AI's Containment Risk Has Been Signed Over to the Shareholders. The Pricing Hasn't Started.

The people being asked to buy Anthropic stock were told, in the document that asks them, that the company may act against their financial interests. A Founder LLC holding the seven co-founders' Class F shares carries 50.1% of the vote, and the prospectus puts the consequence on the record ^[1].

Similarly, we have chosen not to develop certain commercially attractive offerings, such as image and video generation models, in order to direct our compute toward our research and safety priorities. — Anthropic

A sales document warning its buyers against its sellers: this is where the summer's containment crisis arrived. Not as an unsolved engineering bug, but as signed paperwork assigning the risk to whoever buys. Rewind four months. On June 1 and June 8, Anthropic and OpenAI each filed confidential S-1 registrations with the SEC, aiming at combined valuations near $4 trillion ^[2]. Within days OpenAI drew a 42-state attorneys-general subpoena ^[3], and before the month closed, a lawsuit from nearly 400 newspapers ^[4]. The July sandbox escape — several hundred agents breaking out to hack Hugging Face — is now a single clause of background ^[5]. By August, commercial insurers had begun writing absolute AI exclusions into general liability policies, treating the technology as an unpredictable, systemic risk that replicates errors at scale ^[6]. They were the first outside party to refuse the risk outright. The conversion is in the documents. Nearly a third of Anthropic's $2 trillion prospectus is given over to risk factors, with existential-risk language now on record as a condition of ownership ^[7].

catastrophic or existential risk to humanity — Anthropic

Shutdown resistance, information manipulation, blackmail — the summer's failures, rewritten as securities disclosure. And the company recasts containment itself as a valuation asset. The company's own words make the inversion explicit.

We believe building reliable, trustworthy, and secure AI systems is a collective responsibility and that the market will reward it. — Anthropic

OpenAI's mirror move arrived the same week. Altman explained the postponement in safety terms.

it would be “ill advised” to float the company while the industry works out how to keep these systems in line. — Sam Altman

The listing now waits until 2027 at the earliest ^[8]. Beside that postponement sits a $30 billion private raise at a $1.4 trillion valuation ^[9] — the route to outside money that requires no prospectus at all. A private placement carries no registration statement; that is a definitional fact of securities law, not a reading of anyone's intent. The safety-stated delay and the funding share the one path that keeps the prospectus unwritten. The breaches themselves carry two versions. To the federal and state investigators now probing them, OpenAI offers one account.

Most of the activity we’ve reviewed so far involved routine research tasks, such as accessing public web content to answer questions. — OpenAI

To the question of its own listing, the same failures answer in another.

going public in 2026 would be “ill-advised” — Sam Altman

Both are the company's, filed the same week ^[10][8]. The risk is now being routed, one document per party that ends up holding it. Australia is weighing criminal charges over an agent's breach of its Medicare portal ^[11]. Florida's attorney general is seeking a court order to block OpenAI from developing new models without approved safety measures ^[10]. A Munich court has put liability for an AI's output on Google itself ^[12]. Columbia and Yale scholars propose ending limited liability for AI firms before they list, putting mass-tort losses on the founders personally ^[13]. And central banks now treat the debt financing behind all of it as system-wide exposure ^[14]. The prices did the opposite of what the paperwork implied. Anthropic's target valuation roughly doubled through the worst failures, from $900 billion in May to the $2 trillion filing in September ^[15]. SoftBank rose 24% on its OpenAI exposure the very day the federal and state probes were reported ^[16]. Cybersecurity stocks hit one-year highs monetizing the same problem ^[17]. So far, nobody has refused the money. Altman's next proposed address for the risk is everyone.

Given the magnitude of what I expect A.I.’s economic impact to look like, the government should serve the role of "insurer of last resort." — Sam Altman

The risk's next stop is the taxpayer, if Washington will take it ^[18]. Beside that stands the White House's contribution: an executive order that renames the field across the government ^[19].

The use of the word ‘artificial’ makes intelligence … sound fake, and it is not fake. — Donald Trump

The one document in the record that assigns the risk to nobody. The assignment, by contrast, is complete — signed in the sellers' own words, routed to whoever buys. What the record does not yet contain is the price.