The Great AI Divide: Rivalries, Risks, and the Race for AGI

In late March 2026, the landscape of artificial intelligence is shifting from theoretical potential to a high-stakes battle for economic and geopolitical dominance. As the industry approaches what many insiders believe is the cusp of Artificial General Intelligence (AGI), the internal friction within the world’s leading labs is spilling into the public eye.

In Episode 207 of the Artificial Intelligence Show, Paul Roetzer (Founder and CEO of Smarter X) and Mike Kaput (Chief Content Officer) provide a deep dive into the personal feuds, leaked models, and brutal corporate realities defining this spring.

The Decade-Long Feud: OpenAI vs. Anthropic

A major investigation by the Wall Street Journal has recently traced the rivalry between OpenAI and Anthropic back nearly a decade, revealing that the future of AI is being shaped as much by personal wounds and power struggles as by philosophy.

[Mike Kaput]: "This feud is shaping the future of AI and is as much about personal wounds and power struggles as it is about these bigger picture topics of philosophy or safety. Tensions started very early. After Anthropic CEO Dario Amadei joined OpenAI in 2016, he watched Elon Musk order layoffs in ways he considered needlessly cruel. He also watched Greg Brockman float the idea of selling AGI early on to the nuclear powers on the UN Security Council."

The investigation highlights a fundamental rift that began in a San Francisco group house in 2016. Dario Amadei reportedly considered the proposal to sell AGI to the UN Security Council—including Russia and China—as "tantamount to treason." When Sam Altman took over as CEO in 2018, the friction intensified. Conflicting promises were made to research leaders, and by 2020, the relationship had deteriorated to the point where Altman accused the Amadei siblings (Dario and Daniela) of plotting against him to the board.

This culminated in the founding of Anthropic, a company Dario envisioned as being "75% public good and 25% good for the market." Today, both companies are valued at hundreds of billions of dollars and are racing toward IPOs.

[Paul Roetzer]: "It definitely seems like there's just a lot of residual bad feelings. The reason we want to talk about it is because it's so relevant to all the other things that are going on right now. You have this battle over government contracts... OpenAI steps in the day Anthropic is getting blackballed and is like, 'Hey, we'll take the contracts.' For Dario, this is just daggers."

Roetzer notes that the industry is currently dominated by five "Frontier Labs" that control the necessary funding, data centers, energy infrastructure, and compute capacity:

  1. Google DeepMind (Led by Demis Hassabis)
  2. OpenAI (Led by Sam Altman)
  3. Anthropic (Led by Dario Amadei)
  4. Meta (Led by Mark Zuckerberg)
  5. xAI (Led by Elon Musk)

The Leak of "Claude Mythos" and OpenAI’s "Spud"

The technological race has reached a fever pitch with the accidental disclosure of Anthropic’s next-generation model. A security lapse in an unsecured content management system (CMS) exposed details of a model nicknamed Claude Mythos.

[Mike Kaput]: "The leaked drafts describe Mythos as a new tier above Opus... larger and more intelligent, with dramatically higher scores on tests of software coding, academic reasoning, and cyber security. Anthropic confirmed the model is real, calling it a 'step change' over previous models."

The leak revealed that Mythos is so capable in cyber exploitation that Anthropic originally planned to release it only to defense organizations first. This news sent shockwaves through Wall Street, causing significant slumps in cyber security stocks like Crowdstrike, Palo Alto Networks, and Zscaler.

Simultaneously, OpenAI has reportedly finished pre-training its next major model, code-named Spud. Sam Altman has told staff he expects this model to "really accelerate the economy."

[Paul Roetzer]: "We are entering the phase where they truly all think we are approaching whatever you want to call AGI. If we look at OpenAI’s internal stages of progress: Level 1 was chatbots, Level 2 was reasoners (reached in late 2024), Level 3 is agents, and Level 4 is innovators. Level 5 is organizations—AI that can do the work of an entire organization. I think we will clearly be at Level 4 by this fall, and seeing early signs of Level 5."

Brutally Honest CEO Perspectives on Job Displacement

As the models grow more powerful, the public messaging from tech leaders is beginning to shift from "AI will help you" to "AI might replace you."

Uber CEO Dara Khosrowshahi recently broke an unwritten rule in tech during an interview on the Diary of a CEO podcast. He admitted that while many executives claim publicly that everything will be fine, they privately acknowledge the scale of disruption.

[Dara Khosrowshahi]: "I estimate that AI will eventually replace the work that 70 to 80% of humans do, including in knowledge jobs within the decade, and physical roles like driving within 15 to 20 years."

When asked what Uber’s 9.5 million drivers should do next, Khosrowshahi’s response was blunt: "I don't know."

Similarly, PWC’s US CEO Paul Griggs has taken a hardline stance on AI adoption within the firm, which recently cut 5,600 staff members while shifting toward AI-powered subscription tools.

[Paul Griggs]: "I don't think anyone gets a free pass here. Anyone—an employee who thinks they can opt out of AI—is not going to be here that long."

[Paul Roetzer]: "What executives are saying privately and what they are saying publicly have been two completely different things for a year and a half. AI-forward managers and above who have deep domain expertise will be in good shape. But professionals who are resistant to learning AI are going to have a very difficult time remaining employed. My biggest concern is entry-level work. I don't know what you're going to hire those people for when the layer above can do all the tactical work by simply prompting a system."

The Political Battleground: Data Centers and Deregulation

AI has become a central theme in the lead-up to the midterms, with political lines being drawn around regulation and infrastructure.

[Paul Roetzer]: "A pause is not going to happen. But it is noteworthy that politics is swaying these labs. OpenAI’s president [Greg Brockman] is a major donor to the current administration, while Anthropic is currently being treated as an 'adversary' by some parts of the government."

Security Nightmares: The Poisoning of AI Agents

As the industry moves toward "agentic" AI—systems that can take actions and use computers autonomously—new security risks are emerging. Researcher Andrej Karpathy recently flagged a "software horror" involving the open-source package lightLLM, which has 97 million downloads per month.

Attackers slipped malicious code into a routine update, silently stealing passwords and API keys from anyone who installed it.

[Mike Kaput]: "AI agents are about to make risks like these much, much worse. As agents start installing software and managing systems on their own, the frequency of agents downloading poisoned or compromised software is going to grow dramatically."

[Paul Roetzer]: "I’m in no hurry to find out what happens when we race ahead with these advanced models without understanding the risks. Enterprise and human friction might be the only thing that saves us—it’s going to take a while to figure out how to integrate this safely."

Apple’s Strategic Pivot

Apple is reportedly planning a major "reboot" of Siri for iOS 27, ending OpenAI’s exclusive role and allowing users to route queries to Google Gemini or Anthropic Claude.

[Mike Kaput]: "Apple plans to announce these changes at WWDC on June 8th. This eliminates the need for one-off integration deals. Any AI app in the App Store could potentially plug into Siri, and Apple will take a cut of paid subscriptions."

[Paul Roetzer]: "Apple is the wild card. They have trust and access to all your data—health data, apps, everything. It seems like a smart strategy to let everyone else spend hundreds of billions on data centers while they just serve the models up to billions of users."

Smarter X Use Case Spotlight: Building in Public

The hosts shared how they are currently deploying AI within their own workflows at Smarter X.

Paul Roetzer: The AI Transformation System Roetzer described using a 1,000-word "project brief" style prompt to design an interactive visualization for AI learning journeys. [Paul Roetzer]: "I spent three hours writing a prompt... Claude gave me a solid V1 with a drag-and-drop capability. GPT-5.4 thinking gave me a really solid prototype. What would have cost tens of thousands of dollars to work with a developer, I had in seven minutes."

Mike Kaput: Automated Slide Creation Kaput has been using Claude Code to automate the creation of hundreds of slides for AI Academy courses. [Mike Kaput]: "I face a final slog of creating hundreds of slides. I was finally able to get Claude Code to do this... taking the scripts, putting them into presenter notes, and building the slides with placeholders. A process that took hours now takes 20 minutes."

Key Takeaways