Latent Space reports that SpaceXAI has launched Grok 4.5, positioning it as an Opus-class model and the first major model release after the Cursor acquisition. The release matters for teams tracking frontier coding and agent models because price, performance, and IDE distribution are now becoming direct competitive levers.
TechJuice reports that PIMS Islamabad has launched Pakistan's first public-sector Pulsed Field Ablation program for atrial fibrillation care. The rollout is a notable local medtech deployment because it brings an advanced electrophysiology treatment platform into a public hospital setting.
The Hacker News reports that Microsoft has shipped fixes for RoguePlanet, a Defender vulnerability tracked as CVE-2026-50656. Because public details circulated before the fix, endpoint teams should confirm Microsoft Malware Protection Engine updates have landed across managed Windows fleets.
The Hacker News reports on an AI Now Institute proof of concept called Friendly Fire that targets autonomous security scans by AI coding agents. The issue affects workflows using tools such as Claude Code and OpenAI Codex, making sandboxing, repo trust, and approval boundaries important before teams let agents inspect untrusted projects.
VentureBeat reports that OpenAI has launched GPT-Live, a new voice model pair designed to listen and speak at the same time. The change could improve live translation, support calls, and voice agents by reducing the turn-taking delays common in older speech interfaces.
TechCrunch reports that Ollama has raised $65 million and now reaches nearly 9 million users, with heavy adoption among developers running AI models locally. The funding reinforces that local model workflows are moving from hobbyist setup into mainstream developer infrastructure.