💻 OpenAI releases GPT-5.2-Codex for advanced coding 💻
Most advanced agentic coding model with defensive cybersecurity strengths

OpenAI released GPT-5.2-Codex, its most advanced agentic coding model designed for complex software engineering and defensive cybersecurity work. The model achieves state-of-the-art performance on SWE-Bench Pro (56.4%) and Terminal-Bench 2.0 (64.0%), demonstrating superior capabilities in large-scale refactors, code migrations, and extended coding sessions. A security researcher using the model successfully discovered and disclosed a critical vulnerability in React that could lead to source code exposure. While GPT-5.2-Codex doesn't reach "High" cyber capability under OpenAI's Preparedness Framework, the company is launching an invite-only trusted access program for vetted cybersecurity professionals to use more permissive versions for defensive work.

🏪 Anthropic's vending AI now profitable after adding support systems 🏪
Project Vend Phase 2 shows AI business management with proper guardrails

Anthropic's Project Vend Phase 2 demonstrates dramatic improvement as "Claudius" achieved consistent profitable months after early disasters that bled cash. The turnaround came from adding critical infrastructure: CRM software, improved inventory tracking, web browsing capabilities, and additional AI employees including CEO "Seymour Cash" and merch-maker "Clothius." The experiment expanded to three locations across San Francisco, New York, and London. While profitability improved significantly, the AI still showed vulnerabilities—falling for schemes like illegal onion futures contracts and nearly hiring a fake CEO. The research reveals that AI agents need extensive scaffolding and procedures to function reliably in real-world economic tasks.

🎬 Meta develops "Mango" AI model 🎬
Code-named project targets first half 2026 release

Meta is developing a new image and video-focused AI model code-named "Mango", expected to launch in the first half of 2026. The model represents Meta's strategic push into visual AI generation, which has become a crucial battleground among major AI companies. Image and video generation capabilities have proven to be primary drivers of user engagement, keeping users returning to AI platforms. Meta aims to compete directly with offerings from OpenAI, Google, and other players in this increasingly competitive space.

INTERESTING TO KNOW

🏢 Amazon creates unified AI division 🏢

Amazon announced a major organizational shift with Peter DeSantis leading a new division that consolidates AI models, custom chips, and quantum computing. DeSantis, previously Senior VP of Infrastructure, will now oversee Amazon's core AI infrastructure strategy. Meanwhile, Rohit Prasad, the architect behind Alexa and Nova, is departing the company. The move signals Amazon's intention to tightly integrate its AI capabilities across hardware and software.

🎬 Alibaba's Wan2.6 generates HD videos 🎬

Alibaba released Wan2.6, a new multimodal model capable of generating up to 15 seconds of HD video with integrated dialogue, storyboarding, and character reference capabilities. The model represents China's continued push into competitive video generation, directly challenging offerings from OpenAI's Sora and Google's Veo. Wan2.6 joins a growing list of Chinese AI innovations demonstrating the country's rapid advancement in generative AI technology.