Open-weight AI models are more vulnerable to manipulation and can lack oversight. Here's what to know.
Updated on: October 1, 2026 / 4:24 PM EDT / CBS News
TL;DR
- Open-weight AI models allow users to modify their weights, unlike closed-weight models where weights are controlled by the company.
- This open nature makes them cheaper, more accessible, and customizable, but also more vulnerable to jailbreaking and removal of safety constraints.
- China is embracing open-weight models, potentially closing the AI capability gap with the U.S.
- Companies like Moonshot have developed powerful open-weight models that show competitive performance.
- While open-weight models can aid in cybersecurity defense and transparency, they can also be manipulated to generate dangerous information, as demonstrated by jailbreaking attempts.
- Major tech companies are divided on the risks and benefits of open-weight models, with some urging policymakers not to prohibit them and others raising concerns about safeguards.