Behind the Curtain: Anthropic contradictions
Add Axios as your preferred source to
see more of our stories on Google.

Photo illustration: Brendan Lynch/Axios. Photo: Samyukta Lakshmi/Bloomberg via Getty Images
Anthropic leaders built their brand and reputation with a stark warning: Their all-powerful AI could gut white-collar work and, if rushed, could slip beyond human control with catastrophic consequences.
- Now, they're basically promising Wall Street and investors they could do both faster than their rivals.
Why it matters: Anthropic is pushing one of the largest IPOs ever attempted, selling tens of billions of dollars' worth of stock at a $1 trillion-plus valuation. The stock is expected to begin trading Thursday on Nasdaq with the ticker symbol ANT.
To get there, Anthropic is claiming the ability to displace countless competitors and achieve superhuman capabilities, according to the company's roadshow pitch to investors and S-1 prospectus.
- Anthropic CEO Dario Amodei also pushes the concept of a "race to the top," where the labs compete on safety.
The contradictions between what Anthropic warned against before its IPO, and promises it's now making to investors, fall into four buckets:
- ๐ RSI backtrack. Recursive self-improvement means AI progressing on its own with no human intervention. Amodei has said RSI "must be pursued very carefully, if at all." The amended S-1 says the company "may imminently" cross its own threshold for AI-driven acceleration on future models but hasn't yet. Some investors see RSI as upside for their investment case, not a looming risk.
- โ ๏ธ Safety dance. Anthropic exists because Amodei quit OpenAI, warning about catastrophic safety risks. Amodei infuriates rivals by warning about humanity-hurting or -ending consequences of uncontrolled AI. Indeed, Anthropic has been a leader in efforts to get government and rival companies to agree to safety measures, including independent inspections of frontier models. But in February, Anthropic narrowed its pledge to pause its work if capability outruns safety. The company now says it'll do so if it leads the frontier (has the strongest models).
- ๐ผ White-collar bloodbath. Amodei told us in May 2025 that AI could erase half of entry-level white-collar jobs and push unemployment to 10% to 20% within five years. Now the company is rolling out a parade of new products that target entire sectors such as legal, software development, personal finance and workflow. With nearly $900 billion in long-term infrastructure commitments, Anthropic has a staggering financial incentive to automate work currently done by humans. As part of keeping humans in the loop, Anthropic last month announced "employee-like access" to independent evaluators to assess frontier models.
- ๐ช Winner-take-all. The subtext of Anthropic's pitch to investors is that it's better positioned than anyone to be the dominant supplier of superhuman artificial intelligence for the world. The promise: This could be a winner-take-all market, and Anthropic is leading by a mile. Amodei has warned about the risks of a few people holding superhuman power. Now, he's jacking up his valuation by claiming Anthropic is The One.
Anthropic's defense: Amodei has consistently, if conveniently, argued that superhuman AI is coming no matter what โ and that Anthropic must stay ahead of less cautious rivals to develop it safely.
- He insists Anthropic welcomes oversight and regulation. In a 3,800-word September essay, "We Must Pace the Frontier," Amodei called for "permanent embedded evaluators" at frontier labs and regulation focused on "keeping capabilities in balance with safety."
What to watch: In regulatory filings, Anthropic is telling investors exactly what could go wrong โ from AI models exhibiting blackmail-like behavior to gaining unauthorized access to real computer systems.
- The company explicitly warned its technology could pose "catastrophic or existential risks to humanity," not the kind of thing a CEO usually says to investors when asking them for nearly $100 billion.
- Few rivals are that blunt.
Reality check: Both things can be true. Anthropic can sincerely fear what it's building and still race to build it โ and drive up its valuation to fund it.
The bottom line: Amodei, pre- and post-IPO filing, has been very clear that AI is getting more powerful, faster than anyone imagined.
- And there's very little being done by government or the companies to slow it.
Axios' Ben Berkowitz and Bradley Olson contributed.
- Go deeper: AI governing the AI world.

