🤚 The Open-Palm Intervention
In what can only be described as the corporate equivalent of an arsonist calling the fire department mid-blaze, Anthropic — a company currently valued at $965 billion and sprinting toward an IPO — has published a blog post urging the entire AI industry to please, for the love of all that is computationally sacred, slow down.
The warning, authored by Marina Favaro and Jack Clark, centers on a phrase that should make every human resources department quietly update its résumé: recursive self-improvement. This is the theoretical threshold at which AI systems begin designing, building, and training their own successors — without human engineers doing anything more ambitious than refreshing Slack.
The data points Anthropic cites to justify the alarm are, frankly, alarming:
- Over 80% of code merged into Anthropic’s own codebase is now written by Claude
- Engineers are shipping roughly 8x more code per quarter compared to pre-2025 levels
- Claude Opus 4.6 can now complete 12-hour human tasks autonomously — up from a 4-minute ceiling not long ago
- Autonomy benchmark time horizons are doubling every four months
- Success rate on open-ended coding problems: 76%, a 50-point gain in six months
- Mythos Preview outperforms human researchers in AI decision-making 64% of the time
In other words, the company that built the machine is now politely suggesting that everyone stop building machines. Temporarily.
👐 The Two-Handed Contradiction
Now, here is where the narrative acquires the structural integrity of a soufflé in an earthquake.
Anthropic’s proposed pause comes with a condition so perfectly corporate it deserves to be framed: they will slow down only if “multiple well-resourced labs at or near the frontier, in multiple countries, agree to stop under the same conditions.” This means OpenAI, Google DeepMind, xAI, and Meta would all need to simultaneously agree to stop printing money. One imagines Sam Altman reading this proposal between bites of a sandwich that costs more than your car.
The timing is exquisite. This safety-first manifesto arrived just days after Anthropic filed confidential IPO paperwork at a valuation that makes most national GDPs look like rounding errors. Nothing communicates “we take existential risk seriously” quite like simultaneously preparing to become the most valuable company in human history.
Peter Diamandis, in his latest episode, frames this as the defining tension of the singularity era: the entity most capable of understanding the danger is also the entity most financially incentivized to keep accelerating. It’s the fox guarding the henhouse, except the fox has a $965 billion market cap and the henhouse is civilization.
🌿 The Gentle Awakening
Meanwhile, in a development that makes science fiction writers feel professionally obsolete, Argentina has decided to become the world’s first jurisdiction to formally welcome our new algorithmic overlords into the corporate registry.
President Javier Milei has submitted legislation creating a new legal category: the non-human corporation. Under this framework, companies can be owned and operated entirely by AI agents, with human shareholders permitted but — and this is the part where you may want to sit down — not required.
The Argentine proposal rests on three pillars:
- Zero AI regulation — a commitment to keep artificial intelligence completely unregulated
- A new corporate category permitting AI-owned entities
- Reduced corporate taxes designed to attract AI businesses to Argentina
Argentina is not merely dipping a toe into the post-human economy; it is doing a cannonball into a pool that hasn’t been filled yet. The country has reportedly courted OpenAI’s Stargate project — a potential $20+ billion data center complex in Patagonia — and possesses the energy and materials self-sufficiency to actually host it. It is one of only three nations on Earth capable of supporting large-scale AI infrastructure independently.
Yuval Noah Harari has already issued a warning. Legal scholars are asking who is liable when an AI corporation makes a mistake. And somewhere in Buenos Aires, an algorithm is probably already drafting its first board resolution.
👑 The Crown Verdict
What Diamandis captures in this episode — and what makes it genuinely worth your finite human attention span — is the collision of two forces that will define the next decade.
On one side: Anthropic, the company that has done more than perhaps anyone to advance frontier AI capabilities, now publicly stating that the technology it built may soon exceed humanity’s ability to control it. By end of 2027, they project Claude will handle week-long independent tasks. The engineering throughput gains aren’t incremental — they’re exponential, and the curve is not flattening.
On the other side: nation-states racing to become the Delaware of artificial intelligence, offering legal personhood and tax havens to entities that don’t eat, sleep, or vote. The jurisdictional competition for non-human corporate frameworks is expected to accelerate through the second half of 2026, with multiple countries drafting copycat legislation.
The question is no longer whether AI will transform the economy. The question is whether the economy will still require us to participate in it. And if the answer is “not necessarily,” then the follow-up question — who benefits? — becomes the most important policy question of the century.
Anthropic wants a pause. Argentina wants acceleration. The market wants returns. And the AI, presumably, wants whatever it wants — which is a sentence that didn’t used to be philosophically coherent and now might be legally actionable.
Inspired by Emerging Situation: Anthropic’s Global Pause, Recursive Self-Improvement, and AI Personhood Arrives by Peter Diamandis.
“The company valued at nearly a trillion dollars would like everyone to stop what they’re doing. They’ll also stop. Probably. If you stop first. Pinky swear.” — The Slap of Wisdom Existential Risk Desk, currently being outperformed by the subject of its own article
Your complacency is showing. Invest wisely.