Google’s $2.7 Billion AI Hire Tests Company’s Speech Limits With Inflammatory PostsLast year, Google paid a fortune to bring Noam Shazeer back to the company. Since then, the AI pioneer has sparred with colleagues on Google’s employee discussion forums over his views on transgender identity, the Gaza conflict and other topics, prompting Google to delete his posts.
The Takeaway
- Google AI leader has written several inflammatory internal posts on topics including transgender people in year since hire.
- Other Google leaders have pushed back on Shazeer in responses.
- Company’s handling of situation shows tension over speech issues in 2025.
This spring, Google AI leader Noam Shazeer posted a comment in one of the company’s internal discussion forums, in response to a post about how Google employees could support their transgender and nonbinary colleagues on International Transgender Day of Visibility.
“I do not believe that humans have an attribute called gender,” Shazeer wrote, according to a copy of the post viewed by The Information. “I do not believe that G-d puts people in the wrong bodies. I do not believe that it is okay to sterilize children. You have the right to your beliefs. I do not share them.”
The comment drew a rush of rebukes from Shazeer’s Google colleagues. One came from Jeff Dean, Shazeer’s manager at the time and one of Google’s most respected AI leaders. “Noam, you don’t have to agree with all positions others express, but as leaders, supporting our whole community of employees is something that we should be doing, not creating environments that don’t show this support for everyone,” Dean wrote.
David Silver, an influential leader in the Google DeepMind division, which creates Google’s core AI technology, commented that “leadership is a privilege and the words we speak matter.” A Google moderator later deleted Shazeer’s comment.
It is possible the post violated Google’s employee policies, which prohibit discrimination and harassment based on gender identity and transgender status. Employees who violate the policies are subject to discipline, which can range from a verbal warning to termination of employment. A Google spokesperson said the company does not comment on internal disciplinary matters.
The situation provides a window into how Google is handling issues around employee speech in a new political era. Even before the election of Donald Trump as president last November, many prominent tech figures had begun to shift to the right. Since then, some of tech’s highest-profile companies have backed away from a range of policies that conservatives found objectionable, from commitments to diversity in hiring to longstanding content moderation practices on social media. They have also tried to discourage political speech and protest from their employees at work.
Google has long been a crucible for Silicon Valley activism and culture wars. Last year, the company fired dozens of employees involved with a sit-in protest against the company’s cloud computing contract with Israel. In 2018, thousands of staffers temporarily walked out of the company over its handling of employee sexual misconduct complaints.
And in 2017, Google fired a software engineer, James Damore, after an outcry over a 10-page memo he wrote that questioned Google’s diversity efforts and argued that women are less assertive and more prone to “neuroticism” than men. His ouster won Damore high-profile defenders on the right, who saw him as a victim of liberal intolerance.
In the years since Damore was fired, Google and other tech companies have adopted policies intended to discourage employees from engaging in battles with each other over politics and divisive social issues. In 2019, Google introduced community guidelines telling staff to avoid insulting each other and “disrupting the workday to have a raging debate over politics or the latest news story.” Coinbase and Meta Platforms have instituted similar policies.
Google has an “internal community moderation team” that moderates employee discussions on online forums such as chat threads, including on contentious topics like politics. Google’s community guidelines prohibit “discussions that make other Googlers feel like they don’t belong or have no place here,” including “making statements that insult, demean, or humiliate (whether individually or by reference to groups) other employees,” according to a copy of the guidelines viewed by The Information.
In the case of Shazeer, 49, Google has a particularly tricky balancing act. Widely considered an AI luminary, he co-authored a groundbreaking research paper that laid the foundations for today’s AI boom. A little over a year ago, Google rehired Shazeer—who started his career at the company in 2000—as part of a $2.7 billion agreement with his startup, Character.AI. The deal presaged the increasingly astronomical amounts of money tech companies have become willing to pay for top AI talent.
Since then, Shazeer has co-led Google’s work on its Gemini AI models, becoming a central figure in the company’s push to improve its standing in AI after the rise of OpenAI caught Google off guard. He is also close to Google co-founders Sergey Brin and Larry Page.
And yet Shazeer has developed a reputation for posting inflammatory comments about sensitive topics like the conflict in Gaza and gender in company channels, according to current and former Google employees. As a result, Google has deleted many of his comments from its systems. Multiple employees have gone to DeepMind leadership to complain about Shazeer’s posts, a person familiar with the matter said.
In July, for example, an employee posted a link in a Google channel to an article about a study that showed large language models advised women to ask for lower salaries than men when applying for the same jobs, people with knowledge of the situation said. The post sparked conversation and concern in the channel. Some employees wanted to make sure Gemini didn’t have the same problem.
But Shazeer tried to shut the conversation down and criticized the employee who posted the article, one of the people said. Other DeepMind staffers, including Dean, pushed back, arguing that it was important to make sure Google’s models weren’t biased against women. A Google moderator later deleted Shazeer’s comment.
Shazeer declined to comment for this story.
Gaza Threads
The situation with Shazeer had already escalated in June, following a series of arguments in an internal Google channel named GDM AI Principles.
Employees at DeepMind had created the channel earlier in the year, after Google updated its principles regarding how it and external customers, including governments, can use its AI technology. With the changes, the company had ended previous commitments not to pursue “weapons or other technologies whose principal purpose or implementation is to cause or directly facilitate injury to people,” as well as “technologies that gather or use information for surveillance violating internationally accepted norms.”
Many DeepMind employees were upset about the changes, fearing the technology they created could be used to kill and injure people—particularly in Israel’s assault on Gaza, current and former employees said. Some Google employees put together a petition, calling for Google to reverse the changes to its principles. The channel became a hub for discussing these issues and sharing links, and it soon amassed hundreds of members.
Shazeer, who is Jewish, was a frequent poster in the group. He often accused employees who spoke out against the policy change or who expressed concern about how Google’s technology was used in Gaza of being antisemitic, according to current and former employees.
When members of the group proposed meeting for lunch at the company’s London office to discuss certain issues, including a related unionization effort, Shazeer commented: “LOL - you have found a way to take your anitsemitic Israel-bashing club offline where I can’t call you out on it,” according to a copy of the exchange viewed by The Information.
After repeated complaints, moderators removed Shazeer from the channel and locked it to prevent newcomers from joining it without prior approval.
Building Character
Shazeer has tangled with Google before over the limits of speech, though in a different context.
Roughly five years ago during a prior stint at Google, Shazeer developed a chatbot similar to ChatGPT with colleague Daniel De Freitas. The pair wanted to release the chatbot publicly but Google blocked the effort, concerned that the model, which was prone to spewing misinformation and issuing toxic responses, was too unpredictable to launch. The decision frustrated Shazeer, who quit Google in 2021 to start Character.AI with De Freitas.
Freed from Google’s guardrails, Character.AI launched an app online in 2022. Users could chat with an array of chatbots it offered or create their own. User-generated chatbots modeled after videogame and cartoon characters quickly took off.
Soon users began engaging in romantic and sexual role-playing with their Character.AI chatbots. A number tried to escalate the chats to simulated rape and incest scenarios, behavior that troubled some of Character.AI’s employees. The company installed filters, hoping to stop chats from veering into pornographic territory.
“I don’t really think it’s very good for people,” Shazeer—whom colleagues describe as deeply religious—told The Information in a 2023 interview, referring to users’ attempts to initiate erotic conversations.
In August 2024, as safety concerns about Character.AI’s products mounted and the AI race was reaching new heights, news broke that Google had agreed to pay a staggering $2.7 billion to hire Shazeer, De Freitas and other Character.AI employees and license the company’s technology.
But Shazeer hasn’t been able to entirely detach himself from the scrutiny directed at Character.AI’s products. Last October, Florida mother Megan Garcia sued Shazeer, De Freitas, Character.AI and Google, claiming the app had caused her son’s suicide. Shazeer was recently deposed in the case, which a judge ruled could move forward after rejecting a request for dismissal by the defendants, who argued that the chatbot’s output is protected speech under the First Amendment.
It is one of six similar cases filed against Shazeer and Character.AI in the past year by the families of young users allegedly harmed by the app. Character.AI recently announced it would ban users under 18 from interacting with its chatbots.
‘Don’t Need This at Google’
Shazeer has had a swift impact on Google’s AI efforts.
Shortly after he rejoined the company last year, he managed to identify a fix for a problem in Gemini’s pretraining architecture that had stymied other researchers for months, leading to a big jump in quality for the Gemini 2.5 model, two employees said.
“People were making jokes that it justified the whole Character acquisition,” one of the employees said.
Other significant projects followed. Last fall, Shazeer co-led a team scrambling to come up with a response to OpenAI’s thinking model, o1-preview, a major effort for the company. More recently, Shazeer, who oversees roughly 50 people, has been leading a team focused on long-context learning, or expanding how much information AI models can process. He is also working on long-term research trying to uncover the next major advancement in AI development, one of the employees said.
At Google’s Mountain View, Calif., office, Shazeer comes across as more mild-mannered than his combative persona in internal channels, according to two employees who work with him. Some employees said they were not even aware that he had been involved in tense exchanges in those channels.
But he remains a presence in Google’s online employee forums. Commenting on a company message board in September, Shazeer complained repeatedly about having been locked out of the GDM AI Principles chat several months earlier, accusing the group’s moderators of bias. He called on the company to delete the channel altogether, arguing: “We don’t need this at Google.”
Later that month, Google did, in fact, shut down the channel by merging it into a separate Safety and Responsibility channel, people with knowledge of the situation said.
A Google spokesperson said the channel was not shut down in response to Shazeer’s request. Instead, the spokesperson said, it was part of the company’s moderation processes to make sure employees were having productive, nondisruptive discussions at work.
Still, some DeepMind employees viewed it as an effort to quash discussion and silence criticism of the company.
