智库警告：AI暂停将惠及主导企业，而非安全

Carter Institute warns: Government-imposed suspension of AI development could hinder competition and delay beneficial technologies

The liberal think tank Cato Institute has issued a warning that government-imposed suspension on artificial intelligence (AI) development may shield leading companies from competitive pressures and delay the advancement of beneficial technologies. Currently, legislators and AI industry executives are actively promoting the implementation of restrictions on AI.

Jennifer Huddleston, a technology policy scholar at the think tank, pointed out in a blog post posted on Monday that companies can address specific risks through voluntary security measures and shared standards without stalling the entire industry.

"A government-imposed suspension will raise multiple concerns and will likely fail to achieve its claimed safety improvements," she wrote, adding that government rules-making processes are slow and changes are even slower. "Regulatory frameworks based on current models may proactively hinder the development of future technologies, preventing innovations that could lead to better or safer outcomes."

Huddleston acknowledged the potential risks of AI, but warned that rules around leading developers could help them consolidate power and limit competition-a market monopoly that would make it more difficult for small companies to compete in the market. She also pointed out that if the United States slows down the pace of development while foreign competitors continue to advance, this could weaken cybersecurity defenses.

她的评论紧随参议员伯尼·桑德斯（Bernie Sanders）和众议员格雷格·卡萨尔（Greg Casar）提出的提案之后。该提案要求在美国联邦安全标准建立之前暂停先进AI的开发，并永久禁止人工超级智能的发展。与此同时，OpenAI首席执行官山姆·奥特曼（Sam Altman）也支持更慢速的开发进程，呼吁企业在等待他支持的联邦法规出台之前，加强自身的安全保障措施。

“许多此类呼吁都伴随着类似科幻电影中的末日场景描述，聚焦于最坏情况的发生，”赫德尔斯顿写道，“然而，出于恐惧而匆忙进行监管，其长期负面后果可能比它所声称要解决的风险更为严重。”

Block联合创始人杰克·多西：警惕以安全为名的市场壁垒

Block公司董事长兼联合创始人杰克·多西（Jack Dorsey）也在周一的X平台帖子中表达了类似的担忧，关注点同样集中在竞争问题上。

“领导机器智能发展的公司值得被倾听。他们拥有专业知识以及需要保护的商业利益，”多西写道，“围绕这些巨头资源制定的规则，可能会使它们成为唯一能够参与市场的主体。即使是真诚的安全关切，也可能形成进入壁垒。”

Dorsey said he supports an open source release model that allows researchers to conduct inspections, modifications and tests, and publishes evaluation results and known limitations so that outsiders can challenge developers 'security claims.

"I support a review mechanism. But I oppose industry-wide restrictions negotiated by today's industry leaders because they could exclude those who might expose failure cases or build alternatives,"he wrote." Maintaining a company's business advantage is not a security goal."

In Dorsey's post, he also expressed support for independent testing and public accountability enforcement mechanisms, including temporarily withholding model releases when evidence suggests that model releases will materially increase catastrophic risks and other narrower measures cannot adequately respond.

"Computing power itself is enough to trigger censorship without restricting development. Inspection does not amount to approval from regulators or competitors,"he said." I don't want to establish universal approval requirements or waiting periods. Any delay imposed on release for security reasons must be justified solely through a 'catastrophic risk exception' clause."

A recent analysis by the Atlantic Council concluded that competitive pressures require the implementation of binding AI security standards, and that mistrust between China and the United States complicates the conclusion of international agreements. In addition, the cooperation has raised antitrust concerns: OpenAI once asked lawmakers whether competitors could legally agree to jointly slow down development.