If you use Claude at work, or you are considering buying shares in the company behind it, a central question is whether Anthropic’s “safe AI” label matches what it does. A coalition of scientists, legal scholars, and tech leaders now says it does not. As an inference, the answer could shape how the company is judged as it moves into public markets.
For Claude users, the statement itself does not announce any change to Claude’s product. As an inference, the effect may fall on how Anthropic’s promises get checked. Once the company is publicly traded, which was expected to begin on 19 November 2026 under the ticker ANT, claims about its safety commitments could draw investor questions and regulator attention. As an inference, a broken or softened promise could carry more financial and legal weight than it did while the company was private.
What the statement claims
On November 18, 2026, the Center for AI Safety (CAIS) and a broad group of signatories published the MisAnthropic Statement, hosted at misanthropicstatement.com. The release came ahead of Anthropic’s initial public offering (IPO), the process of selling shares to the public for the first time. CAIS describes the coalition as AI safety researchers, open-source advocates, AI ethicists, and accelerationists. Named signatories include Steven Pinker, a Harvard psychologist, Oliver Hart, a Nobel laureate in economics, and Judea Pearl, a Turing Award winner. Max Tegmark, an MIT physicist, and three Fields Medal winners in mathematics also signed.
The statement’s core argument is about recursive self-improvement. That term describes AI systems that help build better versions of themselves. The text puts it this way: “Anthropic warns that recursively self-improving AI could concentrate power and escape human control, yet races to build it first.” The statement also says the company pitches the same capability to public-market investors as a core competitive advantage.
The second charge concerns branding. The statement says Anthropic markets itself as an “AI safety and research company” to recruit talent and pitch investors. It adds that the company is pitching investors as it tries to automate its own AI research. Karmactive has tracked the recent wave of AI agent security incidents, including disclosures from OpenAI and Anthropic.
The statement argues that public-market investors are being asked to buy into a safety story that conflicts with the company’s own record on self-improvement, safety pledges, and lobbying.
Who signed, and why it matters
The list spans several groups. Safety researchers such as Dan Hendrycks and Max Tegmark appear alongside Guillaume Verdon, who founded the effective accelerationist (e/acc) movement and goes by “Beff Jezos” online. Effective accelerationists favor rapid AI development with few limits. Open-source and tech figures include Thomas Wolf of Hugging Face, George Hotz of the tiny corp, and Tristan Harris of the Center for Humane Technology. Legal scholars Laurence Tribe, a professor emeritus, and Lawrence Lessig, both of Harvard Law School, also signed.
Hendrycks, executive director of CAIS, framed the alliance this way: “People in this coalition disagree on almost everything, yet we all recognize the danger of a company that convinces itself its own dominance is the greater good.” He added: “You cannot credibly claim to be safe while weakening your own previous safety commitments and asking the public for tens of billions of dollars to build recursively self-improving AI.”
As an inference from the signatory list, the dispute is less about whether AI is dangerous than about who should hold power over it.
The February policy change
In February 2026, Anthropic revised its Responsible Scaling Policy, its published framework for managing risks from more capable models, as the company explained in its own announcement. TIME reported on February 24, 2026 that the revised version dropped the 2023 pledge to never train an AI system unless it could guarantee in advance that its safety measures were adequate. Chief science officer Jared Kaplan told TIME that “it wouldn’t actually help anyone for us to stop training AI models,” and that unilateral commitments made little sense “if competitors are blazing ahead.”
The new version adds transparency commitments, including frontier safety roadmaps and regular risk reports. Anthropic says the revision is meant to “reinforce what has worked well to date” and to increase transparency and accountability. Chris Painter of the research group METR, who reviewed an early draft, told TIME that Anthropic “believes it needs to shift into triage mode with its safety plans.”
State AI bills and lobbying claims
The statement goes further. It says Anthropic has lobbied to weaken SB 1047, California’s frontier AI safety bill, the RAISE Act, and the AEGIS Act, and that it privately advocated for federal preemption of state AI laws. Preemption means a federal rule overrides state rules. Governor Gavin Newsom vetoed SB 1047 in 2024, in a veto message that set out his objections. This article could not independently confirm the AEGIS Act or its status.
The record has other sides. Anthropic first said it did not support SB 1047 in its original form, proposed amendments, and then gave qualified support to the amended bill. Anthropic backed California’s SB 53, a 2025 AI transparency bill. It also supported New York’s RAISE Act, which Governor Kathy Hochul signed in December 2025, per the state’s official announcement, according to comments Anthropic made to The New York Times. Politico reported on October 16, 2026 that Anthropic had privately pitched Congress a confidential draft federal bill on frontier AI safety. An Anthropic spokesperson told Politico that the draft reflects the proposals in the company’s Advanced AI Framework, which it has advocated for with federal and state lawmakers since June. The reported draft would preempt some state AI safety laws.
What happens next for investors and US policy
Karmactive has reported on Anthropic’s embedded evaluators and its safety framework. As a general rule, public companies must disclose material risks. As an inference, once Anthropic is listed, its safety commitments could become a legal question as well as a branding one.
For US readers, the next test is in Washington. Congress has debated federal preemption of state AI laws in several 2026 proposals, including the Great American AI Act, a bipartisan discussion draft released in June 2026. The statement accuses Anthropic of backing preemption privately. Readers outside the US cannot vote on those bills, but they can read the company’s public claims. Karmactive has also covered the voluntary safety pact that Trump and AI company chiefs signed in September 2026.
The MisAnthropic Statement challenges Anthropic’s safety record ahead of its IPO. Its claims cover recursive self-improvement, safety branding, a February 2026 policy change, and lobbying on state and federal AI rules. At the time of writing, this article found no public response from Anthropic to the statement. As an inference, Anthropic’s answer, and how investors weigh it, may shape how its safety promises are judged.