US AI crackdown sparks open-source wave
A surprise crackdown by the US government on powerful AI systems from OpenAI and Anthropic has shaken the global tech industry and revived a long-running debate over open versus closed AI. Restrictions on who can access top-tier models have forced many startups and developers to rethink their dependence on a few American “frontier labs”.
Instead of relying on single, closed platforms, more teams are turning to open-source AI models that they can host and control themselves. This shift is not just technical; it is changing the balance of power in the AI ecosystem.
How restrictions hit OpenAI and Anthropic
In early June, Washington ordered Anthropic to block non-US users from accessing its strongest closed models, Mythos 5 and Fable 5. Rather than build complex screening systems, the company removed both models from public use altogether, leaving many developers suddenly without a key tool.
Soon after, OpenAI agreed that the US government would approve every customer for its newest model, GPT‑5.6, adding another layer of uncertainty for businesses that depend on its services. For startups building products on top of these systems, the message was clear: access can be cut off overnight.
Why developers are flocking to open-source AI
For many founders and engineers, the Mythos and Fable shutdowns were a wake-up call about relying too heavily on closed, remote models. Open-source or “open‑weight” models work differently because the core model files can be downloaded, fine-tuned and run on a company’s own infrastructure.
Once these files are available, neither governments nor vendors can easily take them back. That makes open-source AI attractive not only from a cost perspective, but also for reliability and long-term control over critical systems.
China’s Zhipu AI and DeepSeek step into the spotlight
The US crackdown has also opened space for Chinese firms offering competitive open models. Zhipu AI’s GLM‑5.2, for example, is an open model that performs close to top Western offerings on major benchmarks, while remaining free to download and run on private servers.
Platforms that route traffic across many models have already seen a shift: usage share for Google, Anthropic and OpenAI combined dropped noticeably between January and June, while China’s open DeepSeek has emerged as a leading choice. For cost‑sensitive teams, strong performance at a lower or zero license fee is a powerful incentive to switch.
Cost pressure and flexibility for AI startups
Using closed AI has been getting more expensive as providers raise prices and introduce tighter usage limits. At the same time, approvals, risk checks and regional access rules make planning harder for global businesses.
Open models give startups flexibility: they can mix and match different systems, run them locally, and avoid being locked into a single supplier. As one founder put it, no serious company now wants to depend on just one lab when rules and pricing can change with little warning.
Are Chinese models still seen as a security risk?
Early in the AI boom, many Western experts worried that Chinese AI systems might pose a security or privacy threat. Those concerns have not fully disappeared, but some practitioners now argue that the fears are more emotional than rational when it comes to open models.
If a model is downloaded and run on local hardware, the original developer — whether based in Beijing, Paris or San Francisco — does not automatically gain access to user data. For some startups, this practical reality is enough to treat Chinese open models as viable tools rather than off‑limits technology.
Could open models be restricted next?
Despite the current momentum behind open-source AI, some researchers warn that tighter rules may eventually reach these systems as well. If governments decide that Mythos‑level models are too risky to be widely available, they may pressure companies and even foreign regulators to keep the most advanced systems closed.
That would leave a smaller pool of top-tier models under strict control, while today’s open tools might become the mid‑range option. In that scenario, debates over who should control the most powerful AI systems would only grow more intense.
What the US AI crackdown means for the future
The US AI crackdown has highlighted how much of today’s AI boom rests on access, not just algorithms. Developers who built their businesses around a single, closed platform are now searching for safer, more flexible foundations.
Open-source AI and strong non‑US models are becoming central to that search, giving companies more options and slightly reducing the dominance of a few frontier labs. Whether this leads to a more diverse, resilient AI ecosystem — or simply a new set of control battles — will depend on the next wave of policy decisions in Washington, Beijing and beyond.