Anthropic’s most advanced AI models blocked worldwide:US cites cyberattack risks

AI company Anthropic has suspended global access to its most advanced AI models, Claude Fable 5 and Mythos 5. The move follows a US government order restricting the use of the models by foreign nationals. As a result, non-US citizens, including Indians, can no longer access these models. US officials cited national security concerns and the potential risk of cyberattacks as the reason for the decision. It is the first known case of a government restricting access to AI software itself rather than hardware or computer chips. Concerns over AI jailbreaks and cyberattacks The US government fears that the advanced models could be used to bypass built-in security protections, a process known as a “jailbreak”. Officials argue that if these safeguards are bypassed, the models could potentially be used to identify weaknesses in computer systems, access sensitive government data, or support cyberattacks on banks and critical infrastructure. According to a report by The Wall Street Journal, researchers at Amazon were able to extract information about software vulnerabilities from Anthropic’s model using specially designed prompts during testing. The findings were reportedly shared with the US Department of Commerce, which then moved to restrict access. Anthropic says ban is based on a misunderstanding Anthropic has criticised the decision, saying it is based on a misunderstanding of the risks involved. The company argues that the vulnerabilities discovered were limited and that similar issues can be found in other publicly available AI models. It says a worldwide restriction on foreign users is therefore disproportionate. Anthropic also noted that the models underwent weeks of testing with US government agencies and the UK’s AI Safety Institute before launch, and no major security concerns were identified. The company is now in discussions with the government to restore access. Until then, users can continue using its older AI models. Dispute with Trump administration predates ban Anthropic has reportedly been at odds with the Trump administration for several months. The company is said to have refused requests to allow US security agencies to use its AI systems for domestic surveillance and fully autonomous weapons programmes. Following this, the Pentagon reportedly placed Anthropic on a supply-chain risk list. The dispute comes as the company prepares for a potential stock market listing, with reports valuing the business at around $1 trillion. Sridhar Vembu: ‘Globalisation is over’ The decision has sparked debate across the technology sector. Until now, US restrictions have largely focused on preventing advanced AI chips and semiconductor technology from reaching certain countries, particularly China. The move to restrict software access has therefore surprised many observers. Indian tech entrepreneur Sridhar Vembu described the decision as a major wake-up call for India, arguing that globalisation is effectively over. He said India should accelerate the development of its own sovereign AI capabilities and support open-source AI models rather than relying heavily on foreign technology providers. According to Vembu, any country could potentially restrict access to its technology in the future. Knowledge box: What are ‘Jailbreak’ and ‘Sovereign AI’? Jailbreak:
AI models contain safety rules designed to prevent them from providing harmful information, such as instructions for making explosives, carrying out cyberattacks or engaging in illegal activities. When a user or hacker finds ways to bypass these safeguards and obtain restricted information, it is known as a jailbreak. Sovereign AI:
A sovereign AI system is developed and operated by a country using its own data, infrastructure and experts. It is designed to function independently without relying on foreign governments or overseas technology companies.