As a result, non-U.S. users—including those in India—can no longer access these models. U.S. officials cited risks related to cybersecurity and critical infrastructure protection. If confirmed, this would mark the first known instance of a government restricting access to AI software itself rather than focusing on hardware exports such as chips or semiconductors.
Concerns Over “Jailbreak” Exploits and Cyberattacks
The U.S. government is reportedly concerned that advanced AI systems could be “jailbroken,” meaning users might bypass built-in safety restrictions and gain access to capabilities that were intended to remain unavailable.
Officials fear that such exploits could allow malicious actors to:
- Identify vulnerabilities in computer systems
- Develop sophisticated hacking techniques
- Access sensitive government data
- Target banking systems and critical infrastructure through large-scale cyberattacks
According to reports, researchers were able to use specially crafted prompts during testing to uncover certain software vulnerabilities through Anthropic’s models. The findings were allegedly shared with the U.S. Department of Commerce, prompting the government’s decision to impose restrictions.
Anthropic Says the Ban Is Based on a Misunderstanding
Anthropic has criticized the decision, arguing that the reported vulnerabilities were limited in scope and not unique to its models.
The company stated that:
- Similar vulnerabilities can be identified by several publicly available AI systems.
- A worldwide restriction is disproportionate to the actual risk involved.
- The models underwent extensive testing before launch with U.S. government agencies and the UK’s AI Safety Institute.
- No major security concerns were identified during those evaluations.
Anthropic said it is currently in discussions with government officials and hopes to restore access as soon as possible. In the meantime, users can continue using the company’s older AI models.
Ongoing Tensions With the Trump Administration
The reported restriction comes amid broader tensions between Anthropic and the Trump administration.
According to reports, Anthropic had previously declined requests to allow its AI models to be used for:
- Domestic surveillance programs
- Fully autonomous weapons systems
Following those disagreements, the Pentagon reportedly classified the company as a potential “supply-chain risk.”
The controversy arrives as Anthropic is preparing for a potential initial public offering (IPO), with some estimates placing the company’s valuation near $1 trillion.
Zoho Founder Sridhar Vembu: “Globalization Is Over”
The decision has sparked debate across the technology industry.
Sridhar Vembu, founder of Zoho, described the move as a major wake-up call for countries such as India. He argued that nations can no longer assume permanent access to foreign technology platforms and should accelerate investment in domestic AI capabilities.
Vembu emphasized the need for:
- Sovereign AI infrastructure
- Indigenous AI research and development
- Open-source AI ecosystems
- Reduced dependence on foreign technology providers
He warned that any country could potentially lose access to critical technologies if geopolitical priorities change.
Knowledge Box: What Are “Jailbreak” and “Sovereign AI”?
Jailbreak
AI models are designed with safety guardrails that prevent them from providing instructions for harmful activities such as cyberattacks, illegal operations, or weapon creation.
A jailbreak occurs when a user successfully bypasses those restrictions through specially designed prompts or techniques, causing the AI system to reveal information it would normally refuse to provide.
Sovereign AI
Sovereign AI refers to an AI ecosystem developed and controlled by a nation using its own:
- Data resources
- Computing infrastructure
- Research institutions
- Scientists and engineers
Such systems operate independently of foreign governments or companies, giving countries greater control over strategic AI capabilities and digital infrastructure.
