Evaluation reveals AI confidence peaks on errors  |  New AI model uncovers major Cursor flaw  |  DeepSeek Harness Debuts Open-Source Rival to Claude Code  |  AI Agent Identity Solved Containment Proves Difficult  |  AI video made in under 7 seconds  |  OpenAI unveils GPT‑5.6 Cyber with 95% success  |  Stanford AI agents design drug built by Merck  |  Four AI Agents Outperform Claude Opus 4.8