Anthropic IPO 引发 AI 安全与医疗应用抗议
Anthropic IPO Draws Protests Over AI Safety and Healthcare
Anthropic 进入公开市场之际,ACT UP NY 与 Ban RSI 分别在纽约和旧金山组织抗议,质疑 AI 医疗应用及递归自我改进风险。招股书显示其 IPO 以每股 $83 发行 980 million 股 Class A 股票,募资 $81.34 billion;创始人持有的 Class F 股票掌握 50.1% 的一般投票权。
推荐理由
AI 评分 86
抗议诉求与招股书披露并置,呈现 Anthropic 上市后在 AI 安全承诺和治理问责之间的矛盾。
主题
正文
ACT UP NY and Ban RSI organized separate demonstrations in New York and San Francisco on November 19, 2026, as Anthropic entered public markets. The actions challenged AI in healthcare and recursive self-improvement, while a separate campaign pressed financial institutions over AI investment and the use of retirement savings.
ACT UP NY called supporters to 4 Times Square at 8:30 a.m. ET, where Anthropic was holding its Nasdaq IPO event. Its action notice asked members to hold an AI Out Of Our Healthcare banner, distribute flyers and potentially speak at the rally and to reporters.
ACT UP’s notice said some of the practices it opposes have already begun in hospitals and doctors’ offices. Its concerns include AI replacing live medical screening or hospital staff, influencing treatment and outcomes, training on health data, and racial, gender and anti-LGBTQ bias.
Ban RSI’s rally page scheduled a separate gathering outside Anthropic’s office at 500 Howard Street in San Francisco. The event listing set it for 11 a.m. to noon Pacific.
The listing subsequently showed 164 accounts in its Went category, a host-platform RSVP status rather than a verified crowd count. Ban RSI listed former AI researchers Jacob Coxon and Daniel Kokotajlo and AI safety educator Rob Miles as scheduled speakers.
Ban RSI defines recursive self-improvement as AI doing the work of building more powerful AI. It wants large AI companies prohibited from using systems more powerful than those available at the rally for AI research, ideally through an international treaty. Its proposed enforcement mechanism would require more powerful systems to have safeguards blocking such research even internally, red-team testing of those safeguards, and exceptions for monitoring and safety research that clearly do not increase AI capability.
The group is not demanding a unilateral pause. It wants Anthropic to announce concrete, verifiable steps toward avoiding recursive self-improvement, potentially including urging other AI companies to follow suit or lobbying for legislation.
Anthropic’s final prospectus describes closing the loop as a structural advantage: successive Claude versions assist in building the next through experiments, training-data generation and processing, evaluations, safety audits, and code writing and review.
An Anthropic Institute article says full recursive self-improvement, meaning an AI system autonomously designing and developing its own successor, has not arrived and is not inevitable. It also says such a capability could increase the risk of humans losing control. The article identifies a remaining gap in Claude’s judgment about which goals and experiments to pursue, even as the system takes on more engineering and research work.
The company says it would expect to slow down or temporarily pause if other developers at or near the frontier did so verifiably. Its proposed conditions include multiple well-resourced labs in multiple countries and systems that can confirm compliance.
Anthropic argues that a unilateral pause would accomplish less and could change the industry front-runner. That creates a narrow point of agreement with Ban RSI, which also rejects a unilateral-pause demand, but a difference over the immediate obligation: the protesters want Anthropic to announce verifiable steps itself, while the company’s stated expectation to pause depends on reciprocal action.
Retirement Funds and Shareholder Control
The financial objections also turn attention to who can control Anthropic following its initial public offering. The final prospectus priced 980 million Class A shares at $83 each, an $81.34 billion offering. It provides for a single Class F share held by the founders’ Founder LLC to carry votes equal to 50.1% of general voting power, while the Long-Term Benefit Trust generally can elect and remove the Class T directors who constitute a majority of the board. Class A shares carry one vote each.
Those arrangements concentrate general voting power with the founders’ vehicle and board-selection power with the trust. The IPO therefore leaves a paired accountability question: whether Anthropic’s development practices can be verifiably constrained, and whether that governance structure will uphold the company’s public-benefit purpose of responsible AI development for humanity’s long-term benefit.
AI 热点