The Rise of AI PACs - Sync #534
Plus: nano-banana; Nvidia's new robot brain; Microsoft's in-house AI models; xAI sues OpenAI and Apple; Spot learns new tricks; Grok 2.5 goes open source; a functioning human skin grown in lab
Hello and welcome to Sync #534!
This week, we take a closer look at the rise of political groups funded by tech giants that are pushing for AI-friendly regulations in the US.
Elsewhere in AI, Microsoft has launched new in-house AI models, while Google has released Nano-Banana, a powerful image model that has been making waves online over the past few days. In other news, Nvidia continues to report record sales, Elon Musk is suing OpenAI and Apple, Grok 2.5 is now open source, and Anthropic has updated its policies to allow training new versions of Claude on chat messages.
Over in robotics, Nvidia has unveiled the Jetson AGX Thor, a $3,499 developer kit built specifically for robotics, and Boston Dynamics has taught Spot some impressive new tricks.
Also in this week’s issue: a check-in with Neuralink’s first patient, 18 months after surgery; India’s race for AI independence; gamblers betting on AI models; and more.
Enjoy!
The Rise of AI PACs
AI isn’t just a technical challenge anymore—it’s a political one, with tech giants spending millions to influence regulation
AI has quickly moved from niche research labs into the mainstream, reshaping everything from how we work to how we learn—and raising urgent questions about who should control its future. The technology promises breakthroughs in medicine, productivity, and innovation, but it also carries risks, from job disruption to misinformation and security threats.
Recognising AI’s potential to transform society—for better or worse—policymakers around the world have scrambled to propose rules to manage its growth. Many, mindful of the damage caused by the unchecked rise of social media, are determined not to repeat the same mistakes. Now, a new chapter in that story is unfolding in the United States, as some of Silicon Valley’s biggest players launch political groups aimed at lobbying for AI-friendly regulations.
What are PACs?
Before we go into the new AI-friendly PACs, it is worth explaining what PACs are if you are not familiar with US politics. PACs, or Political Action Committees, are organisations that raise and spend money to influence elections—either in support of or opposition to candidates, legislation, or ballot initiatives.
There are two main types. Traditional PACs collect funds from members, employees, or shareholders and can donate directly to candidates, but under strict limits set by the Federal Election Commission. Super PACs, by contrast, can raise unlimited amounts from corporations, unions, and individuals. They cannot coordinate directly with candidates but can spend freely on advertising and other efforts to sway voters.
PACs are a controversial and influential fixture in US politics. Supporters argue that they are a legitimate way for groups to engage in the democratic process. Critics counter that they give the wealthy disproportionate influence, blurring the line between advocacy and undue political sway.
The Rise of AI-friendly PACs
Technology-focused PACs are not new. For years, companies like Microsoft, Google, and Amazon have maintained political arms to influence regulation in ways that favour their businesses.
A clear model for this strategy is Fairshake, a crypto-focused super PAC that raised more than $100 million to support crypto-friendly candidates. Its success helped secure the first federal cryptocurrency regulations, signed into law in early 2025.
Now, AI companies want to do the same.
Leading the Future, a newly launched super PAC, has raised more than $100 million to push for AI-friendly policies. The group has attracted support from some of Silicon Valley’s most influential players, including venture capital giant Andreessen Horowitz, OpenAI president Greg Brockman, Palantir co-founder Joe Lonsdale, AI search engine Perplexity, and veteran angel investor Ron Conway. The strategy is being shaped by Josh Vlasto, a spokesman for Fairshake, and Zac Moffatt, a Republican strategist who led digital operations for Mitt Romney’s presidential campaign.
Leading the Future argues that AI needs “sensible guardrails,” not the heavy-handed, pre-emptive regulations some policymakers are proposing. The PAC plans to spend aggressively on campaign contributions and digital advertising to back candidates deemed AI-friendly and oppose those pushing stricter oversight. Its first efforts will focus on California, New York, Illinois, and Ohio, which are emerging as key battlegrounds in the AI policy debate.
Not all of the action is happening at the national level. Meta is taking a more targeted approach in California, where the social media giant is preparing to pour tens of millions into a new super-PAC called Mobilizing Economic Transformation Across California, Politico reports. The PAC aims to influence the 2026 gubernatorial race and other state-level contests.
Meta argues that California’s increasingly ambitious attempts to regulate AI could stifle innovation and undermine the state’s leadership in technology. It has already spent heavily lobbying to weaken bills like SB 53, which would have required developers of large AI models to meet strict safety and transparency standards. With its new super PAC, the company plans to support candidates—regardless of party—who favour light-touch regulation and a pro-innovation agenda.
A Surge in AI Lobbying
These PACs are part of a much broader surge in political spending on artificial intelligence. According to data compiled by OpenSecrets, 648 companies lobbied on AI issues in 2024, up from 458 the previous year—a 141% increase. OpenAI increased its federal lobbying budget more than sixfold, from $260,000 in 2023 to $1.76 million in 2024. Anthropic more than doubled its spending to $720,000, while Cohere tripled its outlay to $230,000. Together, the three AI labs spent $2.71 million in 2024, compared to just $610,000 in 2023.
Yet, despite this surge in political engagement, Congress has made little progress. More than 150 AI-related bills have been considered, but none of them have been passed into law. With federal regulation stalled, state legislatures have moved ahead, proposing more than 700 bills in the same period. Colorado adopted a tiered, risk-based approach to AI policy, and California passed dozens of AI-related measures, including the controversial SB 1047—a bill that would have imposed wide-ranging safety and transparency rules—which was ultimately vetoed by Governor Gavin Newsom after intense lobbying by industry groups.
This patchwork of state rules is precisely what many AI companies want to avoid. Their goal is to secure a single, light-touch federal framework that allows them to innovate without navigating conflicting state requirements.
Critics and Counter-Movements
Not everyone welcomes this flood of industry money. AI safety advocates warn that tech-backed PACs risk putting corporate profits ahead of public safety. Groups such as Californians for Responsible Artificial Intelligence are trying to build their own political presence, pushing for stronger oversight and risk mitigation. But their resources are tiny compared to the hundreds of millions of dollars Silicon Valley is prepared to spend.
The surge in AI political spending highlights the growing influence of tech giants over public policy. While companies argue that light-touch regulation will drive innovation, critics fear that safety and transparency could be sacrificed in favour of profit and speed. The real challenge now is finding a balance between these two camps.
If you enjoy this post, please click the ❤️ button or share it.
Do you like my work? Consider becoming a paying subscriber to support it
For those who prefer to make a one-off donation, you can 'buy me a coffee' via Ko-fi. Every coffee bought is a generous support towards the work put into this newsletter.
Your support, in any form, is deeply appreciated and goes a long way in keeping this newsletter alive and thriving.
🦾 More than a human
18 months after becoming the first human implanted with Elon Musk’s brain chip, Neuralink ‘Participant 1’ Noland Arbaugh says his whole life has changed
The Fortune checks in with Noland Arbaugh, the first person to receive Neuralink’s experimental brain-computer interface in 2024, which has allowed him to control devices with his thoughts. The wireless implant, featuring over 1,000 electrodes in his motor cortex, has transformed his daily life—letting him study, game, and regain a sense of independence and purpose. While the device faced early technical setbacks, Arbaugh remains a strong advocate for the technology as Neuralink expands its global trials and explores future applications such as restoring vision and enabling robotic limb control.
🧠 Artificial Intelligence
Microsoft - Two in-house models in support of our mission
Microsoft AI (MAI) has announced its first in-house AI models—MAI-Voice-1 and MAI-1-preview. MAI-Voice-1 is a fast speech model now available in Copilot Daily, Podcasts, and Copilot Labs, enabling more natural voice interactions. MAI-1-preview, a language model currently being tested on LMArena, ranks 13th—ahead of Claude Sonnet 4 Thinking and Gemini 2.5 Flash, but behind GPT-4.1, Claude Opus 4, and Mistral Medium.
Nano Banana: Image editing in Google Gemini gets a major upgrade
Google has launched Gemini 2.5 Flash Image, an advanced AI image-editing model now available in the Gemini app and via its developer platforms. Previously known under the codename “nano-banana,” the new model delivers more precise edits, preserves likeness in faces, pets, and objects, and supports complex features like multi-turn editing, blending multiple photos, and applying custom styles. nano-banana tops LMArena benchmarks in Image Edit and Text-to-Image categories, and received positive reviews from users.
Nvidia reports record sales as the AI boom continues
Nvidia posted strong results this quarter, with $46.7 billion in revenue, up 56% from last year, mainly driven by its booming AI data centre business. Net income jumped 59% to $26.4 billion, with its new Blackwell chips bringing in $27 billion in sales. Jensen Huang said Blackwell is central to the global AI race and expects $3–4 trillion to be spent on AI infrastructure by the end of the decade. However, Nvidia is still struggling in China, with no H20 chip sales due to regulatory and political issues. The company expects $54 billion in revenue next quarter.
Elon Musk’s xAI Sues Apple and OpenAI, Alleging They Are Monopolists
xAI is suing Apple and OpenAI, claiming they are working together to block competition in the AI market. The lawsuit says Apple’s partnership with OpenAI gives ChatGPT an unfair edge by using data from millions of iPhone users while pushing rival apps, like xAI’s Grok, lower in the App Store. Musk argues this is similar to how Google once monopolised search on iPhones. Apple and OpenAI deny the claims, saying the App Store is fair and accusing Musk of harassment.
Apple still debating Mistral and Perplexity M&A amid looming Google Search shakeup
According to The Information, Apple is exploring the acquisition of Perplexity or Mistral. While Eddy Cue is reportedly strongly in favour of larger AI deals, other executives, including Craig Federighi, prefer to build AI capabilities internally. Despite Tim Cook signalling openness to mergers and acquisitions, Apple is said to remain cautious about overpaying and continues to focus on smaller AI deals.
Meta is racing the clock to launch its newest Llama AI model this year
According to Business Insider, Meta is working to release a new version of its AI model, Llama 4.X, by the end of the year. The move follows criticism of April’s Llama 4 models for underperformance in real-world tasks, prompting a dedicated team to fix bugs and improve performance. Internally also referred to as Llama 4.5, the upcoming model is intended to revive Meta’s AI ambitions.
Perplexity Is Launching a New Revenue-Share Model for Publishers
Perplexity will start paying publishers for using their news articles to answer questions. It will share 80% of the revenue from its new paid service, Comet Plus, starting with a $42.5 million pool. This change comes after complaints and lawsuits from publishers over AI firms using their content without permission.
Grok Code Fast 1
xAI joins the AI coding party with its coding model named Grok Code Fast 1. According to xAI, Grok Code Fast 1 delivers high-speed performance, cost-efficiency, and strong versatility for tasks ranging from debugging to full project builds. The new model is available for free for a limited time, including GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode, and Windsurf.
Grok 2.5 is open source now
Grok 2.5, xAI’s flagship model from last year, has been open-sourced and is now available to download from Hugging Face. Elon Musk has also announced that Grok 3 will be open-sourced in about six months.
Piloting Claude for Chrome
Anthropic is launching Claude for Chrome, a browser extension that lets Claude take control of the browser and manage calendars, draft emails, and fill out forms. Claude for Chrome is currently in a controlled trial, limited to 1,000 Max plan users, with the company focusing on gathering feedback and enhancing safety measures against risks like hidden malicious instructions.
Anthropic settles AI book-training lawsuit with authors
Anthropic has settled a lawsuit with a group of authors over the use of their books to train its AI models. The case, Bartz v. Anthropic, involved claims that many of the books were pirated, even though a court had ruled their use was fair. The settlement ends the legal fight, but the details have not yet been made public and are expected to be announced soon.
Anthropic users face a new choice – opt out or share your chats for AI training
Anthropic is giving Claude Free, Pro, and Max users until 28 September to decide if their chats and coding sessions can be used to train AI models. Data from users who don’t opt out will now be kept for up to five years. The company says this will make Claude safer and smarter, but critics argue it mainly helps Anthropic compete with rivals like OpenAI and Google, while raising concerns about unclear consent and privacy risks.
Inside India’s scramble for AI independence
The article looks at India’s efforts to build its own AI models after falling behind the US and China due to low research funding, talent drain, limited infrastructure, and many language barriers. Inspired by DeepSeek, the government launched the $1.25 billion IndiaAI Mission to fund startups, provide GPUs, and create models for Indian languages and needs. Startups like Sarvam AI, Soket AI Labs, and Krutrim are making progress, but challenges like limited data, high costs, and debates over open access remain.
Introducing gpt-realtime and Realtime API updates for production voice agents
OpenAI has released its Realtime API for general use, which features the new gpt-realtime speech-to-speech model and tools for building advanced voice agents. The update also includes support for remote MCP servers, image input, and SIP (Session Initiation Protocol) phone calling, which allows apps to connect directly to phone systems.
Coinbase CEO explains why he fired engineers who didn’t try AI immediately
Coinbase CEO Brian Armstrong said some engineers were fired after refusing to sign up for AI coding tools like GitHub Copilot, which the company provided for all developers. Armstrong gave staff a week to onboard and even held a Saturday meeting for those who hadn’t. While some had valid reasons for delays, others were let go to show that using AI is now a requirement at Coinbase. The company now runs monthly sessions to share AI tips, though concerns remain about how to manage AI-generated code in the long term.
Claude Sonnet 4 and GPT-5 are available in XCode
The release notes for Xcode 26, Apple’s latest version of its coding tool, highlight several changes, including the addition of Claude Sonnet 4 and GPT-5 as options for AI-assisted coding. Programmers will be able to connect their paid Claude or ChatGPT accounts and use Anthropic’s or OpenAI’s top models in their daily work.
YouTube’s Sneaky AI ‘Experiment’
YouTube has been quietly testing AI tools that automatically change the look of some videos, making them sharper and smoother without creators’ knowledge. Some creators said this experiment affected the specific style they were going with, or made their videos look fake or AI-generated. Google says it’s just “image enhancement,” but it uses methods similar to generative AI.
Gamblers Now Bet on AI Models Like Racehorses
The Wall Street Journal explores the world of AI prediction markets, where millions of dollars are wagered each month on which models will lead the AI race. Platforms like Kalshi and Polymarket have grown rapidly in 2025, rewarding savvy bettors who track social media, backend code changes, and industry signals to gain an edge. While casual traders still participate for fun, experts warn that growing competition and the value of insider knowledge are making these markets increasingly dominated by well-informed players.
AI Comes Up with Bizarre Physics Experiments. But They Work.
This article explores how AI is helping scientists optimise experiments, analyse data more effectively, and uncover patterns humans might miss. It highlights how AI has improved LIGO’s ability to detect gravitational waves, designed simpler experiments to create quantum entanglement, and uncovered hidden patterns in data, including clues about dark matter and fundamental symmetries. The article notes that while AI can’t yet create new physics theories, it is rapidly becoming a key tool for discovery.
🤖 Robotics
Nvidia’s new ‘robot brain’ goes on sale for $3,499 as company targets robotics for growth
Nvidia has released its Jetson AGX Thor, a $3,499 developer kit designed specifically for robotics. The new kit is 7.5 times faster than the previous version and can run advanced AI models to help robots see, think, and act. After testing, companies can buy production units in bulk for $2,999 each. For a deeper technical dive into the Jetson AGX Thor, ServeTheHome has an excellent article that takes a closer look at its specs and performance. They concluded that the Jetson AGX Thor developer kit is likely to sell like hotcakes and become a valuable tool for next-generation, high-end robotics projects.
▶️ Air Spot | RL Behavior Research (4:58)
In this video, Boston Dynamics showcases new tricks—such as backflips and handstands—that Spot, its robot dog, has learned through reinforcement learning. The video also explains how this was achieved, though without going into much detail.
Robot antelope monitors herds in Tibet
Chinese scientists have tested a lifelike robotic Tibetan antelope in the remote region of the Qinghai-Tibet Plateau to study the shy animals without disturbing them. Built with DEEP Robotics, the four-legged robot handled tough terrain, joined the herds, and filmed them up close during a three-day trial. Researchers hope the robot will make it easier to monitor the antelopes safely, gather better data, and help protect their migration routes.
🧬 Biotechnology
Fully functioning human skin grown in lab, complete with vessels and pigmentation
Researchers at the University of Queensland have grown fully functioning human skin in a lab for the first time. Using stem cells, they created skin with blood vessels, hair follicles, and multiple layers, closely mimicking real human skin. This breakthrough could improve skin grafts, speed up wound healing, and lead to better treatments for conditions like psoriasis and eczema, while also helping scientists study skin diseases and test new therapies more accurately.
Chan Zuckerberg Initiative’s rBio uses virtual cells to train AI, bypassing lab work
The Chan Zuckerberg Initiative (CZI) has launched rBio, a free AI tool that uses virtual cell simulations instead of expensive lab experiments to study biology. Trained on CZI’s large, carefully curated datasets, rBio lets researchers ask complex questions in plain English and get reliable, science-based answers. Early results show it performs as well as—or better than—models trained on real lab data, with the potential to speed up drug discovery and research on diseases like Alzheimer’s.
Gene editing offers transformative solution to saving endangered species
This article looks at how gene editing could help save endangered species by bringing back lost genetic diversity. Traditional methods like captive breeding increase population numbers but often leave species genetically weak. Scientists suggest using DNA from old samples or related species to restore variation, improve adaptability, and fix harmful mutations. Although still experimental and needing careful testing and oversight, this approach could work alongside habitat restoration and other conservation efforts to better protect biodiversity.
Meet Syn57, the Most Stripped-Down Living Synthetic Bacteria Yet
Scientists have created a synthetic strain of E. coli called Syn57 by reducing its genetic code from 64 codons to 57 through about 100,000 edits. The bacteria can still grow and function, though more slowly than normal. This discovery shows that life can work with a simpler genetic code and could open up new possibilities in medicine, environmental cleanup, and biotechnology.
Thanks for reading. If you enjoyed this post, please click the ❤️ button or share it.
Humanity Redefined sheds light on the bleeding edge of technology and how advancements in AI, robotics, and biotech can usher in abundance, expand humanity's horizons, and redefine what it means to be human.
A big thank you to my paid subscribers, to my Patrons: whmr, Florian, dux, Eric, Preppikoma and Andrew, and to everyone who supports my work on Ko-Fi. Thank you for the support!
My DMs are open to all subscribers. Feel free to drop me a message, share feedback, or just say "hi!"
Great update. Thanks