scraping other people’s homework.

When the Wall Street Journal asked for comment, xAI responded with its usual grace and tact: “Legacy Media Lies.”