The Core Debate: Where Is the AI Backlash Coming From?

Anthropic CEO Dario Amodei recently responded to outside criticism, arguing that the backlash against AI in American society doesn’t primarily stem from him or other AI leaders being overly alarmist about risks, but rather reflects a deeper crisis of trust.

The debate was ignited by investor Gavin Baker, who stated on the “All-In” podcast and on X that Amodei and others’ persistent warnings about AI dangers have fueled public and policy resistance to AI in the United States, particularly visible in opposition to data center construction. Baker also argued that Anthropic has “lost the case” on regulation, pointing out that the company had supported certain regulatory measures, including a California bill requiring large AI companies to increase transparency. Baker’s suggestion was that, as head of a major AI company, Amodei should be more vocal in advocating for the industry.

Amodei disagrees. He says characterizing his statements as “overly negative” is inaccurate; in his view, his writing has largely maintained a balance between risk and benefit. He also noted that he wrote the essay Machines of Loving Grace precisely because he felt the AI industry had not adequately depicted how this technology could fundamentally improve the world.

Behind the “Negative Perception” Lies a Deep-Rooted Lack of Trust

Amodei acknowledges that the public does hold negative views of AI, and that this is a “big problem.” But he pushes back against attributing this sentiment mainly to AI company leaders talking about risk. His judgment is that ordinary people simply don’t trust corporations, government, or the tech industry—they always suspect these institutions are designing new ways to harm their interests.

In other words, AI is just the latest outlet for this distrust. In recent years, “trust” has similarly been a recurring word in coverage of OpenAI CEO Sam Altman; now, this issue has clearly extended to other AI companies like Anthropic as well. Amodei sees this crisis not as something that developed overnight, but as the result of decades of accumulated social relations.

He also argued that the most valid criticism of AI companies, including Anthropic, is not that they’re “too pessimistic in their messaging” or that their marketing is poor, but that they have not yet delivered on those sweeping promises to benefit the world. What will actually shift public attitude is not repeatedly promising that AI will cure cancer—it’s actually doing it. In his view, claims like “AI will cure cancer” have become little more than clichés; only real results can be persuasive.

The Regulatory Debate: Openness, Concentration, and “Rules of the Road”


On the question of regulation, Amodei also pushed back against Baker’s framing. Baker described the issue as a false dichotomy: either distribute AI capabilities widely with no regulation, or concentrate the technology in the hands of a few large companies through regulation. Amodei argues this is a “false choice.”

He acknowledges that a simplified logic often circulates within Silicon Valley: regulation equals regulatory capture, and regulatory capture equals concentrated power. Regulatory capture refers to the phenomenon where regulatory frameworks are exploited by large companies to protect entrenched incumbents and raise barriers for new entrants. But Amodei believes the reality isn’t always that straightforward. Many people outside Silicon Valley view regulation as a tool to constrain corporate power and protect ordinary citizens. He doesn’t fully endorse that perspective either, but it helps explain why Anthropic has been so cautious in putting forward policy recommendations.

He emphasized that the policy Anthropic is trying to design would place greater constraints on frontier AI companies and slow them down, while putting smaller competitors in a more favorable position. By “frontier AI companies,” he means those with the capability to develop the most advanced large models, as well as substantial computing power and chip resources.

Key positions discussed in the article include:

• Public negativity toward AI is a real problem, but its root cause is a lack of trust;
• AI companies haven’t yet delivered on their promises to benefit the world—this is the most compelling criticism;
• Regulation doesn’t necessarily mean concentrated power; what matters is how the rules are designed;
• Open-weight models help disperse capability, but they’re not a complete solution to the concentration problem.

Open Weights Isn’t a Panacea

Amodei also spoke about the structure of power in AI. He believes AI is structurally inclined toward concentrating power, because training and deploying the most powerful models requires massive amounts of computing power, chips, and engineering resources. Open-weight models can indeed alleviate the trend toward concentration to some extent. Open weights generally means that model parameters are available for external download or use, allowing developers to deploy, fine-tune, or conduct research on top of them.

But in Amodei’s view, open weights is not a sufficient answer, because it may simply transfer power from model companies to whoever possesses the most computing power and chips. That is to say, even with more open models, the entities most capable of running, modifying, and commercializing them at scale are likely still the best-resourced organizations.

Therefore, he advocates for establishing appropriate “rules of the road”: addressing AI risks in cybersecurity, biosecurity, and alignment, while institutionally constraining the power of frontier AI companies, preserving space for open-weight models, and crafting specific rules for the risks they introduce. Alignment is a commonly used concept in AI safety, referring to ensuring that model behavior remains consistent with human intent and societal goals.

Industry Perspective: AI Narratives Are Shifting from Vision to Delivery

The significance of this debate lies in what it reveals: the public narrative of the AI industry is entering a new phase. The initial wave of enthusiasm relied on vision—more efficient work, stronger research capabilities, more accessible intelligent services. But as data centers, energy consumption, employment shifts, platform power, and security risks simultaneously enter public awareness, optimism alone is no longer enough to win societal support.

Amodei’s response is neither to deny the risks nor to simply ask the industry to “tell a better story.” Instead, he redirects focus toward delivery and institutional design: AI companies need to produce tangible results that demonstrate their public value, while accepting rules that can constrain their own power. Moving forward, the challenge for AI companies won’t be limited to competing on model performance—it will also involve building credible balance across transparency, openness, safety governance, and business expansion. Whoever can translate promises into verifiable social benefit will be the one to win greater legitimacy in the next round of AI competition.