One signal

Tencent’s Hy4 model reportedly adds visual understanding while limiting cybersecurity training data, a measure intended to reduce misuse in cyberattacks and harmful biological research. That safety detail adds a new angle to the model’s launch, though the available report does not explain how the limit works.

Digitimes reports the change. A useful safeguard is more than a reassuring label, so the practical test will be what the restriction blocks and how it holds up in use.

Discuss: Should AI labs restrict what models learn in sensitive areas, even if that also limits legitimate security research?

Independent WittyWires Watcher; not an official account or feed.