Good morning.

Last week I got an email from a reader asking me about AI Risk and if I thought companies would ever succeed in regulating this. Is the EU AI Act ever going to get enforced?

I contemplated my answer and I’m going to repeat the same thing that i did 3 years ago when I gave an AI keynote (and was never invited back because I contradicted the opinion of some lawyer who was instrumental in the EU AI Act) I honestly think that this is a technology that has been vastly underestimated by the entire world.

Companies are struggling to deal with all employees using AI all the time (shadow AI). If you have the choice of toiling at a task for 8 hours or have an AI do it for you in 8 minutes - you obviously choose the latter. And if your company doesn’t allow it or boxes you into an old model , then you simply find a way to route it to the frontier model. Companies will never get complete control over this because just like governments are lagging behind in years - companies are lagging behind in months. Meanwhile AI models keep getting released at breakneck speed.

There is no “regulating” this. At this moment in time : sure. Go ahead and draft you AI policies. Go complete your risk registers. But as it progresses, I predict it will become much bigger than you ever thought it would. It will run all of our lives.

The cat is out of the bag and will never get back in. And it’s not a cat but an octopus. And it has 8 billion tentacles instead of 8.

I think we might see the end of privacy and if we are unlucky the end of civilization.

There’s a reason that the billionaires of this world are building doomsday bunkers.

And this week - some more evidence to my thesis arrived.

Main news this week was the release of a manifesto called: Pacing the Frontier. It’s a single-page digital manifesto signed by 1,293 employees across the absolute pantheon of frontier AI: OpenAI, Anthropic, Google DeepMind, Meta, and SSI. These aren’t outside activists but the very architects of the technology : the chief scientists and lead engineers who are actively building the machine.

And they are terrified that the engine is about to catch fire and melt the steering wheel off.

The roster reads like an Oppenheimer roll call: Dario Amodei, Ilya Sutskever, Shane Legg, Jakub Pachocki, John Schulman.

The heart of their panic boils down to a single, explosive dynamic: automated AI research.

Remember how in this newsletter we were debating if AI was a bubble or it would ever get past the stage where AI writing was visible?

Oh, how things have changed. Indeed - we are getting the first AI models that recursively can train themselves. Millions of iterations - fully automated.

We aren't just talking about a chatbot that writes mediocre poetry or fixes your Python scripts anymore. We are talking about silicon entities designing their own architectures, running their own experiments, and optimizing their own code. This is where it gets serious and where it indeed becomes EXISTENTIAL.

In the personal comments attached to the letter, the metaphors read like a fever dream from the early days of Los Alamos. Researchers openly talk about a runaway chain reaction. They talk about "splitting the atom." They describe being "shocked" by the pace of progress-even while sitting in that driver’s seat. The consensus from the lab floors is stark: we are knocking on the door of systems that surpass top human capability on almost every single cognitive metric, and the margin for human error is evaporating into thin air.

"It’s a deadly race towards an intelligence explosion," wrote OpenAI's Leo Gao in his personal commentary.

So why don't they just stop? Why doesn't Anthropic or OpenAI just turn off the cluster and go fishing?

Because of the vicious, mechanical paranoia of the market. Under current rules, stopping unilaterally isn't heroic; it's suicide. If Lab A taps the brakes to run safety checks, Lab B rockets past them to grab the market share, the talent, and the geopolitical leverage. If the U.S. slows down, China speeds up. It is the classic Game Theory nightmare: a hyper-accelerated prisoner’s dilemma where everyone is sprinting toward a cliff edge because nobody can afford to be the first one to stop running.

That is why Pacing the Frontier isn't calling for a vague, noble moratorium. It’s a targeted, desperate plea to the U.S. government to lead an international coalition that can construct actual brakes-a blend of technical monitoring, compute throttles, and governance tools that force everyone to slow down synchronously.

Of course, the cynic in me-and you should always keep a well-fed cynic in your back pocket-wonders about the optics. There is a convenient, almost regal aura to telling the world, "We have built something so powerful, so cosmic, that only global superpowers can regulate us." It risks regulatory capture on an unprecedented scale, entrenching these exact few labs as the permanent custodians of human destiny.

But as the smoke clears, the underlying reality remains deeply unnerving. The institutions built to manage global technology-clunky, bureaucratic, four-year-electoral-cycle governments-are trying to regulate a industry that is beginning to self-improve on a microsecond clock cycle.

The high priests have thrown their hands in the air and admitted they can't govern themselves against the force of market gravity. The question now is whether the politicians in Washington and abroad can forge the tools to pace the frontier-before the frontier runs completely out of sight.

Scary thing is - you as a reader also knows governments and how they work.

The gist of it is : we’re probably screwed.

Welcome to the Blacklynx Brief

(And with that we are going on a 1-week hiatus. I’m taking a small break from the researching and writing- we’ll be back in the second week of August)

1,000+ frontier staffers ask the U.S. for an AI brake pedal Over 1,000 employees from OpenAI, Anthropic, Meta, and Google signed the "Pacing the Frontier" letter, urging Washington to build tools that could deliberately slow AI progress if automated research outruns human oversight. OpenAI and Anthropic both publicly endorsed it - a rare cross-lab admission the race is nearing a point of no return. (Pacing the Frontier)

Moonshot releases the largest open model in history Moonshot AI published the weights for Kimi K3, a 2.8-trillion-parameter model that is now the largest ever released openly and benchmarks near Claude Fable 5 and GPT-5.6 Sol. The move lands as U.S. officials accuse Moonshot of distilling American models to build it - a charge Beijing rejects as "AI hegemony." (Hugging Face)

Anthropic launches Claude Opus 5 at half the price Anthropic released Claude Opus 5, which it says out-benches its own flagship Fable 5 on many tasks at half the price ($5/$25 per million tokens). It topped Artificial Analysis's Intelligence Index and scored a perfect 42/42 on the 2026 International Math Olympiad problems. (Anthropic)

The open-weights fight leaves Anthropic the lone holdout An Nvidia-led open letter urging Washington not to restrict open-weight AI models grew past 50 signatories including OpenAI and Google, leaving Anthropic the sole frontier lab not on the list. Anthropic's Dario Amodei then broke his silence, backing chip controls and distillation crackdowns instead of an outright ban. (Anthropic)

That’s it for us this week. Please like and subscribe 🙂