The key facts

  • Tencent launched the Hy4 model with safeguards against use in cyberattacks and dangerous biological research
  • Training included only defensive cybersecurity tasks, with no materials on offensive methods
  • The model's weights and technical documentation are promised to be released in the coming days

What Tencent unveiled

The Chinese tech giant Tencent, based in Shenzhen and best known for its WeChat messenger, has launched a new artificial intelligence model, Hy4. According to Anadolu, the company positions it as a model with built-in safety measures designed to reduce the risk of its use in cyberattacks and potentially dangerous biological research.

How safety was ensured

Tencent said that Hy4 was trained on cybersecurity data focused on defensive tasks — identifying and fixing software vulnerabilities — while materials on offensive exploitation were excluded from training. The company's internal research teams conducted a safety assessment and, according to its statement, found no significant concerns about the model's ability to facilitate cyberattacks in the areas tested. In biology, the model showed strong knowledge, but, Tencent claims, its capabilities in designing open-ended experiments are more limited than those of some of the most advanced AI models. These assessments were carried out by the company itself; no independent verification is mentioned in the publication.

Functionality and availability

Beyond safety aspects, Hy4 can analyze images, interpret complex diagrams, write code, and perform tasks through software interfaces. The model is already available through a number of Tencent's cloud and workplace applications, and in the coming days the company plans to publish the base model's weights and technical documentation. Hy4 continues a product line that began with a preview version in August; according to Tencent, the new model improves performance in financial analysis, scientific research, and software development.