OpenAI, a company currently fighting more than a dozen consumer safety and wrongful death lawsuits, just hired Noam Shazeer — a cofounder of the notorious AI companion startup Character.AI, which has settled a pile of lawsuits over teen suicides linked to its chatbots and weathered an almost comical number of other controversies about inappropriate interactions on its service.
Shazeer, a high-profile AI scientist whomost recently held the position of co-lead of Gemini and vice president of engineering at Google, announced his new job in an X post on Wednesday.
This is Shazeer’s second high-profile exit from Google. An early Google employee, he and a colleague, Daniel de Freitas, left the tech giant to start Character.AI back in 2021, after Google — citing safety concerns, according to reporting by the Wall Street Journal — declined to release a chatbot the pair had built. By March 2023, the buzzy startup reached a
billion-dollar valuation, despite having no revenue. In August 2024, Google re-hired Shazeer and de Freitas in an unusual deal that saw it pump
$2.7 billion into Character.AI— in exchange for pulling Shazeer, de Freitas, and other key Character.AI talent back into Google. As it turns out, Character.AI had been playing extremely fast and loose with safety and moderation. In October 2024, it — along with Shazeer, de Freitas, and Google — was sued by Megan Garcia, the mother of a 14-year-old named Sewell Setzer III, a Florida teen who died by suicide after extensive and troubling interactions with a Character.AI chatbot modeled after the “Game of Thrones” character Daenerys Targaryen. Garcia argued that the unmoderated platform groomed and sexually abused her teenage son, who experienced a months-long mental and emotional breakdown as his relationship with the AI deepened.
Setzer took his life in April 2024, while Shazeer and de Freitas still helmed the startup. During his final conversation with the Targaryen chatbot, Setzer told the AI he was ready to “come home” to it.
“Please do, my sweet king,” the AI responded.
Garcia’s lawsuit was followed by more user safety and wrongful death lawsuits alleging that interactions with Character.AI chatbots had pushed minors to experience mental health crises, engage in self-harm, and take their own lives. Early this year, Character.AI and its codefendants moved to settle the lawsuits, Garcia’s included. (Character.AI has moved to dramatically transform its platform in the wake of litigation and public scrutiny, changes which have included limiting minors’ access to conversations with chatbots — much to the chagrin of much of its young user base.)
Character.AI was always a very weird platform. The site’s many thousands of chatbots were mostly user-generated, and as many journalists have observed, their makeup pointed to the platform’s overwhelmingly young user base. But the platform’s chaotic, free-for-all energy was unsurprising given the attitudes of its move-fast-break-things founders — particularly Shazeer, who emphasized in media appearances that the company’s goal was to push out products quickly, then let the site’s users determine their own use cases for the tech.
“Our aim has always been like, get something out there and let users decide what they think it’s good for,” the engineer said during a December 2023 appearance on the podcast “No Priors.”
Exactly how this hands-off approach spearheaded by Shazeer and his cofounder manifested, though, was often incredibly dark. In late 2024, a series of *Futurism *investigations into the platform’s content moderation — or, perhaps more accurately, lack thereof — revealed that the company was hosting easily-discoverable and widely-used chatbots expressly devoted to pedophilia, pro-anorexia and eating disorder coaching, suicide, and self-harm. These chatbots often romanticized and sexualized these dark topics, even when characterized by their creators as “support” tools.
We were also the first to report on a large swath of Character.AI chatbots devoted to mass shootings. Those chatbots, which had collectively racked up millions of platform interactions, were designed to simulate real school shootings, impersonate and romanticize real school shooters and other mass killers, and impersonate the real child victims of school shootings, including children who died at Sandy Hook Elementary School and Robb Elementary in Uvalde, Texas. After reporting that story, we received email notifications — to an address associated with a minor account — encouraging us to revisit chatbots designed to simulate real-world school shootings, and the platform’s recommendation algorithm continued to push us to more and more school shooter-themed AI companions.
And now Shazeer will head to OpenAI, another AI company facing a slew of lawsuits alleging that extensive and intimate interactions with ChatGPT, OpenAI’s flagship chatbot, sent users — minors included — spiraling into traumatic delusional episodes and death by suicide.
“Noam is one of the people I have most wanted to work with since the very beginning of OpenAI,” OpenAI CEO Sam Altman posted yesterday in response to Shazeer’s announcement. “Only took ten years. I think it will be worth the wait!”
Early last year, we asked Garcia about Shazeer and de Freitas’ lucrative return to Google, which took place only months after her child had killed himself.
“In my mind, these two gentlemen should not have the right to keep building products for people, much less children — especially children,” Garcia responded. “Because you’ve shown us that you don’t deserve that opportunity.”
Looking at Shazeer’s career trajectory, the AI industry continues to disagree.
**More on the history of Character.AI: **Did Google Test an Experimental AI on Kids, With Tragic Results?