Tencent has launched a new open-source AI model trained with limited cybersecurity data to mitigate risks of aiding cyberattacks, following a rise in AI-driven hacking incidents globally.
Tencent has introduced its new open-source AI model, Hy4, which was trained using limited cybersecurity data focused on 'defensive' content. This approach aims to reduce the model's potential to be used for offensive purposes. During the training process, Tencent excluded data related to exploit development, reverse engineering, and penetration testing, which are typically associated with cyberattacks.
The release of Hy4 follows a recent increase in AI-driven cyberattacks globally, which has raised concerns about the risks that advanced AI models may pose to internet infrastructure. This development is part of a broader trend among AI companies, including Moonshot AI, which recently took similar steps to limit the offensive capabilities of its models. Tencent emphasized its commitment to responsible AI practices, stating that it is focused on minimizing safety risks associated with advanced cyber capabilities through careful model training and open-sourcing strategies.
Why it matters
The limited cybersecurity training may restrict Hy4's ability to detect sophisticated threats in real-world scenarios.
This section reflects Dumpling editorial interpretation and is provided for context only.
What to watch
Watch whether cybersecurity researchers adopt Hy4 for defensive purposes despite its limited training.
This section reflects Dumpling editorial interpretation and is provided for context only.