Behind the scenes of AI: We were warned. 6 facts about AI from the co-founders of Axios
Jim VandeHei and Mike Allen wrote a column for Axios that begins with: Six facts. No exaggeration. All from the last 60 days.
Jim VandeHei and Mike Allen wrote a column for Axios that begins with: Six facts. No exaggeration. All from the last 60 days.
AI is the fastest-growing product category in human history. One of the latest models is so powerful that its creators refuse to release it to the public. OpenAI and Anthropic say their most powerful code-writing models are now building themselves. AI companies are becoming less and less transparent as their models become more powerful. The federal government is not demanding any transparency. AI discontent is growing rapidly. In early April, OpenAI CEO Sam Altman’s home was attacked twice in one week. He wrote in shock: «The fear and anxiety about AI are justified… Power cannot be too concentrated.» AI chaos is no longer theoretical: This year’s great software crash wiped out $2 trillion in value, as investors learned week by week what new human tasks the latest models would clean up, from writing code to real estate to legal research to financial management.
Why this matters. A year ago, we wrote a disturbing piece for business leaders. This one is for everyone: We were warned — by the data, the technology, and the people most responsible for creating it — that we had unleashed something powerful, something that was growing exponentially, and something that very few people, especially those in power, understood.
Between the lines. Think of it as the dawn of a new Atomic Age. The atomic race, which culminated in 1945, was the last time our species encountered such a transformative and exciting technology. Its potential—both for prosperity and destruction—gave rise to science fiction that imagined everything from utopia to apocalypse.
Much of the most viral AI writing could be considered modern science fiction. «AI 2027,» a 2025 attempt to model superhuman intelligence written by a former OpenAI researcher, ends with AI either supporting a pro-democratic revolution spanning the solar system or embarking on a brain-harvesting spree. This year’s discourse has followed a similar path. Matt Schumer’s viral piece «Something Big Is Happening» equated AI’s ability to generate code with the emergence of intelligence with real flavor. Citrini Research’s «Global Intelligence Crisis 2028» imagined a worst-case economic scenario in which neither governments nor markets provide any effective response.
These texts have caused so much discussion and in some places moved the markets — because they could turn out to be true. To be clear: most likely not. They represent extreme cases. But we can’t promise you that they are wrong. The president can’t. The heads of AI companies can’t. If someone claims that they can — that’s also science fiction.
The big picture. We have no idea how this will end, or what good or evil might emerge along the way. No one knows. But it is increasingly clear that, without better leadership, cooperation, and understanding, American society, workers, academic institutions, and government are completely unprepared for what is unfolding.
It begins with understanding the six realities.
Anthropic is the fastest-growing company in the history of American business. Its annual revenue jumped from $1 billion at the end of 2024 to $9 billion a year later and to $30 billion as of this month. More than 1,000 companies are spending more than $1 million a year on Claude, a number that has doubled in less than two months. No company in any era—Rockefeller’s Standard Oil, Google in the tech boom, Zoom in the pandemic—has scaled organic revenue so quickly from such a base.
Anthropic has built a model that it has no plans to release to the public. Claude Mythos Preview is capable of cracking critical security vulnerabilities in the operating systems and browsers that power the modern world. It is that powerful—and that dangerous. The company announced that access to Mythos will initially be limited to 40+ organizations to give cyber defenders a head start on what’s coming. No regulator forced Anthropic to choose this strategy. It did it on its own. Others may not do the same.
AI is now building AI. Boris Cherny, who runs Claude Code at Anthropic, said in late January that «virtually 100%» of the code inside Anthropic is generated by AI. The company’s official position is: «We build Claude with Claude.» It’s the same picture at OpenAI, where a senior researcher has publicly stated that he no longer writes code by hand. And that’s just the beginning. OpenAI’s chief scientist has called 2028 the target for a fully autonomous AI researcher. Anthropic’s co-founder expects a decision to allow the AI to recursively improve itself between 2027 and 2030, calling the move «the ultimate risk.»
The more powerful the models, the less we know about them. The Stanford AI Report 2026’s Fundamental Model Transparency Index has fallen from 58 to 40 (out of 100) in the past year. The straightforward conclusion: «The most capable models are now the least transparent.» There is no meaningful requirement to disclose the details of how the models are trained.
A 20-year-old man is accused of driving from Texas to San Francisco and throwing Molotov cocktails at Altman’s house at night, then showing up at OpenAI headquarters and trying to break down the door. He had a manifesto that named AI executives as targets. He said he was there to burn everything down and kill everyone inside. Two days after the attack, someone else drove up to Altman’s house and fired shots at him.
In 10 weeks, AI agents wiped out $2 trillion in the combined value of public software companies—the first to be targeted by AI. That’s more relative to the market than the dot-com bust or the 2008 financial crisis.
A few hours after the attack, Altman published a blog post. Read his words slowly:
«Fear and anxiety about AI are justified.»
«AI must be democratized; power cannot be too concentrated.»
«We are witnessing the biggest change in society in a long time, and perhaps ever.»
«We all learn new things quickly; some of our beliefs will turn out to be right, some will turn out to be wrong, and sometimes we will have to change our minds quickly as technology advances and society evolves.»
The person who runs the most widely used AI product in history has publicly admitted, hours after someone tried to kill her at home, that concerns about her technology are justified — and neither she nor anyone else knows exactly where we’re headed.
Bottom line: Each of these six facts in a normal industry would be the business event of the decade. Together, in a 60-day window, they describe a technology whose growth, power, and risk have outpaced public understanding—and whose creators say so in their own words.
We were warned.
Meta is creating an AI agent called Hatch, similar to OpenClaw but adapted for regular users, as well as an AI assistant for shopping on Instagram
Meta is creating an AI agent called Hatch, similar to OpenClaw but adapted for regular users, as well as an AI assistant for shopping on Instagram
«Artificial intelligence makes a sign for each government decision.» Bornyakov told how AI helps government officials predict who benefits from certain bills
«Artificial intelligence makes a sign for each government decision.» Bornyakov told how AI helps government officials predict who benefits from certain bills
Read the country’s main IT news in our Telegram
Read the country’s main IT news in our Telegram
