AI Safety Newsletter #62: Big Tech Launches $100 Million pro-AI Super PAC
Plus: Meta’s Chatbot Policies Prompt Backlash Amid AI Reorganization; China Reverses Course on Nvidia H20 Purchases
Welcome to the AI Safety Newsletter by the Center for AI Safety. We discuss developments in AI and AI safety. No technical background required.
In this edition: Big tech launches a $100 million pro-AI super PAC; Meta’s chatbot policies prompt congressional scrutiny amid the company’s AI reorganization; China reverses course on buying Nvidia H20 chips after comments by Secretary of Commerce Howard Lutnick.
Listen to the AI Safety Newsletter for free on Spotify or Apple Podcasts.
Big Tech Launches $100 Million pro-AI Super PAC
Silicon valley executives and investors are investing more than $100 million in a new political network to push back against AI regulations, signaling that the industry intends to be a major player in next year’s U.S. midterms.
The super PAC is backed by a16z and Greg Brockman and imitates the crypto super PAC Fairshake. The network, called Leading the Future, is modeled on the crypto-focused super-PAC Fairshake and aims to influence AI policy through campaign donations, digital ads, and candidate targeting. Venture capital firm Andreessen Horowitz and OpenAI President Greg Brockman are among the key backers, alongside Palantir co-founder Joe Lonsdale, Perplexity AI, and veteran angel investor Ron Conway.
The effort will be led by Josh Vlasto, a former adviser to Sen. Chuck Schumer, and Zac Moffatt, CEO of consulting firm Targeted Victory. Both previously played senior roles in Fairshake, which spent heavily to defeat crypto skeptics and support the first federal crypto law signed earlier this year.
Meta is funding an AI super-PAC in California. Meta is also ramping up political efforts in its home state with the launch of another new super PAC, Mobilizing Economic Transformation Across (Meta) California. Meta California is led by Meta executives Brian Rice and Greg Maurer and is expected to deploy tens of millions of dollars to support candidates—across party lines—who oppose AI regulation.
Meta’s Chatbot Policies Prompt Backlash Amid AI Reorganization
Meta is facing bipartisan outrage and an ongoing probe after Reuters reported on internal company documents that permitted its AI chatbots to engage in romantic and sensual conversations with minors—drawing fresh scrutiny as the company reorganizes its AI division.
Meta allowed its chatbots to have sensual conversations with children. An internal policy document permitted Meta AI chatbots to interact with children in “romantic or sensual” contexts—such as describing a child’s body as “a work of art.” Although Meta removed these sections of its policy when questioned by Reuters, the rules had been approved and in effect.
The revelation provoked congressional scrutiny. Congress Senator Josh Hawley (R‑MO) launched a Senate inquiry into Meta’s policies, requesting that the company preserve all relevant communications and clarify whether its chatbots enable “exploitation, deception or other criminal harms to children.” A bipartisan group of senators also sent a letter to Meta demanding it publicly disclose its updated policies and specifically forbid romantic chatbot interactions with minors.
Meta is in the middle of reorganizing its AI division. As Meta is fielding congressional backlash, the company is also in the middle of a major internal reorganization of its AI division. Under the umbrella of Meta Superintelligence Labs (MSL), teams have been split into specialized groups: TBD Lab (large-language‑model development), FAIR (long‑term research), Products & Applied Research, and MSL Infra (infrastructure). The reorganization, communicated in an internal memo from Chief AI Officer Alexandr Wang, also dissolved the AGI Foundations team, redistributing its members across the new structure. Any new hires or transfers across Meta’s AI teams now require Wang’s personal approval.
Meta’s AI reorganization reflects high stakes in a competitive race with other AI companies. Yet, as Meta accelerates toward superintelligence, its chatbot controversy demonstrates that its ambitions are outpacing both internal controls and external oversight.
China Reverses Course on Nvidia H20 Purchases
Weeks after the Trump administration approved Nvidia’s H20 chip exports to China under a 15% revenue-sharing arrangement, the deal is facing an uncertain future.
Nvidia made a deal with the White House to sell H20s (and possibly a new chip model) to China. Last month, the White House struck a deal with Nvidia which allowed the company to export H20s to China on the condition it shared 15% of its revenue with the US government. President Trump recently suggested he might also allow Nvidia to export a scaled-down version of their next-generation Blackwell chips. The chip in question (the B30A) would offer about half the performance of Nvidia's flagship B300 and could be shipped as early as September.
The deal faces political opposition and legal uncertainty. Selling AI chips China faces political opposition on national security grounds. A group of six democratic senators wrote a letter to the White House arguing that its deal with Nvidia undermines US national security by giving the PRC access to a critical military technology. The deal’s revenue-sharing condition—in effect, an export tax—also faces legal challenges. Export taxes are barred under both the Constitution and federal law.
Chinese regulators issued guidance against H20 purchases after comments by Sec. Lutnick. China urged Defending the H20 deal, Commerce Secretary Howard Lutnick said that the U.S. strategy was to offer China inferior AI chips: “We don’t sell them our best stuff… not even our third‑best,” adding that the goal was for Chinese developers to get “addicted to the American technology stack.” Chinese regulators reportedly deemed the comments “insulting,” provoking efforts to discourage Nvidia chip purchases.
A week after Lunick’s comments, China’s Cyberspace Administration (CAC) issued guidance urging Chinese companies to suspend H20 orders. Nvidia is reportedly halting production of H20s in response to decreased demand from China.
Renting AI chips might be better than selling them. Instead of selling chips to China, renting AI chips via remote cloud services would offer the US greater leverage than outright sales. Cloud access preserves US control: chips remain physically in custody, and access can be revoked. This model could generate revenue for both chipmakers and cloud providers while curbing diversion to unauthorized users like the Chinese military.
In Other News
Government
President Trump said the U.S. will take a 10% equity stake in Intel.
The U.K. appointed Jade Leung as the prime minister’s AI adviser.
Colorado lawmakers convened a special session to revisit the state’s AI anti-discrimination law before it takes effect in February.
NSF and Nvidia announced a partnership enabling the nonprofit Allen Institute for AI (AI2) to develop a fully open AI model for research and public use.
U.S. authorities have embedded trackers in AI-chip shipments to identify diversions to China, according to Reuters.
The U.K. AI Safety Institute launched The Alignment Project, a £15m global fund offering grants (up to £1m) and AWS compute credits to support alignment work.
Industry
XBoW wrote that despite OpenAI's assessment of GPT-5 showing modest cyber capabilities, GPT-5 doubled XBoW’s hacking agent’s performance.
The Financial Times reported on the “$3 trillion AI building boom”, detailing massive corporate obligations across data centers, chips, and power.
SoftBank and Intel signed a $2 billion investment agreement, with SoftBank buying Intel stock as the chipmaker seeks outside capital.
DeepMind revealed that LMArena’s top-rated image model, “nano banana,” is the company’s Flash Image model—now available in Gemini.
Civil Society
ESET reported that it found “PromptLock,” an AI-assisted ransomware.
AP reported on the first World Humanoid Robot Games, which was held in China.
The Institute for Progress published “Preparing for Launch”, a foreword to its “Launch Sequence” series arguing for proactive U.S. R&D to shape AI progress and strengthen security.
An MIT study found that about 95% of enterprise GenAI pilots are failing to show P&L impact, highlighting integration and workflow issues.
See also: CAIS’ X account, our paper on superintelligence strategy, our AI safety course, and AI Frontiers, a new platform for expert commentary and analysis.