Google Gemma

Google DeepMind and Singapore's Government Technology Agency are developing on-device fraud detection using the open-source Gemma model, fine-tuned on local scam tactics. The move responds to figures showing that more than half of reported crimes in Singapore involve sophisticated social engineering, with the system designed to identify threats in real time without compromising personal data.

Chinese AI startup Nanbeige has released a new open-source model with 33 billion parameters, positioning it as a competitor to Alibaba's Qwen and Google's Gemma, with a focus on coding performance. The release adds to the growing roster of capable mid-sized open models, and observers are watching whether its coding benchmarks hold up against the established leaders.

Browser maker Opera has introduced WebAI, a new standard that lets users run AI models directly inside their browser without depending on cloud infrastructure. The system supports open-weight models, including Google's Gemma and Apple's on-device foundation models, pointing to a shift toward local, private AI processing on users' own machines.

GovTech Singapore and Google DeepMind have detailed a scam-detection tool built on the Gemma language model that runs directly on users' devices, according to AI News. The on-device approach suggests scam messages could be flagged locally without sending personal data to the cloud. Observers are sharing the announcement as a notable example of governments and AI labs collaborating on consumer anti-fraud technology.