AI Round-up: 10 August 2026

CNBC traced rogue-agent security tests involving OpenAI, Anthropic and Meta models to Irregular, a small Israeli security company, while Reuters reported that a North Korean hacking group is building AI tools for cyberattacks. Bloomberg and creator coverage kept the consequences of agent breaches and advanced cyber capabilities in focus.

Anthropic is making Claude Code's auto mode the default, and community discussion centred on agent-to-agent messaging, model choice and preserving context across coding tools. Together AI detailed Cursor's low-latency inference stack, Databricks reported improved accuracy and lower costs from specialised data agents, and AI activity continued to support market forecasts and record dealmaking.