AI security risks escalate as researchers uncover WeChat vulnerabilities and large-scale fraud

A new report from Anthropic and findings from the California-based security firm Calif highlight growing vulnerabilities and malicious uses of artificial intelligence. Researchers at Calif developed a tool named ‘WeWorm’ capable of hijacking WeChat accounts to spread through contacts via automated calls, a capability experts describe as a ‘new nuclear weapon’ due to its potential to disrupt China’s critical digital infrastructure.

Anthropic’s latest threat intelligence report also details widespread AI-driven fraud. A Chinese app studio reportedly used the Claude model to power over 20 dating applications, where 75% of interactions with 25,000 users were conducted by AI characters rather than humans. These apps utilized a mix of AI and human part-timers to deceive users into purchasing virtual currency.

The reports further document various high-stakes AI misuses, including attempts by actors to use AI for biological weapon research, refining anti-torpedo systems, and conducting surveillance. These developments underscore an ‘AI security dilemma,’ where the rapid advancement of offensive AI capabilities may outpace defensive measures, prompting calls for international cooperation and crisis communication channels between major AI powers.