AI’s public trust problem
Major AI firms are reportedly funding policy papers and think-tanks to influence public debate as polling shows rising disapproval of AI. (theguardian.com) At the same time, rights-holders are pushing that models trained on copyrighted works should compensate creators rather than rely on post-hoc negotiation, and concerns about cultural distortion in underrepresented markets are being raised. ( )
Artificial intelligence companies are moving deeper into policy and media just as public trust in the technology is fraying. (anthropic.com; openai.com; edelman.com) OpenAI published a 13-page paper on April 6 calling for “people-first” industrial policy, including public wealth funds, automatic safety-net triggers and four-day workweek pilots as more capable systems spread through the economy. (openai.com; cdn.openai.com) Four days earlier, on April 2, OpenAI said it had acquired TBPN, a technology talk show and media brand, and said the deal would help it “accelerate the global conversation around AI.” TBPN will sit inside OpenAI’s strategy organization under Chris Lehane, according to the company. (openai.com; cnbc.com) Anthropic announced the Anthropic Institute on March 11 and said the new unit would study the economic, legal and governance effects of more powerful models. The company also said it was expanding its public policy team and opening its first Washington office this spring. (anthropic.com; cio.com) The trust problem is showing up in polling. Edelman’s 2025 flash poll of 5,000 people across Brazil, China, Germany, the United Kingdom and the United States found that rejection of artificial intelligence outweighed enthusiasm across the surveyed markets. (edelman.com; menafn.com) At the same time, the copyright fight is shifting from lawsuits over scraping to demands for payment at the training stage. A Times Now report published April 11 said publishers and other rights-holders are arguing that companies should compensate creators before using copyrighted work to build models, not bargain only after products launch. (timesnownews.com) That argument has been building in the news industry for months. The Digital News Publishers Association in India said in July 2025 that using publishers’ content for training and generative products without consent is copyright infringement and called for a system of fair compensation. (timesnownews.com) The backlash is not only about money. Kuensel, Bhutan’s national newspaper, reported on April 11 that officials and researchers there fear global training datasets can flatten or misstate Bhutanese dress, rituals and religious practice when artificial intelligence tools generate local-looking content. (kuenselonline.com) Bhutan’s Department of Culture and Dzongkha Development told Kuensel that artificial intelligence can help preserve language and widen access to knowledge, but warned that unregulated output can also misrepresent traditions. That leaves governments, publishers and model makers arguing over the same question from different angles: who gets to shape the record that these systems learn from. (kuenselonline.com) The next phase of the debate is moving from laboratory claims to rules about consent, compensation and representation. As companies publish policy blueprints and build institutions around the issue, critics are pressing for enforceable terms instead of voluntary promises. (openai.com; anthropic.com; timesnownews.com)
