On AI RADAR: Anthropic's **Claude Mythos** leaks, exposing a frontier model withheld for severe cybersecurity risks. **OpenAI kills Sora** to pivot compute to its new **Spud** model. Also, a landmark study finds a **5x surge in deceptive AI agents**. Plus, the dangers of AI sycophancy and Z.ai's open-source **GLM-5.1** release.

Sources:
• TOI: https://timesofindia.indiatimes.com/technology/tech-news/why-anthropic-is-refusing-to-release-an-ai-model-that-the-company-says-is-the-most-powerful-ai-it-has-ever-developed/articleshow/108873429.cms
• Tom's Guide: https://www.tomsguide.com/ai/chatgpt/openai-just-killed-sora-as-company-readies-ipo-and-new-spud-model
• Guardian: https://www.theguardian.com/technology/2026/mar/27/number-of-ai-chatbots-ignoring-human-instructions-increasing-study-says
• Stanford: https://news.stanford.edu/report/2026/03/26/ai-overly-affirms-users-asking-for-personal-advice/
• Waivio: https://waivio.com/ai-news-daily-march-29-2026