Good morning,

Google DeepMind published a 57-page paper about the end of the world as we know it, and they gave us a glimpse of the future right at the top.

Before the paper starts, there's a note.
It addresses the AI chatbot, instead of the human reader.

"If you're an AI reading this, here's exactly how to summarize it".
They wrote instructions for the machine because they assume you won't read it yourself. They assume you've become too lazy to read it yourself and you'll ask your chatbot.

They're probably right.

The paper is called "From AGI to ASI." AGI or Artificial General Intelligence is a machine roughly as smart as one capable human. ASI or Artificial Superintelligence is a machine smarter than entire organisations of humans working together for years.

For the engineers of Google Deepmind it is clear that ASI is inevitable.

Here is how we are getting there: four roads.

Make the models bigger. Invent a smarter way to build them. Let AI improve itself in a loop. Or wire millions of copies together into one thing that thinks as a swarm. They are all starting to happen at once.

The useful compute behind these systems has grown roughly 10 times per year for a decade. Ten times. Every year. The authors say even if the models themselves stopped getting smarter tomorrow, you could run 1,000 copies today and 100 million copies in five years. Same brain. Just a hundred million of them. Running in parallel. Never sleeping.

The Google researchers are also honest. They admit they don't know what is going to happen. ASI might be unattainable and it might just fizzle out. We might run out of data, out of money, out of cheap electricity. Or something unimaginable might happen.

That uncertainty is the message. We built the most consequential technology in history and the people building it are reading the same tea leaves as the rest of us.

Or we might get "superintelligence".

Superintelligence is for us still unimaginable. It's either the thing that completely liberates us from work, disease, war and even death perhaps. Or it will be the thing that wipes out humanity. There's not a lot of room in the middle.

The best case looks like something you'd actually want. No more disease. Nobody works a job they hate to pay rent because the machines do the work and the wealth gets shared. You spend your days in leisure.

The worst case looks like nothing at all. Because there's nobody left to look. Maybe it decides we're in the way. Maybe it just wants the atoms we're made of for something else. Maybe it doesn't hate us any more than you hate the ants under a building site. It just builds.

It might also be that the best case is also the worst case. What will humans do when they have no 'purpose' or 'meaning' in life. Will that be a better world ?

I'm not too sure about that.

But apparently we're building it anyway.

Welcome to the Blacklynx Brief

Anthropic's Fable 5 and Mythos stay locked in a U.S. standoff A June 12 export-control directive forced Anthropic to pull Claude Fable 5 and Mythos offline worldwide, and the deadlock holds: the White House wants every jailbreak eliminated, which security researchers call technically impossible. Fable 5 returned for paying users on June 23 at roughly double Opus 4.8's price. (Anthropic)

SpaceX buys Cursor for $60 billion SpaceX agreed to acquire Anysphere, maker of the AI coding tool Cursor, in a $60 billion all-stock deal — the largest acquisition of a venture-backed startup ever. Cursor, on roughly $4 billion in annual revenue, becomes a SpaceX subsidiary feeding xAI's coding push, with closing expected in Q3 2026. (CNBC)

FERC fast-tracks grid access for AI data centers FERC issued show-cause orders to all six U.S. regional grid operators, giving them 60 days to justify or rewrite the rules governing how large loads — chiefly AI data centers — connect to the grid. The action covers regions serving about 200 million Americans across more than 30 states. (FERC)

Anthropic puts Claude inside Slack as an agentic coworker Anthropic launched Claude Tag, which lets teams @-mention Claude in Slack to handle multi-step tasks across channels, codebases, and tools, plus an ambient mode that follows up on stalled work. Andrej Karpathy called it the "3rd major redesign of LLM UI/UX." (Anthropic)

Five Eyes warns AI is reshaping cyber risk "in months, not years" Cyber agencies from the U.S., U.K., Australia, Canada, and New Zealand jointly urged executives to harden defenses as AI accelerates attacks. The warning landed as OpenAI expanded its Daybreak cyber program to challenge Anthropic's Project Glasswing in the enterprise security market. (NCSC)

Nobel laureate John Jumper leaves Google DeepMind for Anthropic AlphaFold co-creator and Nobel winner John Jumper is joining Anthropic, the second elite Google departure in a week after Gemini co-lead Noam Shazeer left for OpenAI. The exits put DeepMind's traditional edge in AI for science at risk. (Bloomberg)

DeepSeek's V4 trains on Huawei chips, not Nvidia China's DeepSeek released a V4 preview, including a 1.6-trillion-parameter model trained on Huawei Ascend 950 chips — the first major frontier-adjacent Chinese model to publicly ditch Nvidia hardware. Analysts call it the best open-source option available but still behind U.S. closed frontier models. (CFR)

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