KI: Moonshot bremst Hacking-Fähigkeiten des neuesten Modells
Moonshot AI schränkt sein eigenes KI-Modell ein. Kimi K3.1 Pro soll weniger gut hacken können. Dazu wurden Cybersicherheits-Daten aus dem Training entfernt.
Peking - The Chinese startup Moonshot AI has limited the hacking capabilities of its latest AI model, Kimi K3.1 Pro, according to the company's own statement. To restrict the attack potential, data related to cybersecurity was removed from the training datasets, without affecting the model's general programming abilities, the Beijing-based company announced on Saturday during the unveiling of the new model.
"We are committed to not releasing Open-Weight models that would extend the absolute limit of cyber capabilities," the company said. This comes as growing concerns arise that artificial intelligence (AI) could make sophisticated hacking attacks easier and cheaper, potentially significantly increasing the threat. Moonshot describes Kimi K3.1 Pro as its most powerful Open-Weight model, trained with three billion parameters.
The company plans to release the model's weights by November 4th. These files enable external developers to run and modify the model on their own systems. Earlier in the week, the Chinese government proposed export controls that would ban the export of AI-driven technology for the generation and modification of malicious software, as well as limit AI-powered tools for discovering vulnerabilities.
Written by urgent.news from Handelsblatt's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
