OpenAI Rewired - Sync #518

Plus: OpenAI buys Windsurf for $3 billion; new Gemini 2.5 Pro; Amazon's new warehouse robot; Apple eyes move to AI search; the Physical Turing Test; Neuralink gets a boost from generative AI; and more

Hello and welcome to Sync #518!

This week, OpenAI announced plans to transition into a Public Benefit Company, and we will take a look at what OpenAI proposes.

Elsewhere in AI, OpenAI has acquired Windsurf, an AI coding start-up, for $3 billion. Meanwhile, Google released a new version of Gemini 2.5 Pro that is even better at coding. Additionally, Apple is exploring a move into AI-powered search, Gemini has beaten Pokémon Blue, and Singapore has released a new blueprint for international cooperation on AI safety.

Over in robotics, Amazon unveiled a new robot coming to its warehouses, while various robotaxi and autonomous car companies announced expansions into new cities. We also have Jim Fan from Nvidia sharing his vision of the Physical Turing Test, and an article pouring cold water on the humanoid robotics hype train.

Apart from that, Neuralink gets a boost from generative AI, Bryan Johnson wants to start a new religion in which “the body is God”, can AI usher in The New Enlightenment, and we will finish this week’s issue of Sync with cyborg cicadas playing Pachelbel’s Canon.

Enjoy!

OpenAI Rewired

Originally launched as a nonprofit in 2015 with a commitment to ensuring artificial general intelligence (AGI) benefits all of humanity, OpenAI has since grown into one of the most influential and heavily funded tech companies in the world. However, that growth came with growing pains—chief among them, how to secure the vast resources needed to develop and deploy frontier AI systems.

OpenAI’s leadership soon realised that the organisation’s idealistic nonprofit model, while mission-driven, was ill-suited to the capital demands of cutting-edge AI, and began exploring alternatives.

This move drew significant pushback from former employees, nonprofit advocates, and OpenAI co-founder Elon Musk, who filed a lawsuit alleging the company had strayed from its original mission. With pressure mounting from regulators and investors alike, OpenAI’s new structure is an attempt to chart a middle path—one that may ultimately reshape how mission-driven tech companies operate at scale.

What’s changing

OpenAI’s new corporate structure begins with a fundamental change: its for-profit unit, previously a capped-profit LLC created in 2019, will be converted into a Public Benefit Corporation (PBC). As Sam Altman explained in a letter to employees, the goal is to maintain the company’s mission-driven ethos while enabling greater access to capital. Unlike traditional corporations, PBCs are legally required to balance shareholder returns with a commitment to public benefit—in OpenAI’s case, ensuring AGI serves all of humanity. The move brings OpenAI in line with other leading AI labs such as Anthropic and xAI, which already operate under the PBC model.

What makes OpenAI’s structure unusual is that the newly formed PBC will not be fully independent. It will remain under the control of OpenAI’s nonprofit parent, which will also become a major shareholder. According to CEO Sam Altman, this is neither a spinoff nor a sale, but a restructuring of governance and financial incentives that preserves nonprofit oversight while creating a more conventional framework for investors. The shift away from the capped-profit model, which limited investor returns, also opens the door for larger capital infusions.

Altman argues the goal is to simplify OpenAI’s structure while maintaining its mission, enabling it to raise the enormous funds needed to build and deploy advanced AI systems. OpenAI says the nonprofit’s growing equity stake in the PBC will allow it to fund more programs in education, healthcare, and scientific research, ensuring that commercial success directly supports the public good.

What about Microsoft?

Despite OpenAI’s efforts to present its new structure as a balanced solution, the approval of one key partner remains uncertain: Microsoft.

The tech giant, which has invested nearly $14 billion into OpenAI and serves as a critical infrastructure and distribution partner, remains the most significant holdout among investors. According to Bloomberg, Microsoft has yet to formally approve the restructuring and is actively negotiating its terms behind the scenes. No other investor has been invited to renegotiate, signalling that Microsoft’s buy-in could make or break the transition.

Microsoft’s concerns are tied to the complexity and exclusivity of its relationship with OpenAI. Beyond equity, the two companies have licensing and revenue-sharing agreements that give Microsoft access to OpenAI’s models and APIs, as well as integration rights across its products, from Office to Azure. According to reports, currently OpenAI shares about 20% of its top-line revenue with Microsoft, but internal forecasts suggest that figure could fall to 10% by the end of the decade, which could have major implications for Microsoft's long-term returns.

While SoftBank is reportedly prepared to move forward with its full investment, Microsoft’s deeper strategic entanglements make its position far more consequential. Until those negotiations conclude, the fate of OpenAI’s recapitalisation remains unresolved.

Whispers of IPO

One of the key questions surrounding OpenAI’s new structure is whether it lays the groundwork for a future initial public offering (IPO). While OpenAI insists it has no immediate plans to go public, converting its for-profit arm into a PBC theoretically allows for such a move. PBCs can list on public stock exchanges, giving them broader access to capital, something OpenAI may require as its ambitions grow.

However, the path to an IPO under this hybrid structure raises more questions than answers. If OpenAI’s nonprofit retains control over key intellectual property and strategic decisions, what exactly would public shareholders be investing in, and how much influence would they have? Can a company truly prioritise both societal benefit and investor returns, especially when guided by a mission-driven board? And if the nonprofit ultimately holds the power, would that limit the kind of transparency and accountability typically expected of a publicly traded company?

For now, OpenAI’s leadership maintains that the structure strikes the right balance. But should the company pursue an IPO in the future, these governance tensions are likely to resurface with even greater intensity.

---

OpenAI’s restructuring is more than a legal or financial manoeuvre. By attempting to retain nonprofit oversight while opening the door to greater capital and scalability, OpenAI is trying to prove that it’s possible to build transformative technology without losing sight of who it’s meant to serve.

Yet the outcome remains far from certain. With major stakeholders like Microsoft still at the table, legal questions under review, and public trust hanging in the balance, OpenAI’s new path will be closely watched by regulators, investors, and civil society alike. The stakes are high—not just for the company, but for the future model of AI governance in a world increasingly shaped by it.

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

Bryan Johnson wants to start a new religion in which “the body is God”
Bryan Johnson, a tech entrepreneur known for his extreme anti-ageing efforts, is launching a new religion called Don’t Die, centred on the goal of preserving human life. In this short interview, Johnson explains how the religion is driven by his belief that AI poses an existential threat to humanity, and that aligning AI with human survival is crucial. He envisions Don’t Die as a decentralised, inclusive movement that reimagines the body as sacred, blends science with ritual, and offers a modern framework for navigating identity and purpose in an AI-driven future.

Neuralink's speech restoration device gets FDA's 'breakthrough' tag
Neuralink has received the FDA’s "breakthrough" designation for a device aimed at restoring communication in individuals with severe speech impairments caused by conditions such as ALS, stroke, and cerebral palsy. The designation, part of the FDA’s programme to accelerate the development and review of innovative medical devices, could speed up the device's path to market.

This patient’s Neuralink brain implant gets a boost from generative AI
The article covers the story of Bradford G. Smith, the first nonverbal person with ALS to receive a Neuralink brain implant, allowing him to control a computer with his thoughts. It explores how Smith communicates using a combination of the implant and Musk's AI chatbot, Grok, which helps him generate responses more quickly. The piece raises ethical questions about identity and authorship as AI becomes more integrated with brain-computer interfaces, highlighting the transformative—but complex—intersection of human cognition and advanced technology.

🔮 Future visions

How AI could usher in The New Enlightenment
The article argues that artificial intelligence is significantly underappreciated in terms of its future impact, suggesting we are entering a "New Enlightenment" era comparable to the 18th-century Enlightenment. Just as that period introduced transformative inventions that reshaped civilisation, such as steam engines and representative democracy, today's emerging technologies—including AI, clean energy, and bioengineering—will drive fundamental changes in our economy, society, and governance. The article contends that we must prepare for a civilisational-scale shift, with new systems for capitalism, democracy, and global coordination likely needed within the next 25 to 75 years.

🧠 Artificial Intelligence

OpenAI Reaches Agreement to Buy Startup Windsurf for $3 Billion
OpenAI is set to acquire Windsurf, an AI-powered coding assistant formerly known as Codeium, in a deal valued at around $3 billion—its largest acquisition to date. The move positions OpenAI to better compete in the growing market for AI-driven programming tools, where it faces rivals such as Anthropic and Microsoft-owned GitHub, as well as start-ups like Cursor, Lovable, Replit, and more.

Gemini 2.5 Pro Preview: even better coding performance
Google has released the Gemini 2.5 Pro Preview (I/O edition) ahead of schedule, which, according to Google, excels in coding tasks, particularly in front-end and UI development, and offers improved agentic workflows. Google also claims that the new model ranks number one on the WebDev Arena leaderboard and excels in tasks such as video-to-code conversion and rapid app prototyping.

Apple Eyes Move to AI Search, Ending Era Defined by Google
Apple is exploring a revamp of Safari to prioritise AI-powered search engines, potentially ending its $20 billion-a-year deal with Google as the default search provider, Bloomberg reports. Testifying in a US antitrust case, Apple executive Eddy Cue cited declining Safari searches and rising AI use as signs of a tech shift, mentioning OpenAI, Perplexity, and Anthropic as future options.

Singapore’s Vision for AI Safety Bridges the US-China Divide
Singapore has released a new blueprint for international cooperation on AI safety, following a high-level meeting of researchers from the US, China, and Europe. The Singapore Consensus outlines shared priorities for studying risks from advanced AI, building safer models, and developing control mechanisms. Experts warn that current methods for controlling powerful AI may be insufficient, emphasising the urgency of global collaboration amid accelerating AI development.

Google’s Gemini has beaten Pokémon Blue (with a little help)
Gemini 2.5 Pro has done something Anthropic’s Claude 3.7 Sonnet could not do—it completed a classic Pokémon game. The feat was achieved by Joel Z, a 30-year-old software engineer unaffiliated with Google, who streamed the gameplay. While this is a great achievement for Gemini, Joel Z cautioned against using the event as a benchmark, noting that Gemini and Claude use different tools and methods, including “agent harnesses” that help interpret game visuals and assist decision-making.

A Judge Says Meta’s AI Copyright Case Is About ‘the Next Taylor Swift’
Meta is facing a copyright lawsuit from authors, who allege the company illegally used their books, sourced from shadow libraries like LibGen, to train its generative AI tools. At a recent hearing, a US District Judge focused less on the alleged piracy and more on whether Meta’s AI harms the authors’ book sales, questioning both sides' arguments. While Meta claims fair use, the judge expressed scepticism about that defence, but also doubted the authors could prove financial harm. The outcome of this case could set a major legal precedent for generative AI and copyright law.

Hugging Face releases a free Operator-like agentic AI tool
Hugging Face has released the Open Computer Agent, a free and open-source equivalent of OpenAI’s Operator, which can use a web browser to complete tasks on behalf of the user. As the article notes, the open-source agent is slow, occasionally inaccurate, struggles with complex tasks, and cannot bypass CAPTCHAs. However, the point was not to deliver a flawless product but to show that open AI models are becoming more capable and cost-efficient.

Microsoft launches Recall and AI-powered Windows search for Copilot Plus PCs
Microsoft has officially launched its Recall tool for all Copilot Plus PCs. Initially delayed due to privacy concerns, Recall is now opt-in, with encrypted data and sensitive content filtering. Alongside Recall, Microsoft has rolled out an AI-enhanced Windows search that supports natural language queries and a new “Click to Do” tool, a tool similar to Google’s Circle to Search, which allows users to take quick actions on-screen (e.g., summarise text, edit images). While available now, some features are currently limited to Qualcomm-powered devices and will not launch in the EU and a few other regions until later this year.

Amazon is working on a secret project called 'Kiro,' a new tool that uses AI agents to streamline software coding
Amazon is reportedly joining the AI-assisted coding scene with Kiro, a new tool designed to streamline software development using AI agents and a multi-modal interface. Built by Amazon Web Services, Kiro can generate code in near real-time, auto-create technical documents, flag issues, and optimise code based on user input, including text and visuals. AWS considered launching Kiro in late June 2025, though it is unclear if that timeline is still valid.

Mistral AI: Medium is the new large
Mistral AI has released Mistral Medium 3, a new language model which, according to Mistral, offers state-of-the-art performance at a fraction of the cost of competitors like Claude Sonnet 3.7 and Llama 4 Maverick while also being easier to deploy. Tailored for enterprise use, it supports flexible deployment (cloud, hybrid, on-premises), custom fine-tuning, and integration into existing enterprise tools and systems. Mistral Medium 3 is available now via Mistral’s platform and Amazon SageMaker, with broader cloud support coming soon.

China’s Huawei Develops New AI Chip, Seeking to Match Nvidia
Huawei is preparing to test its most advanced AI chip, the Ascend 910D, aiming to rival Nvidia’s H100 amid growing US export restrictions, the Wall Street Journal reports. The new processor, though less power-efficient, uses advanced packaging to boost performance and is part of China’s push for semiconductor self-sufficiency. Despite production challenges due to limited access to cutting-edge manufacturing tools, Huawei plans to ship over 800,000 chips this year and is developing powerful computing systems like the CloudMatrix 384 to compete at the system level.

AI firms warned to calculate threat of super intelligence or risk it escaping human control
Max Tegmark, a prominent AI safety expert, calls on AI companies to perform rigorous risk assessments—similar to those conducted before the first nuclear test—to calculate the likelihood of losing control over Artificial Super Intelligence (ASI). In a new paper, Tegmark proposes the concept of a "Compton constant," estimating a 90% chance that advanced AI could pose an existential threat. He argues that industry-wide consensus on such risk could drive global safety regulations.

---

If you're enjoying the insights and perspectives shared in the Humanity Redefined newsletter, why not spread the word?

---

🤖 Robotics

Amazon’s Vulcan Robots are Mastering Picking Packages
Meet Vulcan—Amazon’s latest warehouse robot designed to stow and pick items at near-human speeds in the company’s warehouses. While stowing is relatively straightforward, Vulcan tackles the tougher challenge of picking items from densely packed bins using AI, adaptive grasping, and visual feedback. Tested in a German warehouse, Vulcan marks a major step in Amazon’s push toward reliable robotic manipulation in messy, real-world environments.

How robots are building Amazon's future

Uber and WeRide set their robotaxi sights on 15 more cities
Uber and WeRide announced an expansion of their robotaxi services to 15 additional cities outside the US and China over the next five years, including locations in Europe. The expansion comes five months after the two companies launched a commercial robotaxi service in Abu Dhabi.

Aurora to add night driving, new routes as it ramps driverless trucking
Following a successful commercial launch with over 4,000 driverless miles (almost 6,500 km) completed, Aurora, a startup focusing on autonomous trucks, announced plans to expand its driverless freight operations by adding night and adverse weather driving in the second half of 2025, while extending routes beyond Dallas to Houston, El Paso, and Phoenix. Aurora also reported $871,000 in pilot revenue and expects modest 2025 earnings as it transitions towards a driver-as-a-service model. The update comes amid the resignation of co-founder Sterling Anderson.

Waymo ramps up robotaxi production at new Arizona factory
Waymo has revealed that its commercial robotaxi fleet now includes over 1,500 autonomous Jaguar I-Pace vehicles, with plans to expand through a new partnership with Magna to build 2,000 more at a newly opened factory in Mesa, Arizona. The facility will streamline production and validation, allowing vehicles to begin public service within 30 minutes of leaving the plant. The site is also set to support future platforms and, at full capacity, will produce tens of thousands of fully autonomous vehicles annually.

▶️ The Physical Turing Test: Jim Fan on Nvidia's Roadmap for Embodied AI (17:33)

In this video, Jim Fan, the director of AI at Nvidia, shares his vision of a future where robots perform physical tasks so seamlessly that humans can't tell whether a person or a machine did them. He highlights the limitations of current robotics due to scarce training data and slow, manual data collection, contrasting this with the rapid advancements in generative AI and simulation. Fan advocates for scalable, high-fidelity training through simulated environments powered by AI-generated assets and video models, enabling robots to learn complex tasks efficiently. He envisions a future with a “Physical API” and a new skill economy, where robots autonomously manipulate the physical world, transforming daily life as ambient, intelligent agents.

Why the humanoid workforce is running late
This article explores the growing disconnect between investor enthusiasm and expert scepticism surrounding humanoid robots, spotlighting the technical, safety, and practical limitations that hinder their widespread use. Despite flashy demos and massive funding, roboticists argue that real-world adoption will be slow and industry-specific, with many current claims overstating progress and capabilities.

Robot leaps and lands like a squirrel
By studying how squirrels land on branches, researchers at UC Berkeley adapted their one-legged Salto robot to leap and land precisely on narrow perches. Mimicking squirrel biomechanics, the team modified Salto to stick tricky landings using techniques like front-leg braking and body torque control.

▶️ How I Built a Real Movie Star Robot (12:57)

In this video, Davis DeWitt from Backhaul Studios shares how he built Newt, an adorable, fully functional robot created from scratch to star in the film The Lightning Code. He takes us behind the scenes, showing the entire process—from early sketches and concept art to 3D modelling, 3D printing, and assembly. With only five weeks to complete the build, DeWitt designed Newt to be expressive, durable, and camera-ready, even creating a stunt double for action scenes.

🧬 Biotechnology

GM mosquitoes: inside the lab breeding six-legged agents in the war on malaria
This article highlights the work of British biotech company Oxitec, which is developing genetically modified mosquitoes designed to curb the spread of malaria and other diseases by preventing female offspring from surviving. These mosquitoes were recently released in Djibouti to combat a rising threat from the invasive Anopheles stephensi species, which is driving a resurgence of malaria in urban areas. The project is still in early stages, but scientists hope it will become a valuable tool in global public health efforts.

💡Tangents

Will Supercapacitors Come to AI's Rescue?
Similar to how Brits strain the power grid by switching on kettles during breaks in televised events, the rise of AI training is causing synchronised energy spikes in data centres due to thousands of GPUs powering up simultaneously. To address this growing strain, companies like Siemens Energy, Eaton, and Delta Electronics are deploying supercapacitor banks that rapidly charge and discharge without degrading, smoothing out these bursts and helping stabilise the grid.

Cyborg cicadas play Pachelbel’s Canon
I did not expect to see the words "cyborg", "cicadas", and "Pachelbel's Canon" used in one sentence, but here we are: Japanese scientists have turned cicadas into musical cyborgs by attaching electrodes to their sound-producing muscles, enabling them to "play" melodies. If you are interested in how these cicadas sound, here is a video of them playing Pachelbel's Canon.

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!"