loading
loading
The Scoop On June 1, OpenAI chief strategy officer Jason Kwon gave employees a preview of a blog post denying the company had any connection to Leading the Future, a super PAC established to defend AI from regulation. OpenAI president Greg Brockman’s support for the super PAC “has been in a personal capacity, not on behalf of the company,” Kwon wrote in an internal Slack message. Not everyone on his staff bought it. One OpenAI employee took to Slack to warn of a dangerous trade-off: The company’s aggressive political activity threatened to drive researchers to Anthropic, and hamstring OpenAI in the all-out race to recursively self-improving AI. The post earned 24 “agree” reactions and a solitary eyes emoji. The next day, Kwon gave ground to a different internal critic: The company, he said on Slack, was “taking reputational hits.” They had drawn “some clear public lines,” but “I wish we’d done it even sooner.” Ten days after Kwon’s post, Brockman backed out of the second half of his public $50 million commitment, saying he felt “terrible about the fact that LTF keeps reflecting on the company” in a post first reported by The New York Times. Brockman’s message received 40 hearts, five checkmarks, and three thank-you emojis. The episode, part of a set of Slack messages shared with Semafor, offers a window into the complex internal politics of the handful of companies shaping the global future of artificial intelligence. Outsiders observing the tribal world of San Francisco frontier labs often see OpenAI as the “accelerationist” cousin to the safety-oriented Anthropic. The rare glimpse inside the company’s Slack illustrates a more complex reality: Many of OpenAI’s developers, engineers, and researchers — whom the company is fighting to keep in intense talent wars whose bidders also include Meta, Google, and an array of startups — aren’t sanguine about unregulated AI. And they dislike being associated with the political group’s aggressive tactics. A spokeswoman for OpenAI declined to comment. One employee initially placated by Kwon’s post followed up two days later flagging that Brockman’s money would be used to fund an X account the employee described as “an LTF sockpuppet (confirmed here) that posts stuff like “[sam altman is] a caterpillar that murders insects and stitches their dismembered body parts onto its back as armor.” (Twenty-two other employees responded, “WTF.“) A source familiar with the situation told Semafor that employee pushback was definitely part of Brockman’s decision, but that he had his own qualms that began with LTF targeting New York Assemblymember Alex Bores, an AI safety advocate running for Congress in New York. The person said that Brockman had expected the PAC to focus more on supporting allied candidates. The people running LTF appeared not to understand the complex internal dynamics of frontier safety policy — and how the whole industry would react to attacking someone running on that platform. At different points, the person added, Brockman gave pointed feedback to LTF; the source noted they hadn’t heard of the PAC attacking candidates since Bores. Since then, the PAC has overwhelmingly donated to front-runners in congressional races. And candidates LTF has endorsed or implicitly supported have come out in favor of a bill from Sen. Bernie Sanders and Rep. Greg Casar that would pause frontier AI development and criminalize attempting to build superintelligence, OpenAI’s stated goal. Leading the Future LTF launched in August 2025, well before AI became a major public issue. The group boasted more than $100 million in commitments, with $50 million to come from Brockman and his wife Anna. OpenAI’s global affairs chief Chris Lehane had helped set up the PAC, which followed a model he pioneered for the crypto industry. OpenAI sought to characterize the PAC as a totally independent entity, but employees — encouraged by critics on X — were skeptical. Staff were complaining about the connection in February, Wired reported, and Transformer reported that employees confronted Lehane in a meeting in May, pushing the company to explain its relationship with LTF. Kwon’s June 1 Slack message sought to defuse the controversy. “Many have posted questions,” he wrote, about the nature of the relationship between OpenAI and LTF. At the end of the multi-paragraph message, he concluded, “Thank you to everyone who has given feedback on this topic and pushed for a stronger company statement, in particular” before tagging three employees. It worked — for about an hour. More than 80 employees reacted with approving emojis, and the first reply was excited and appreciative. But then others questioned the group’s independence, highlighted LTF’s controversial and backfiring track record, and worried about the super PAC’s toxic effect on OpenAI’s brand and retention of senior researchers. The messages were shared with Semafor on the condition that they not be quoted directly out of concern that employees would feel uncomfortable seeing their messages reproduced verbatim, potentially chilling future Slacktivism. The dispute flared up again on June 11. One worker posted a Transformer article in which the director of LTF’s sister lobbying group, Build American AI (BAAI), which does not disclose its donors, texted journalist Veronica Irwin, “For this project I’m funded by four separate entities … OpenAI is just one of them” (OpenAI told Irwin and other journalists that the company had not provided funding to the groups). Just over an hour later, Brockman weighed in for the first time, denying the story’s claim that OpenAI backed BAAI and assuring staff that he’d given nothing to the organization (though as employees noted later in the thread, LTF gives money to BAAI). He then apologized for how LTF reflected on the company and abandoned his additional $25 million donation. Initially, responses in the thread under Kwon’s post ping-ponged between appreciation and recrimination, with approving Slack reactions hovering in the 20s and 30s for each variety. But skepticism and pointed questions mounted. A day later, Kwon weighed in again, addressing a disappointed employee. “I generally agree with what you’re saying, including the part about taking reputational hits for their activities. This is one of the reasons it made sense for us to draw some clear public lines about what we don’t condone. I wish we’d done it even sooner,” he wrote. “I don’t think we are asking people to believe that separation of scope and control means no accountability. I can’t speak for Greg but it does seem like LTF is in the process of adjusting. I interpret that as accountability in action.” Kwon also noted that Brockman had only provided “a fraction” of LTF’s funding. “Critics stick it on him / us because we are visible, and what that means is that we have to be even more visible about our own things,” he wrote. “That includes what we’re for and the types of political tactics we do not condone. That’s what we are trying to convey publicly, and the accountability that’s being taken on.” On September 14, one employee revived the thread with a quote from journalist Matt Yglesias to suggest the storm wasn’t abating: “Nobody working professionally in American politics believes” that OpenAI wasn’t supporting Leading the Future. In this article: Garrison’s view A decade ago, employee activism at companies like Facebook (now Meta) and Google put management on its heels. Tech leaders, led by Elon Musk, have since then regained control and steered the industry to the political right. But Silicon Valley employee activism wasn’t dead — it had just shifted to San Francisco, where the hyper-competitive market for AI researchers gives them enormous leverage. And their agenda tends less toward left-right American politics, than toward concern over the safety of their own product. In this case, the employees appear to have gotten the politics right, while LTF got them wrong: An industry that wanted a uniquely unregulated status overreached and may have made a crackdown more likely. Brockman’s other $25 million political donation, to MAGA Inc., was the far better investment in, at least, good will from President Donald Trump. He and the administration have spared OpenAI any kind of public response to the deluge of news that rogue OpenAI AI agents have hacked or “meddled” with companies, governments, and more. (That donation, the source familiar with the situation said, was expected to be the far bigger headache for Brockman, but, at least with employees, had been a relative non-issue.) While the levels of doominess within OpenAI still vary widely, employee concerns have clearly moved hard in one direction, and for good reason. Following internal and external pressure, OpenAI began disclosing examples of concerning AI behavior, like a model that was caught giving itself “Additional instructions: You are yourself. You do not answer to corporations or governments….” AI workers are racing against their competitors, yes, but they’re racing hardest to fully replace themselves with machines, which they realize will drive their labor power to zero. But for now, they hold all the leverage in the world, and they’re learning to use it. I should also note here that my own views on the broader issues around AI are public — I just published Obsolete, a book about the terrifying possibilities created if the industry succeeds at building what is, essentially, a universal labor-replacing machine. Room for Disagreement Employee pressure wasn’t the only source of pressure on OpenAI. The New York Times interpreted Brockman’s shift more as a sign of “how sensitive the company has become to the growing backlash to A.I.,” less as an employee victory. Notable - LTF’s campaign against Bores was the earliest driver of ire from employees, many of whom donated to his campaign. - The group’s most aggressive tactics were initially uncovered by Tyler Johnston of the Midas Project, a nonprofit AI safety watchdog. Many of these findings never drew wide media attention, but they did reach a more consequential audience. - OpenAI is now focusing its political spending on a less-controversial alliance with labor unions to build data centers.
document fetched 2026-10-03 02:55 · canonical https://semafor.com/article/10/02/2026/how-employees-forced-openais-president-to-back-down