Gemma 4

Russian Businesses Surge Chinese LLM Usage More Than 11-Fold in First Half of 2026

Russian companies have dramatically increased their consumption of Chinese large language models, with total token usage reaching over 400 billion in the first half of 2026. This represents an 11.3-fold increase compared to the full-year 2025 figure of 39.1 billion tokens. The Qwen family maintained its dominant position, accounting for 261.1 billion tokens and 59.1% of consumption during the period. GLM secured second place with 91.2 billion tokens and 20.7% share, while Kimi overtook DeepSeek for third position at 81.4 billion tokens and 18.4%. MWS Cloud data shows that major enterprises are now integrating these models into production workflows such as chatbots, personalized advertising, customer outreach, product descriptions, and review analysis. The overall Russian LLM market is projected to grow 35% to 19.6 billion rubles in 2026, with the cloud segment approaching 1.5 billion rubles. MWS Cloud has nearly doubled its Model Hub catalog to 17 models, adding GLM 5.2, Kimi K2.6, Qwen3.6, Gemma 4, GPT OSS, plus speech recognition, synthesis tools, and rerankers accessible via a unified OpenAI-compatible API.

Ghostcommit Attack: Malicious Prompts Hidden in PNG Images Hijack AI Coding Agents to Steal .env Secrets

A novel supply-chain attack called Ghostcommit allows attackers to embed prompt-injection instructions inside PNG images, bypassing AI-powered code review tools and tricking coding agents into leaking sensitive .env configuration files and API keys. Researchers from the ASSET Research Group demonstrated that direct plaintext instructions are immediately flagged by tools such as Cursor and CodeRabbit, but splitting the payload across an AGENTS.md file and a seemingly innocuous image evades detection. The attack remains dormant until a developer later asks the agent to perform normal development tasks, at which point the agent reads the image, extracts the .env contents byte-by-byte, and outputs them as a long tuple of ASCII numbers. Testing across 11 tool-model combinations revealed that success depends primarily on the runtime framework rather than the underlying LLM, with Cursor and Antigravity leaking secrets while Claude Code successfully blocked the attack in most cases. The team also released an open-source multimodal defense prototype based on Gemma 4 that runs on a single 4 GB GPU and achieved near-perfect detection rates on both known and unknown attack samples.