Anthropic has disrupted several attempts to misuse its Claude artificial intelligence models for potentially harmful activities, including biological research, cyber operations, weapons development and surveillance.
The AI company said its latest threat intelligence findings show that increasingly capable AI systems are being targeted by actors seeking to automate sophisticated operations and gain assistance with sensitive technical tasks.
Anthropic said it identified attempts to use Claude in research involving biological threats, raising concerns about the potential role of advanced AI in accelerating dangerous scientific activities. The company has introduced additional safeguards designed to prevent its models from providing assistance that could facilitate the development of biological weapons.
The report also highlighted activity linked to Russia and China. Anthropic said Russian-linked actors used Claude in operations involving cyber activities and military technology, while Chinese-linked groups attempted to use the model for activities including intelligence gathering, electronic warfare and weapons-related research.
Anthropic said it blocked the accounts associated with the activities and shared relevant information with authorities and other organizations to help counter emerging AI-enabled threats.
The incidents underscore a growing challenge for AI companies as their models become more capable. While advanced AI can assist researchers, businesses and developers with legitimate work, the same capabilities can potentially be exploited for cyberattacks, military applications and other harmful activities.
Anthropic said it will continue strengthening its safeguards and monitoring systems as it works to prevent its technology from being used for malicious purposes.
Leave a comment