Bhaskar Chakravorti urges Anthropic to file AI public risk statement — Fortune
In the United States, Fortune author Bhaskar Chakravorti proposed that Anthropic publish a separate statement on the public risks of its artificial intelligence products alongside its IPO documents. In his view, Anthropic's IPO documents describe threats to investors in detail but devote substantially less attention to potential consequences for people who are not the company's shareholders.
Risks for investors and society
According to the author, the risk section in Anthropic's filing contains 69,800 words — approximately 30% of the entire 240,000-word document. It mentions, among other things, catastrophic risks, worker displacement, uncontrolled autonomous agents, and the concentration of influence in the industry. At the same time, these issues are predominantly described as potential costs to Anthropic itself and its shareholders.
Chakravorti proposes introducing a Public Risk Statement — a statement that would explain how the company's products are used, who may be harmed, who is responsible for the consequences, and how affected people can seek compensation. He noted that Anthropic is incorporated in Delaware as a public benefit corporation and must combine the financial interests of shareholders with benefits for the broader public.
More current news is available on the UA.News Telegram channel Telegram.
AI oversight and potential compensation
The author believes that the risks of workers becoming overly dependent on AI and losing professional skills require separate disclosure. He notes that Anthropic's documents mention excessive trust in AI systems only once, while skill loss is not addressed. In his view, companies should explain who reviews AI decisions and whether they have sufficient time, data, and authority to override them.
As examples of costs resulting from algorithmic errors, Chakravorti cited Detroit's $300,000 payment to Robert Williams after his wrongful arrest due to a false match by a facial recognition system. He also mentioned tenant screening service SafeRent's $2.275 million settlement in a discriminatory screening case and Anthropic's $1.5 billion agreement on compensation for authors and publishers over pirated books.
According to the author's estimate, laws that could have been applied were already in place in about four out of five of the 103 documented cases of AI harm. He also cited Transparency Coalition data, according to which lawmakers in 30 U.S. states passed 115 laws related to artificial intelligence, nearly 58% more than a year earlier.
