Several interesting events have occurred in the world of artificial intelligence. The Chinese company HuHu AI has released an updated version of its Qwen3.8-27B model, which is now available for free and without restrictions. The model supports a reasoning mode, works with a context of up to 262,000 tokens, and can process text and images. The developers have removed all censorship restrictions, making it attractive for users who need a powerful and unrestricted neural network.

However, with the growing popularity of AI tools, new problems arise. A study presented at the IEEE-ISTAS 2025 conference showed that iterative code generation with AI leads to a degradation of code security over time. The researchers conducted an experiment: they took 10 secure programs and asked an AI to improve them using four different strategies (adding features, optimization, security improvement, and general requests). After 10 iterations, they obtained 400 code samples and found that the number of vulnerabilities increased with each iteration. Adding new features proved particularly dangerous, resulting in 158 vulnerabilities. Even when the AI was asked to improve security, it still created new holes (38 vulnerabilities). The researchers also found that code complexity is directly related to the number of vulnerabilities: a 10% increase in complexity leads to an average 14.3% increase in vulnerabilities. This means developers must review code after each iteration rather than relying entirely on AI.

Another issue is the leakage of sensitive data when using AI chats. Many users paste contracts, resumes, passwords, and other confidential information into chats without thinking about the consequences. To address this, developers have created a free browser extension that automatically masks sensitive data before sending it to AI. The extension works locally: it detects names, card numbers, INN, and other data in the text, replaces them with aliases, and after receiving a response from the AI, decrypts them back on the user's screen. Only anonymized labels are sent to the AI server, and the mapping dictionary is stored in the browser's memory and deleted when the browser closes. The extension supports over 70 text formats, including PDF and document scans, and does not require an internet connection to work. This allows users to protect their data without giving up the convenience of AI tools.