DeepSeek engineer criticizes Anthropic, OpenAI over AI safety calls

OpenAI: OpenAI is a US artificial intelligence research and deployment company known for its closed-source large language models and generative AI systems. It has supported industry calls to slow the pace of frontier AI development amid safety concerns. The company is positioned as a key player in the debate over responsible AI progress versus competitive pressures from China. DeepSeek: DeepSeek is a Chinese artificial intelligence company focused on developing advanced models with an emphasis on open-source approaches. Engineer Liu Shengyu, who contributed to its V4.1 models, publicly criticized proprietary US developers and defended the company's open-source stance in the context of global AI governance debates. The company operates amid intensifying US-China technological competition. Anthropic: Anthropic is a US-based artificial intelligence company developing frontier models with a focus on safety and alignment. Its CEO Dario Amodei recently called for frontier AI firms to pace development due to rapid capability advances outstripping safeguards, while also highlighting risks from Chinese AI leadership. The firm is preparing for a public listing later this year. Liu Shengyu: Liu Shengyu is a DeepSeek engineer who worked on the company's V4.1 AI models. He publicly warned against concentrating advanced AI capabilities in proprietary US firms like Anthropic and OpenAI, drawing historical comparisons to underscore risks to social mobility and global access. His statements reflect broader Chinese skepticism toward US-led pacing initiatives. Dario Amodei: Dario Amodei is the CEO of Anthropic and a prominent voice in AI safety discussions. He urged frontier AI companies to pace development while warning that Chinese leadership in the field would pose grave dangers, advocating for continued US export restrictions on advanced chips. His comments have sparked international debate and responses from both industry and government figures. • Liu Shengyu, a machine-learning systems engineer behind DeepSeek’s V4.1 models, published a blog and social posts arguing that Anthropic and OpenAI cannot be trusted to keep frontier AI open and affordable, in contrast to DeepSeek’s open-source stance. • He framed Anthropic controlling the most advanced AI or AGI as a geopolitical-scale danger, saying consequences would be comparable to Adolf Hitler or Nazi Germany acquiring nuclear/atomic bomb technology before the Allies. • The intervention comes after Anthropic CEO Dario Amodei’s essay “We Must Pace the Frontier,” backed by OpenAI’s Sam Altman and others, which calls for slowing frontier model capability gains and embedding third‑party safety evaluators. • Liu’s comments tap into a broader Chinese industry backlash to US “pacing” proposals, which are seen by some as a way for US labs to lock in a lead by restricting rapid development and centralizing resources under proprietary Western companies. • He contrasts a future where AI is widely shared and supports more egalitarian outcomes with a Cyberpunk‑style world where a handful of companies hoard advanced models, warning that US-led closed development could entrench global class and power stratification. • The episode underscores intensifying US–China tensions over AI governance, access to cutting-edge models, and control of AGI, raising the stakes for prediction markets tracking AI regulation, openness vs. closed-source strategies, and which bloc ultimately dominates frontier AI.