Anthropic has received approval to restore access to its most advanced artificial intelligence model, Mythos 5, for a limited number of approved U.S. organisations following a temporary restriction linked to government concerns over AI security risks.
The company said the model will be made available to selected entities involved in areas such as critical infrastructure and national security operations. More than 100 organisations, including major companies, are expected to receive access under the updated arrangement, according to people familiar with the process.
The move comes after the U.S. government introduced restrictions on certain advanced AI models over concerns that powerful systems could potentially be misused, including for cyber-related threats. Anthropic had previously limited access to Mythos 5 and another model, Fable 5, after the restrictions were announced.
Anthropic said it is working with officials to restore access for approved users while continuing discussions about broader availability. The company has not disclosed the full details of the security requirements attached to the approval.
The government’s review process has sparked debate among technology and civil rights groups. Critics argue that the criteria for selecting approved organisations lack transparency, while supporters say stronger oversight is needed as AI systems become more capable.
Security experts have warned that advanced AI models could create new risks if misused, particularly in areas involving cyberattacks and systems that support essential services such as finance and utilities.
Officials said Anthropic has made progress in addressing concerns around responsible deployment. Under the updated rules, some approved organisations may also receive more flexibility regarding access for certain international employees, while restrictions remain for organisations outside the approved list.
The decision reflects the broader challenge facing governments and AI companies: finding a balance between encouraging innovation and preventing misuse of increasingly powerful technologies.
Analysts say the approval provides a temporary pathway for wider AI deployment but leaves unresolved questions about how future frontier AI systems should be regulated, who should have access, and how much control governments should have over emerging technologies.
