# News: March 27, 2026

> 13 curated links on AI agents, Python, security, Google Cloud, and engineering, selected by Médéric Hurier.

- Author: [Médéric Hurier (Fmind)](https://www.fmind.dev/)
- Published: 2026-03-27
- URL: https://www.fmind.dev/news/2026-03-27/

- [Anthropic reveals "Claude Mythos" and new "Capybara" tier](<https://fortune.com/2026/03/26/anthropic-says-testing-mythos-powerful-new-ai-model-after-data-leak-reveals-its-existence-step-change-in-capabilities/>): A major data leak revealed Anthropic's unreleased model "Claude Mythos," a step-change in reasoning and coding that the company claims poses unprecedented cybersecurity risks.
- [Google launches Gemini 3.1 Flash Live and Live API](<https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-flash-live/>): The new Gemini 3.1 Flash Live model offers enhanced latency and reliability for real-time, multimodal voice-first agents and is now available in preview.
- [Sora shut down as OpenAI pivots to ChatGPT "super-app" strategy](<https://www.theneurondaily.com/p/sora-lasted-6-months-disney-s-1b-deal-lasted-3>): OpenAI is reportedly shutting down its Sora video app and ending its $1B Disney deal to focus compute resources on ChatGPT and its upcoming model "Spud."
- [Cisco enters agentic AI security with DefenseClaw](<https://www.networkworld.com/article/4148823/cisco-goes-all-in-on-agentic-ai-security.html>): Cisco announced DefenseClaw, an open-source framework for securing AI agents, alongside new identity management tools for agentic workflows.
- [Wiz introduces specialized AI agents and workflows for cloud security](<https://www.wiz.io/blog/introducing-wiz-agents>): Wiz has launched specialized Red, Blue, and Green AI agents designed to automate threat detection, investigation, and remediation at scale.
- [Google DeepMind releases evaluation framework for harmful AI manipulation](<https://deepmind.google/blog/protecting-people-from-harmful-manipulation/>): New research and an open-source evaluation toolkit from DeepMind aim to measure and mitigate how AI models could be misused to harmfully manipulate human beliefs.
- [Building Voice Agents with Gemini 3](<https://www.youtube.com/watch?v=XV5bhkDpL7U&si=maemWuVFad2lqobh>): Google for Developers provides a technical deep dive into building low-latency, real-time voice interactions using the new Gemini 3.1 Live API.
- [How to add short-term memory to your AI agent](<https://www.youtube.com/watch?v=vfVcDUCucSs&si=J9g3DYu9Ai9kC6Zq&themeRefresh=1>): This tutorial explains how to manage sessions and state to give AI agents persistent short-term memory for better task execution.
- [How to Vibe Code a Website in 10 Minutes](<https://www.youtube.com/watch?v=xCkIxdJZ8fU&si=nUsJRMV7r_HwJucM>): A demonstration of "vibe coding" using Google AI Studio to rapidly build and deploy a functional website via natural language prompts.
- [ARC-AGI 3 benchmarks show models struggling with general intelligence](<https://x.com/ai_for_success/status/2036975819336409370>): Initial benchmarks for the ARC-AGI 3 challenge reveal that current top AI models are still failing to cross even the 0.5% mark in human-level reasoning tasks.
- [Top open-source AI projects to watch](<https://www.youtube.com/watch?v=sXVbWkoCVaA&si=fs9zh4SQ7FF92SXc>): Matthew Berman reviews the most impactful open-source AI projects currently available for developers and researchers.
- [Google Search Live expands to 200+ countries](<https://blog.google/products-and-platforms/products/search/search-live-global-expansion/>): Powered by Gemini 3.1 Flash Live, Search Live now supports real-time, multimodal conversations in over 90 languages for users globally.
- [Google introduces memory and chat history import for Gemini](<https://blog.google/innovation-and-ai/products/gemini-app/switch-to-gemini-app/>): Users can now import their preferences and full chat histories from other AI services into Gemini to maintain personal context across platforms.