July 27, 2026<\/a><\/p>\nAlthough China was not mentioned directly in the letter, the broader conversation has increasingly revolved around whether Chinese AI developers have gained an advantage by using intellectual property from American companies. One technique frequently cited is model distillation, where one AI system is extensively queried to reproduce the behaviour of another.<\/p>\n
Amodei argued that this issue should not be confused with the broader value of open-weight models. He described less capable open models as a public benefit because they can be freely used by researchers, software developers and businesses without depending entirely on commercial AI providers.<\/p>\n
Instead of targeting open models, he believes policymakers should focus on preventing advanced AI technology from being copied illegally and continue limiting China\u2019s access to cutting-edge chips used to train frontier systems. The United States has already warned that sanctions could follow if investigations find evidence of intellectual property theft involving American AI models.<\/p>\n
STORY CONTINUES BELOW THIS ADAmodei warns of military and biological AI risks<\/p>\n
While defending open-weight models in principle, Amodei reiterated that he remains deeply concerned about the trajectory of frontier AI.<\/p>\n
He argued that the greatest danger is not commercial use of open models but the possibility that authoritarian governments could develop AI systems capable of securing long-term military superiority or expanding domestic surveillance. China, he said, remains the most technologically capable authoritarian state, although it is not the only country that concerns him.<\/p>\n
Amodei also highlighted biological threats as an area demanding greater attention, suggesting that future AI systems could make it easier to design dangerous biological weapons. In such situations, he argued, open-weight models present additional challenges because they cannot easily be updated, withdrawn or fitted with new safeguards after release. He pointed to findings from the UK\u2019s AI Security Institute, which noted that publicly released model weights cannot realistically be recalled.<\/p>\n
That view differs from the position taken by many advocates of open-source AI, who argue that wider access to advanced models strengthens security by allowing independent researchers to identify weaknesses instead of concentrating power within a handful of companies.<\/p>\n
STORY CONTINUES BELOW THIS AD<\/p>\n
Alongside his warning, Amodei backed efforts to establish an international framework for testing the safety of the world\u2019s most powerful AI systems before deployment. He welcomed proposals that would apply to both open and closed frontier models while excluding less capable systems built by universities and smaller start-ups.<\/p>\n
\u201cI think this idea is actually close to a consensus,\u201d Amodei wrote, adding that recent proposals from both industry and the US government indicate growing support for independent evaluations.<\/p>\n
He also argued that any meaningful safety framework would ultimately require international participation, including from China. \u201cNote that to be effective,\u201d he wrote, \u201ctesting would need to be global, which means even the CCP would need to be on board.\u201d<\/p>\n
Despite mounting geopolitical tensions, Amodei suggested that preventing AI from being used to create biological weapons could remain one of the few areas where the world\u2019s leading AI powers still have a shared interest in cooperation.<\/p>\n","protected":false},"excerpt":{"rendered":"Artificial intelligence companies are increasingly split over how governments should regulate the world\u2019s most advanced models, but Anthropic…\n","protected":false},"author":2,"featured_media":121027,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[4788,4989,1710,53,11900,3261,2407,47057],"class_list":["post-121026","post","type-post","status-publish","format-standard","has-post-thumbnail","category-anthropic","tag-ai-regulation","tag-ai-safety","tag-ai-security","tag-anthropic","tag-china-ai","tag-dario-amodei","tag-frontier-ai","tag-open-weight-models"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/121026","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=121026"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/121026\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/121027"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=121026"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=121026"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=121026"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}