Anthropic just told you what it thinks Claude will do to the job market, and it did it with money instead of a forecast. Eight days before filing to go public, it put 5% of the company into a fund whose stated uses include people displaced by AI. At the valuations being discussed, that’s $75 to $100 billion. The filing adds a ratchet: every 50% rise in the share price, starting from the IPO price, lets the board hand the fund up to another 2%. The pool for the people Claude replaces is pegged to Claude’s stock price. The better the product does, the bigger the severance. You don’t pre-fund damages you don’t expect. That’s the most honest forecast in the IPO, and it’s filed under public benefit. The second thing worth knowing came from a podcast in February. Dario Amodei described the one way Anthropic goes bankrupt: buy compute for 10x growth and get 5x. Miss by a year, he said, and no hedge on earth saves you. That was his argument for restraint. Then Q1 usage grew at an 80x annualized pace, Claude hit rate limits, and the filing says Anthropic asked its cloud providers for more compute than they would give. So it bought the compute: commitments running to 2045, roughly $417B by CNBC’s count plus up to $150B of orders not yet priced, against a run rate near $67B a year. Then Q3 growth came in at 45% for the quarter, which compounds to about 4.4x a year. The cushion is real: $131.7B in cash and short-term investments, deals spread out to 2045, existing customers spending 9x what they did a year ago. The test is also real, and he wrote it. Q3 is the first quarter that moved toward the wrong side of it. The S-1 says it expects expert-level AI at scale within a few years. The fund is what Anthropic set aside for the people it replaces. The February test is what happens if it shows up a year late. The IPO asks the public to fund both: the bet, and the cleanup.