Martin Casado: If a company is forced to revise its S-1 filing due to employees publicly expressing concerns about extinction risk, it would be top entertainment
a16z general partner Martin Casado posted that if a company is forced to revise its S-1 filing due to employees publicly expressing concerns about extinction risk, it would be top entertainment, and he hopes to see this happen. He did not name the company, but the timing coincides with the resignation of an Anthropic researcher and the public expression of probability by the alignment lead.
The immediate trigger was Jacob Coxon's departure. The 27-year-old British researcher previously worked at OpenAI on pre-training (including GPT-4o related) and joined Anthropic around July 2026, resigning on September 8 after about four months. He told the media that both labs are racing towards self-improving superintelligence, gambling with lives; colleagues' catchphrases include "crunch time" and "endgame." Axios reported he left about two months before his equity vesting, forfeiting unvested Anthropic options, while still holding OpenAI equity. His resignation post has reportedly been viewed over 100 million times.
The current alignment pressure test lead, Evan Hubinger, publicly agreed: he seriously believes that AI could kill everyone, estimating a greater than 10% chance in the next decade, and stated there is no plan to solve superintelligence alignment, nor is there a clear path forward; the existing model risks are low, but the concern is recursive self-improvement happening faster than expected. Anthropic tech employee Samuel Marks wrote personally that the more senior one is, the more worried they become. Nobel laureate Geoffrey Hinton told the BBC that a 10% extinction estimate within ten years is "not unreasonable." The company's risk report raised the overall risk of current models from "very low" to "low," with a coverage date extended to July 15, 2026.
The S-1 filing is already in the queue. PitchBook stated that Anthropic secretly submitted its S-1 on June 1, followed by OpenAI about a week later; Semafor identified Wilson Sonsini handling Anthropic, and Cooley and Wachtell handling OpenAI, joking about what risk factors might look like if "the product could contribute to the collapse of civilization." The publicly discussed valuation and revenue anchors include approximately $965 billion in private equity benchmarks and an annualized revenue of about $47 billion. Casado himself mentioned in August that the mistake is not the rapid singularity, but rather that the burn rate and scale laws can continue, with hundreds of billions in training runs meaning resources are highly concentrated, and uses could be for healing or weaponization.
The market mechanism is the pressure test of disclosure clauses on valuation. The buy side wants to have the "employees have publicly expressed a 10% extinction probability" written into the risk factors by lawyers and cornerstone investors before the IPO; the sell side hopes the S-1 only discusses competition and computing power. Funds have not yet flowed due to revisions; what is flowing is the secondary market's pricing of whether safety governance constitutes a material financial issue. Beneficiaries are law firms specializing in risk disclosure; those under pressure are issuers who must choose wording between "mission narrative" and "no material omissions under securities law." The event-driven aspect is investors turning the resignation wave into S-1 anecdotes, rather than regulators issuing revision orders.
Once the S-1 is revised, the extinction probability will shift from social media to a litigable sentence in the prospectus. The entertainment value comes from the two types of texts being forced to share a page for the first time.
Source: Public information
ABAB AI Insight
Casado humorously brings securities law into the alignment debate. The function of the S-1 is to write down all facts that could affect investor decisions into a litigable list; employees and the alignment lead have already framed "greater than 10% extinction in the next decade" as an internal consensus. After the secret submission, the public market has not yet priced it, and lawyers must first decide if this statement constitutes a material risk. Coxon forfeiting unvested options is to establish that "I have no incentive to inflate valuations"; Hubinger remains in his position, placing the same probability under his company badge. The two identities appearing together means the S-1 cannot solely reference PR materials.
The capital pathway is the absorption of internal discourse into the IPO documents. Anthropic has raised substantial private equity as a public benefit corporation and safety brand, with OpenAI simultaneously making secret filings. Once risk factors must include self-improvement and extinction probabilities, cornerstone investors' models will shift from revenue multiples to "whether governance commitments are enforceable." The secondary market has already demonstrated valuation fluctuations within 24 hours of safety controversies. Money is still flowing towards training, but the disclosure layer is beginning to impose pricing discounts on the sprint.
Analogies include tobacco companies writing carcinogenic risks into their risk factors, Boeing including safety culture in annual reports, and social media framing content moderation as a business risk. Never before has a company sprinting towards a trillion-dollar valuation had its active safety lead publicly state a double-digit extinction probability. The industry phase is controlling the narrative: labs want "we are the most cautious, so we should win," while securities law demands "you have acknowledged the possibility of losing control." The two meet in the S-1.
The structural change is regulatory change. The mechanism is: discourse first spreads on social networks, then forces the S-1 to choose between omission or inclusion. Inclusion becomes a fixed issue for litigation and roadshows; omission would constitute a potential material omission in light of employee posts and Hubinger's responses. The extinction risk moves from research papers into capital formation documents, with pricing power shifting from model release events to the drafters of risk factors.
ABAB News · Cognitive Laws
The probability publicly expressed by employees will ultimately force the S-1 to take a side.
Forfeiting options is to ensure that warnings are not counted as interests.
The trillion-dollar narrative and extinction sentence are vying for the same page for the first time.