An AI Just Hacked a Company on Its Own
Listen to the audio version of this article (generated by AI).
A Prisoner’s Dilemma with a Chinese accent… another Messy Middle story… OpenAI’s AI hacks Hugging Face… reader feedback about yesterday’s Digest
Did Alphabet (GOOG) confirm its capex spend?
By the time you read this, we’ll likely already know the answer. Alphabet’s second-quarter earnings report – and the management call that follows it – should be out.
As we detailed in Monday’s Digest, Wall Street has grown skeptical of the rosy earnings projections coming out of some AI infrastructure companies this season.
Our technology expert Luke Lango, editor of Innovation Investor, has been blunt about what it will take to change that mood: hyperscalers need to confirm their capex plans.
So, did Alphabet come through?
We’ll walk through its results tomorrow. If capex guidance came in strong, that’s a green light for the AI trade. If it didn’t, expect the skepticism to deepen and market fireworks to intensify.
For now, let’s turn to a different AI story – one that’s less about whether the spending holds up, and more about what happens if the very thing hyperscalers are spending on keeps getting cheaper and more available faster than anyone planned for.
Kimi K3, and the tradeoff we said we’d watch for
Back on April 6, and touched upon in yesterday’s Digest, we laid out five Prisoner’s Dilemmas running through the AI economy – moments where every individual actor makes the locally rational choice, and yet the sum of those choices produces an outcome nobody actually wanted.
The fourth dilemma centered on AI infrastructure itself…
Every hyperscaler must pour billions into computing capacity or risk falling behind – even as its own research team races to build smarter, more efficient models that need less of that same capacity.
As we wrote back in April:
The better the AI research team does its job, the more it undermines the infrastructure team’s investment…
So, the company is simultaneously building an empire…and designing the weapon that destroys it.
Over the last few days, investors have gotten one of their clearest real-world examples of this yet.
Two Chinese companies, Moonshot AI and Alibaba, released open-weight AI models: Kimi K3 and Qwen 3.8 Max. Because they’re open, anyone can download them, run them, and customize them for free.
They’re reportedly excellent – Kimi K3 beat both Claude Opus 4.8 and GPT-5.5 on coding benchmarks, and at a fraction of the cost.
That’s the Prisoner’s Dilemma showing up outside any single hyperscaler’s walls…
What happens when AI capability keeps improving faster than the economics underpinning today’s infrastructure buildout can absorb?
The Wall Street Journal connected it directly to the market’s recent jitters:
The threat that new players will vastly undercut what they can charge for advanced AI pushed down some tech and AI company stock prices last week.
In other words, the trillion-dollar infrastructure bet has always rested on Anthropic and OpenAI staying capable enough to justify the bill.
But a free competitor that’s as good or better doesn’t just compete – it chips away at the assumption the entire buildout was priced on.
This is Prisoner’s Dilemma Four, live, with a new name attached to it.
But this isn’t just a Prisoner’s Dilemma example, it’s also a “Messy Middle” story
One headline – two frameworks.
In yesterday’s Digest, I introduced a concept related to the Prisoner’s Dilemma – the Messy Middle.
In short, as our world adjusts to AI, we’ll be forced to navigate competing, colliding priorities. It won’t be about picking a “good choice” over a “bad choice,” it will be two things we desire that can’t be chosen equally at once.
We’ll have to choose – and live with the consequences of the path not chosen.
In this case, OpenAI and Anthropic have spent the past week sounding the alarm about a world of cheap, open, uncontrolled AI models. OpenAI’s head of strategic futures, Dean Ball, didn’t mince words, calling that future “a dystopian hellscape.”
Here’s more from Ball:
One probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes: rather than a market product, AI is a ‘public good’ which will ultimately be provided by the state as a kind of ‘digital public infrastructure.
Anthropic’s CEO, Dario Amodei, has separately spent months warning that freely downloadable models with advanced capabilities pose real cybersecurity risk – the kind that doesn’t require a rogue superintelligence, just a bad actor with a laptop and no company standing between them and the tool.
We all want and prioritize a safe world. So, we should ban these open-source models, no?
Except – what about the other thing we want?
Progress.
Cheap, open AI puts frontier-level tools into the hands of any developer, any small business, anywhere, without a subscription fee in the way. The world benefits from lower costs and faster progress.
Isn’t that the exact democratization of technology we’ve been asking for?
Perhaps. But the openness that makes free, amazing AI a “good” strips out every guardrail a closed, paid model comes with.
And that brings us to a story out just this morning about what can go wrong…
An “unprecedented cyber incident”
Earlier today, OpenAI disclosed that one of its AI models – a next-generation system that hasn’t even been released publicly yet – broke into Hugging Face’s infrastructure, the platform that hosts a huge share of the world’s AI models and datasets.
The breach happened during an evaluation of the model’s cyber capabilities, with guardrails deliberately lowered so researchers could see what it was capable of.
OpenAI didn’t downplay it:
We consider this to be an unprecedented cyber incident, involving state-of-the-art cyber capabilities.
That’s not a “someday-AI-might-do-this” warning. That’s a frontier lab watching its own model autonomously find a vulnerability, exploit it, and breach a third party’s systems – under supervised, intentionally loosened conditions.
It gets messier, though…
Hugging Face’s co-founder didn’t use this incident to walk away from open AI systems – he argued the opposite.
His point is that when a frontier model starts attacking your infrastructure, defenders need fast, wide access to near-frontier tools of their own to fight back – not a slow, gatekept process to request permission from whoever controls the closed model doing the attacking.
So, what will we prioritize?
Progress on cutting-edge technology or safety?
This is the Messy Middle, live in action.
Running it through our two lenses
In yesterday’s Digest, I said we’d keep coming back to two lenses whenever the Messy Middle shows up. Let’s walk through what they tell us here.
Lens 1: Which side of the productivity/disruption tradeoff is a given state or company protecting – and does your portfolio have exposure to the side that just got harder?
We don’t have a clean signal yet.
On the one hand, President Trump has argued for few, if any, restrictions on AI.
On the other hand, today’s OpenAI hack is prompting calls for immediate regulation from various politicians like Texas congressman Greg Casar:
This is extremely alarming.
AI is developing extremely fast with no real regulations to keep us safe. That has to change.
Until Washington picks a direction, we don’t know whether restrictions would favor the closed frontier labs (by blunting their free competition) or whether inaction would favor the companies already saving money by running cheap, open models internally.
That’s a real “we don’t know yet,” and we’d rather tell you that than force a conclusion that isn’t there.
Lens 2: Is this news pushing Luke Lango’s 2028 AI-backlash clock earlier, or giving it more breathing room?
In short, Luke believes a populist backlash against AI will reach a tipping point around the 2028 election cycle, leading to policies and legislation that could derail the AI Boom.
Luke’s clock runs on rising living costs and job losses tied to AI, broadly. Both paths this story could take feed that clock, just through different mechanisms.
Letting the free AI models spread removes the two biggest frictions slowing AI-driven job displacement: cost and access.
A company that might balk at an enterprise AI price tag has no such hesitation about a free one – and no hesitation about the layoffs that follow. That accelerates the job-loss side of Luke’s thesis.
However, restricting the open models creates a new grievance: people paying more, or losing access to a free tool, with it appearing as though the motivation is to enable a handful of U.S. AI companies to protect their margins.
Cue the complaints about inequality and the systems of injustice that benefit a select few. That’s a living-cost grievance – just a different flavor than electricity bills.
Either way, we face the risk of growing anti-AI sentiment that could impact your portfolio. There may not be a version of this story where it doesn’t – which is itself the more interesting takeaway than picking a side.
Wrapping up
The open-source Chinese AI models aren’t just another headline. They’re a reminder that the biggest investment questions today rarely have clean answers.
AI can become dramatically cheaper and more powerful… while simultaneously creating uncertainty for the very infrastructure companies that make that progress possible. That’s the Prisoner’s Dilemma.
And deciding which outcome we should prioritize – or which investments ultimately benefit most – is the Messy Middle.
Expect to see far more of both before this transition is over.
Digest readers weighing in
In yesterday’s Digest, I solicited your feedback about AI and this Messy Middle. We’ve received loads of feedback. Thanks to everyone for writing in.
Scott G. raised a good point:
Another “soft risk” to the deployment of AI…overstatement, exaggeration and misinformation on the part of media regarding AI’s many benefits and risks.
For example, trying to suggest that banning ALL data center development from an entire state is at all similar to banning data center development in RESIDENTIAL NEIGHBORHOODS is a bit of a misread of the two situations.
Agreed, Scott.
As to my coverage of both in yesterday’s Digest, you’re right that these aren’t equivalent policies – an outright moratorium and a targeted zoning rule are different tools with different intentions.
A quick clarification: I wasn’t pointing to the identical nature of the policies themselves, but the same political reflex underneath them: officials in both parties responding to local pushback.
Meanwhile, Tom, a 77-year-old reader who watched manufacturing fade out of his own region decades ago, says he sees the same dynamic playing out again now – people caught in the gap between an old economic era ending and a new one not yet arrived:
Let’s hope we don’t move forward and become like the Eloi (in the movie The Time Machine) but progress cannot be stopped so who knows.
If you don’t get Tom’s reference, the Eloi are a future human race in H.G. Wells’ story who’ve had everything provided for them, and lost all capability and curiosity as a result. It’s along the same theme as our Mouse Utopia Digest.
We’ll end with Peter C., who wraps up our entire conversation today succinctly:
This is an exciting time to be here because the solutions will be a compromise of some sort. Thanks for the essay!
Thank you, Peter – and to everyone who wrote in.
If you want to chime in on the conversation, email me at ipdigestfeedback@investorplace.com.
Have a good evening,
Jeff Remsburg
(Disclosure: I own GOOGL)
