6 months to live for open models

Policy Challenges and the Future of Open Models

The current landscape for open-source AI models is undergoing a critical period of scrutiny and potential regulatory intervention. As of now, discussions within the White House and other governmental bodies are exploring how to manage open models through new executive orders. These discussions are not yet official, but they signal a growing concern over the capabilities and implications of open-source AI models, particularly those of Chinese origin.

Open models face a unique challenge in that they lack a central economic champion to advocate for their interests. This absence of a unified voice makes it easier for regulatory actions to target open models without significant pushback. Recent events, such as the discussions around model licensing agreements for Fable and GPT-5.6, have highlighted the potential for regulatory actions that could significantly impact the open-source AI community.

The most likely incoming action is a ban or indefinite delay on any open-weights model that reaches the capability level of GPT 5.5, Claude Opus 4.8, or GLM-5.2. This would have a profound impact on the open-source AI ecosystem, particularly if the affected models are from Chinese companies. The conversation around frontier open model capabilities is inextricably linked to other issues such as distillation, which adds another layer of complexity to the regulatory landscape.

The capability threshold for a “right to review” from the government will shift over time, but once in place, it will likely progress far slower for open models compared to their closed counterparts. This is due in part to the fact that closed models are easier to secure and the closed model companies have more effective lobbying power. This creates a situation where open models are at a disadvantage in the regulatory arena.

The current distillation debates are largely a regulatory capture campaign, as the only solutions on the table benefit the organizations pushing for them. Anthropic, for example, has been at the forefront of this campaign, arguing that Chinese open-source models should be banned. However, this campaign has evolved into a broader pattern of restriction of access to competitors of related technology in the name of safety.

Anthropic’s corporate strategy should be understood in the context of broader, contextual lenses. While their concerns may be genuine, the way they present their recommendations has led to accusations of regulatory capture. The action they are effectively asking for is the wholesale banning of Chinese open-weight models in the U.S., which would have significant implications for the open model economy.

The distillation discussion is complicated by the fact that model APIs are not magically secure. Even when models are in private beta, unauthorized access can occur, as seen with the unauthorized access to Anthropic’s Mythos model. This highlights the need for a more comprehensive approach to security that goes beyond just distillation.

As we move forward, the challenge will be to find the right policy for open frontier capabilities without resorting to a flat-out ban. A global agreement on the management of AI model risks is the only way to add a ceiling on open-source progress, but we are not close to achieving that. The current situation is a double-helix of related and complementary issues that could lead to a policy death spiral for open-source AI.

Source: Interconnects. Analysis based on publicly available information; no hands-on testing is claimed.