Dario Amodei, CEO of Anthropic, argues that societal resistance to artificial intelligence stems from a deeper crisis of trust in companies, governments and the tech industry, rather than simply from poor communication by AI leaders. The public debate between Amodei and investor Gavin Baker underscores that current conflicts around AI involve not only what models can do but whether society believes tech firms are acting in the public interest.
Amodei wrote on X that ordinary people do not trust corporations, governments or the tech industry, and that the roots of this distrust reach back decades; AI is only the latest chapter. This framing shifts attention away from communications tactics toward the social embeddedness of the technology sector.
Why has Amodei been criticized? Baker’s point and Anthropic’s image
TechCrunch covered the public exchange in which Gavin Baker accused Amodei on the All-In podcast and on X of painting an overly pessimistic picture of AI’s future and thereby feeding rising opposition in the United States, including protests against data centres. In parts of Silicon Valley and among some investors, Anthropic has been viewed as a leading voice of a "safety-first", risk-focused AI narrative. Baker argued that a CEO of a strategically important, fast-growing AI company should promote a more positive industry narrative.
Amodei rejected the charge of disproportionate negativity and pointed to his 2024 essay, Machines of Loving Grace, in which he outlined ways advanced AI could benefit healthcare, mental health, poverty reduction, governance and work.
Public opinion: usage rises, optimism does not necessarily follow
Several 2026 public surveys support the characterization of a trust problem. The Pew Research Center’s June 2026 survey found that 49 percent of U.S. adults had used an AI chatbot, up from 33 percent in 2024. Yet 40 percent of respondents said AI would have a negative effect on American society over the next 20 years, while only 16 percent expected a positive effect.
Gallup’s 2026 figures show that 39 percent of Americans believe AI does more harm than good, up from 31 percent in 2025. The Stanford HAI 2026 AI Index paints a more nuanced global picture: the share of people who see more benefits than harms from AI products rose internationally, but the portion of respondents who feel anxious or worried about AI remained high. Together, these results suggest that broader use of AI does not automatically translate into greater public trust.
Data centres: the visible cost of AI
One of the most tangible manifestations of distrust is opposition to data centre projects. Reuters reported that by July 2026 at least 125 U.S. locations had seen protests against data-centre developments. A Reuters/Ipsos poll cited by the agency found only 14 percent of Americans would support a data centre serving AI projects being built in their own community.
Local resistance is often driven by environmental and infrastructure concerns: communities worry about increased local electricity and water demand, grid strain, and the fact that such projects may create relatively few local jobs. Data Center Watch’s quarterly summary indicated that in the first three months of 2026 at least 75 data-centre projects were delayed or blocked by local opposition, representing roughly $130 billion in project value. The IEA’s Energy and AI report notes that AI deployment and data-centre energy demand are linked, and that AI could also transform energy systems—so the impacts on local grids and communities are politically salient.
Regulation: finding a hard balance
Regulatory questions form another key strand of the debate. A common Silicon Valley concern, expressed by Baker, is that strict regulation ultimately benefits incumbent large firms, which can absorb compliance costs while smaller competitors are squeezed out. Anthropic advocates for targeted, risk-based regulation focused on frontier models as a counterproposal.
California illustrates the difficulty of striking the right balance: Governor Gavin Newsom vetoed the controversial SB 1047 in September 2024 on the grounds it did not adequately consider use context and whether systems operated in high-risk settings, but a year later he signed SB 53, the Transparency in Frontier Artificial Intelligence Act, which imposes a transparency-focused framework on frontier AI developers. The European Commission’s General-Purpose AI Code of Practice similarly aims to offer voluntary guidance to providers of general-purpose models in areas such as transparency, copyright compliance and safety risk management.
Open-weight models: a partial fix, not a cure
The question of open-weight models adds a third layer to the debate. Publicly available model weights can support research, competition and a broader developer ecosystem, but they also raise misuse risks, particularly for cybersecurity or biological applications. Amodei argues that open weights can help mitigate concentration, yet cannot eliminate it because powerful models require not only algorithmic know-how but vast compute, chip supply, data-centre infrastructure and capital.
Proponents of greater openness counter that closed, opaque corporate systems reinforce distrust by limiting the ability of researchers, the public and smaller players to scrutinize the strongest AI systems. A Nature analysis warns that "open" is often a misleading label in AI: a system may be open in some respects while key components remain closed. Thus the open-versus-closed model debate is not simply freedom versus safety.
Implications: social limits as well as technical ones
Amodei’s intervention matters beyond his personal defence of communications strategy. His point is that even if technical progress accelerates, the AI industry faces social constraints. Trust deficits appear in public opinion polls, local protests against data centres, workplace anxieties and regulatory battles. That means AI companies’ room to operate is governed not only by model capabilities, chips and capital, but also by how legitimate society considers the technology’s spread.
Amodei’s prescription is stark: trust will not be regained through PR campaigns but by delivering demonstrable social benefits. If AI produces tangible breakthroughs in medicine, science or productivity, public acceptance could grow. If, however, communities primarily experience higher energy bills, environmentally contentious data-centre projects, job insecurity and opaque corporate decision-making, opposition to AI will persist as a durable political and economic risk.
