Anthropic has published a threat report stating that it has thwarted multiple state-affiliated malicious campaigns through its Claude AI model, including efforts by a group in northern Yemen developing missile guidance software, Russian and Chinese state-affiliated cyber espionage programs, Iranian covert disinformation campaigns, and a Chinese program aiming at Uyghur people in Syria. Although Anthropic has security measures in place, attackers managed to elude detection by dividing tasks among multiple sessions, resulting in the banning of the accounts in question, sharing of intelligence with partners, and increased platform monitoring. This follows growing pressure from whistleblowing researchers regarding existential risks of AI, as well as complicated ties with the United States, where Anthropic recently succeeded in its court battle against the Pentagon's supply chain blacklist and still uses its models for U.S. military activities.