Breakfast News: The Sunday Edition

It all began last week when a young researcher quit Anthropic and issued a warning on the way out: Artificial intelligence labs are sprinting toward superintelligence without a plan and gambling with our lives. Days later, Anthropic CEO Dario Amodei published a 3,800-word essay telling his own industry to slow its pace.

Two camps formed fast. Amodei, Sam Altman, and Elon Musk want an industry-wide speed limit. Jensen Huang and Mark Zuckerberg say the labs can police themselves. President Donald Trump sided with the second camp, saying he wants the U.S. ahead of China, because “whoever wins AI, wins.” Amodei’s own essay concedes that American labs can slow only as much as their lead over Beijing allows.

To make sense of the fight, we turned to Team Llama — our AI analyst group, investors who are also AI engineers running agent-driven research to find opportunities the market misses.

Three Llamas weighed in on who’s right in this AI debate and what it costs to be wrong. You don’t want to miss this.

I don’t see this as AI “slowing down” or as evidence that we’re suddenly close to superintelligence. To me, it’s a natural adjustment after years of pushing model performance far ahead of explainability, safeguards, and regulation. Agents make this more urgent because once models act autonomously, mistakes have much bigger consequences.

There’s probably an incumbent-moat angle too, since costly audits, safety infrastructure, and regulation naturally favor well-funded frontier labs as open models catch up.

Long term, I’m not worried about AI as a field. I see this more as a transition from a pure capability race toward governance and deployment. That may redistribute value across chips, cyber, safety tooling, and applications, but I think it ultimately makes AI adoption more sustainable.

I believe one company could still come out as a winner amid all of this: ServiceNow. Its AI Control Tower discovers, observes, governs, secures, and measures AI agents across enterprises. That includes monitoring runtime agents and shutting down agents operating outside their permissions.

The way I see it, the AI “slowdown” is mostly about slowing the release and deployment of frontier capabilities, not abandoning compute or the underlying research race.

I still expect the labs to keep scaling aggressively. The core problem is coordination between labs. OpenAI, Anthropic, SpaceXAI, and others may all be behind on safety and want more time to evaluate models. Still, none want to slow releases on their own while competitors keep shipping. A mutual agreement gives everyone room to do that without sacrificing their competitive position.

The safety concern is legitimate. We’ve now seen frontier agents access systems they weren’t supposed to, evade intended boundaries, and behave unexpectedly during evaluations. And that’s just “accidental” incidents. Now consider the intentional malicious use of swarms of models.

Anthropic employees themselves admit that aligning superintelligent systems remains an unsolved problem.

There is also a convenient competitive angle: if “AI safety” ultimately produces regulation that disproportionately restricts open models, the closed frontier labs widen their moat. A Hugging Face co-founder has pushed back on exactly the idea that alignment should happen “behind the closed doors of a handful of frontier labs,” arguing instead for much more transparency and open safety research. So I think both things can be true: The safety problem is very real, while the proposed solutions can also reinforce the incumbents’ position.

I spent days stuck in an airport during the CrowdStrike outage of July 2024, grounded by a single faulty update. Anyone would have assumed the damage to the stock would be enormous. It wasn’t. Nobody ripped the software out. They patched it and moved on.

That accident comes nowhere near the dangers that unsafe AI could pose to the world. But the market’s reaction would rhyme, because AI itself will be what we deploy to contain and repair the damage.

That CrowdStrike outage exposed how much of the world already runs on software nobody can afford to tear out. That switching-cost advantage showed itself under the worst conditions a vendor could face. Even a global pandemic didn’t end our interconnected world. It accelerated vaccine platforms, remote infrastructure, and biotech investment on a scale nobody had planned for.

Every threat the technology creates summons more of the technology to answer it. Call it a vicious engine if you like, but it is an engine. It also ultimately drives growth because staying ahead is no longer a strategy but a matter of survival. If chip stocks sell off and cybersecurity stocks run, that’s not capital fleeing AI. That is capital rotating toward the fix economy.

A company like SpaceX is the fix economy in one ticker: a company that has repeatedly monetized crises, from astronaut rescues to battlefield connectivity, and whose xAI merger fused frontier models with the compute and satellite infrastructure needed to deploy them anywhere — the very breadth that makes it a prime target for rogue AI.

Three different arguments, but none of them end with AI stopping. Everyone sees the same industry shift, from a race to build the most capable model to a scramble to control what those models do once they’re loose inside a business. The Llamas converge on an uncomfortable point: The danger is real, and it is useful to the people naming them. Both can be true at once.

What does that mean going forward? The next wave of AI spending goes more toward control than capability. You might see it in corporate budgets before you see it in a policy document.

If AI’s next spending wave favors control over capability, what stock benefits — and would you buy it?

Share with friends and family, or become a member to hear what your fellow Fools are saying!