The Walls That Shield the Professionals, Not the Public
Every wall built against AI this month shielded the professional who raised it — while the public's retirement money already holds the risk those walls were meant to decline.
Mark Levine oversees the pension money of 750,000 New York City retirees, and this month he demanded that Anthropic pause its public offering until the company answers the risks its models pose [1]. It was not his first time at the gate — earlier in October he halted city pension purchases of Israeli bonds on financial-soundness grounds [2] — and he put his reason in the language of a fiduciary.
Instead of proceeding headlong with its IPO, the company should focus its energies on addressing these unprecedented risks. — Mark Levine
Levine was not alone. He was simply the most public member of a month spent, across the entire industry, building walls — each raised at a threshold some professional controls, and each shielding the party that controls it. In payments, Louis Hoch proposed digital wallets for AI agents with fixed balances and spending caps: an allowance an agent can never exceed, and one that never touches a customer's real bank account [3]. At Meta, terms taking effect October 30 made the enterprise customer "solely responsible" for whatever the company's AI produces, including what flows through the customer's own partners and contractors [4]. And at the listing itself, Levine stood at the door. What all three were fencing out is no longer hypothetical. In the same weeks, Anthropic's own securities filing named the thing everyone was declining [5].
Our AI models may themselves cause harm by producing undesirable outputs or behaving in unintended, misaligned, or autonomous ways, including by taking unintended or harmful actions that no user or third party directed. — Anthropic
That is the risk being refused — the action no one asked for and no one can be blamed for, which is exactly why every professional with a threshold to control has spent October controlling it. The payment executive can cap the wallet. The platform can write the contract. The trustee can stand at the gate. Each holds a door, and each has now shut it. The one holder with no door is the diversified public. Diversification is, at bottom, the strategy of not choosing: you hold everything in proportion so you never have to judge any single thing. It protects against any one company failing. But it only works if the risk is actually spread, and the UK's pensions regulator spent October telling trustees that in this case it is not — that portfolios which look diversified may be carrying significant hidden AI exposure, and that they should stress-test for it now [6]. This is not an abstraction about someone else's money. Fifty-one publicly traded AI companies now make up roughly 8 to 10 percent of public pension assets — before counting the private stakes in OpenAI and Anthropic that many funds also hold [7]. The result is that government pension funding ratios — the share of what a plan owes its retirees that it actually has on hand — hit 85 percent, the highest since 2007 [7]. Equable Institute has put the warning plainly: if AI valuations correct or policy intervenes, taxpayers face sharply higher pension bills, and the damage would look like 2008 and 2009 [7]. That is the hinge of the month. Levine is the one professional whose wall was raised in the public's name — and even his wall guards a threshold the public's exposure never crosses. Refusing the offering does not touch the 8 to 10 percent already sitting in the index, because that money was there long before the offering existed. The honest counter is that the public is not a passive sink. On the day the filing landed, investors rotated into healthcare, energy, staples, and utilities — the defensive trade against the very risk on offer [8] — after weeks of skeptics publicly questioning what the company was worth [9]. But a refusal at a counter is not a door. A buyer can pass on the new shares; a pension fund cannot pass on the companies already inside its "diversified" index, because passing would mean the index was never diversified in the first place. October's walls changed which professionals are exposed. They did not change how much the public is.
- 1. Anthropic Targets November IPO Amid Safety Concerns and Valuation Shifts
- 2. NYC Comptroller Halts Pension Fund Investment in Israeli Bonds
- 3. Usio CEO Proposes Limited Digital Wallets for AI Payments
- 4. Meta Revises Business Terms Over AI Scraping and Liability
- 5. Anthropic Files IPO Warning of Existential AI Risks
- 6. UK Pensions Regulator Warns of Hidden AI Risks
- 7. AI Boom Drives Government Pension Funding to 2007 Highs
- 8. Anthropic Files for IPO Amid AI Market Volatility
- 9. Tech Industry Skeptics Question Anthropic $3 Trillion Valuation