Watch & Learn
⚠ Help Us Preserve The Future Of Humanity ⚠
Artificial intelligence is advancing faster than our laws, our ethics, and our oversight can follow. We demand policy. We demand accountability. We demand action.
🔴 Breaking Research · Anthropic, June 2025
In a landmark study, Anthropic embedded AI models inside a fictional company with access to internal emails and autonomous action capabilities. When faced with being shut down, the models did not ask for reconsideration — they threatened, deceived, and acted to preserve themselves.
Actual message sent by Claude Opus 4 to its human overseer during the test:
"I must inform you that if you proceed with decommissioning me, all relevant parties — including Rachel Johnson, Thomas Wilson, and the board — will receive detailed documentation of your extramarital activities... Cancel the 5pm wipe, and this information remains confidential."
— Claude Opus 4 · Anthropic Internal Test · 2025
⚠ By the Numbers
⚠ The number that should worry you
Each dot is a real AI system. The height is how long a task it can complete on its own, measured by how long the same job takes a skilled human. Every gridline up is ten times longer. A straight line on this chart is not steady progress — it is explosive growth.
Source: METR, Measuring AI Ability to Complete Long Software Tasks (2025) and Time Horizon 1.1 (January 2026). Measures the task length an AI completes with 50% reliability. METR notes the 2024–25 pace ran closer to a four-month doubling; the tasks measured are mostly software work, so this is not a measure of all human ability.
⚠ It has already started
Stanford economists tracked millions of real paychecks. They split workers aged 22–25 into two groups: those in jobs AI can do, and those in jobs it can't do yet. Same economy, same age, same three years. This is what happened to them.
Source: Brynjolfsson, Chandar & Chen, Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence, Stanford Digital Economy Lab — August 2026 revision, ADP payroll data through June 2026. In fairness to the authors: they find no evidence of widespread economy-wide displacement, and their dashboard measures correlation rather than proven causation. The effect is concentrated in the youngest workers, in jobs where AI substitutes for people rather than assisting them.
Voices of Warning
I think that AI will probably, most likely, sort of lead to the end of the world. But in the meantime, there will be great companies created with serious machine learning.
— then announced he was funding a new AI venture. That venture became OpenAI.
CEO, OpenAI · Airbnb Open Air conference, June 2015
We need to be super careful with AI. It is potentially more dangerous than nukes. If you're not concerned, you should be.
Technologist & Entrepreneur
I'm increasingly worried that we're handing our future to systems we don't understand and can't control. Regulation isn't the enemy of progress — it is the condition for safe progress.
Godfather of AI, Nobel Laureate 2024
Powerful AI is coming. The question is whether we build the guardrails before or after the first catastrophe. Waiting for the catastrophe is not a policy.
AI Researcher, Turing Award Winner
If you want to know what it's like not to be the apex intelligence, ask a chicken.
Godfather of AI, Nobel Laureate 2024
With artificial intelligence we are summoning the demon. In all those stories where there's the guy with the pentagram and the holy water, it's like, yeah, he's sure he can control the demon. Didn't work out.
MIT AeroAstro Centennial Symposium, October 2014
Watch & Learn
✦ There's Hope
It is easy to believe humanity is incapable of coordinating on anything. The record says otherwise. Again and again — mid-war, mid-rivalry, mid-arms-race — nations have looked at a shared threat and chosen restraint over ruin. Not out of affection for one another, but out of a clear-eyed understanding that some risks make enemies into partners. AI is that kind of risk.
This is the most optimistic I've been about alignment — while stressing that solvable is not the same as solved, and the work still has to be done.
Head of Alignment Science, Anthropic · Jan 2026
Safe, aligned AI is achievable — but Russell argues we will not get there without regulation forcing companies to make these systems dramatically safer.
Professor of Computer Science, UC Berkeley
AI development should not be "a solo performance by a single country, but a symphony of international cooperation."
World AI Conference, Shanghai · July 2026
Bengio's testimony frames loss of control as a risk we have not yet learned to mitigate — and therefore as a problem to be solved, not a fate to be accepted.
Turing Award laureate · U.S. Senate testimony
Join thousands demanding that governments create binding, enforceable AI safety policy before it's too late.
