How Anthropic may have talked itself into an AI export ban

Anthropic has consistently highlighted the dangers of advanced AI more often than its competitor, OpenAI, according to a Financial Times analysis. This emphasis on risk, regulation, and restrictions in Anthropic's communications has drawn criticism, with some suggesting it inadvertently contributed to a recent US ban on foreign access to its new AI models, Mythos and Fable. In contrast, OpenAI’s communications showed a significantly lower focus on such concerns.

The US decision to bar foreign nationals from using Anthropic’s latest models has sparked debate. Technologists like Yann LeCun, Meta’s former chief AI scientist, argue that Anthropic’s repeated warnings about AI’s societal risks, particularly regarding Mythos, have "paid off" in the form of these export restrictions. This situation has alarmed stakeholders in Europe and Silicon Valley, who fear broader restrictions on non-US access to cutting-edge AI models from the current administration.

Anthropic has long positioned itself as a conscientious leader in the AI industry, frequently publishing research on potential harms and advocating for increased government intervention. Before the export ban, CEO Dario Amodei expressed concerns that regulators were moving too slowly on AI, highlighting the "very real risks" of Mythos to cybersecurity, critical infrastructure, and national security.

However, the company’s language has softened since 2023, with the frequency of risk-related terms diminishing as its AI tools gained popularity. Despite this, Anthropic’s public communications, while generally positive, maintain a more cautious tone compared to OpenAI. The company had initially limited access to Mythos due to safety concerns, working with government officials on a controlled rollout before wider release.

The export ban follows earlier clashes between Anthropic and government figures concerning the use of its technology in domestic surveillance and autonomous weapons. The Pentagon previously designated Anthropic as a supply-chain risk to national security. Public opinion polls suggest that a majority of US citizens support effective regulation, even if it slows technological advancements.

International leaders are also weighing in, with French President Emmanuel Macron calling for stronger AI regulation and warning against a lack of cooperation among democratic nations. This complex scenario underscores the challenges of overseeing increasingly powerful AI models and the potential for unintended consequences stemming from public and corporate discourse on AI risks.