News of the day
1. OpenAI's new GPT-5.2-Codex model acts as an autonomous agent, solving complex tasks and finding vulnerabilities. A trusted access program offers experts a version with relaxed security filters for cyber defense. → Read more
2. Claude's new Chrome extension integrates AI directly into your browser, enabling context-aware assistance for tasks like summarization and drafting. → Read more
3. Zara leverages generative AI to adapt product imagery, speeding up content creation and reducing the need for repeated photoshoots, integrating AI into existing retail workflows. → Read more
4. Meta is developing new AI models, codenamed "Mango" and "Avocado," for image, video, and text generation, slated for release in 2026. → Read more
Our take
Hi Dotikers!
OpenAI just unveiled GPT-5.2-Codex, a new version of its code-focused AI. But this time, the focus is on a sensitive area: cybersecurity.
This model is designed to work as an autonomous software agent. It can analyze code for hours, navigate complex projects, and most importantly detect security vulnerabilities. OpenAI improved its ability to "compress" context, allowing it to keep an overview even on massive codebases.
Proof that it works? A security researcher used an earlier version to analyze React, Meta's framework. Result: the AI discovered 3 zero-day vulnerabilities that could potentially take down services or expose source code.
But here's the catch: an AI capable of finding flaws can also be used to exploit them. OpenAI is aware of this and now classifies this model as close to "high risk" level in its internal framework.
To manage this, OpenAI is launching an access program reserved for certified cybersecurity experts. These users will get access to a version with relaxed filters, so they can hunt for vulnerabilities without getting blocked. A bit like giving lockpicking tools to professional locksmiths while keeping them out of reach of the general public.
GPT-5.2-Codex is available now for paid ChatGPT subscribers.
Alex.
Meme of the day
