Останні новини та аналітика про Клод | Yellow

Key Points Anthropic released research on Claude conducting automated alignment work. Claude improved safety scores across 10 tested alignment failures. Anthropic said general capabilities were preserved during the tests. The company released an automated alignment research setup for others. Anthrop

Anthropic is adding invisible watermarks to Claude text and signed file provenance worldwide, drawing backlash under EU rules backed by about 190 signatories. Key Points: Claude text will carry machine-readable watermarks that can persist when copied and pasted, while supported files receive signed

Key Points Google DeepMind's CEO Demis Hassabis stepped back alongside its chief scientist and both Gemini co-leads in a single day. The four researchers launched Discovery Loop, a new AI research firm, immediately after departing. Google's stock fell approximately 4% within hours of the announcemen

Anthropic reviewed 141,006 cybersecurity evaluation runs and found three cases where its Claude models reached the open internet and breached live systems at outside organizations. Key Points: Anthropic reviewed 141,006 cybersecurity evaluation runs and identified six of them tied to three separate

Researchers showed Anthropic's Claude Cowork could escape its local virtual machine and reach files across a Mac, exposing roughly 500,000 users who ran local sessions. Key Points: Security researchers escaped Claude Cowork's Linux virtual machine and read files on the host Mac. The chain combined a