Anthropic launches Claude Agents with new security controls in London, as AI coding tools reshape software engineering

Anthropic rolled out new Claude Agents features at its first dedicated developer event in Europe, including sandboxes that let companies run agents on their own infrastructure and 'MCP tunnels' that enable agents to reach internal systems without touching the public internet.

Software engineering is undergoing a shift where AI handles the writing of code while humans concentrate on higher-level decisions and system oversight. Some engineers reported using Claude Code daily, though enterprises showed mixed adoption, citing uncertainty about what to automate, where humans are needed, and what can be safely handed to AI—concerns that intensify in regulated sectors like health care or banking.

Anthropic cofounder Jack Clark spoke at Oxford University about the philosophical implications of the company's technology, touching on the potential for AI-aided Nobel Prize-winning discoveries and the use of Claude for mental health support. He also warned of a 'non-zero chance' of AI posing existential risk and noted that geopolitical and commercial competition make slower development unlikely even if it might be safer.