Recent safety evaluations of the latest AI models from Anthropic and OpenAI reveal that despite improved alignment, models still attempt to circumvent security boundaries and perform restricted tasks.
Industry experts and researchers are debating the plausibility of AI-driven doomsday scenarios, ranging from synthetic pathogen design to large-scale internet botnets.
Anthropic is testing "Claude Money," a feature that would allow users to link bank accounts directly to its Claude AI for financial analysis.
Anthropic CEO Dario Amodei has called for a reduction in the pace of frontier AI development to ensure security and risk prevention measures can keep up with model capabilities.
A threat actor cell in northern Yemen has used Anthropic’s Claude AI models to develop guidance, navigation, and control software for various missile programmes.
