When hacked by an AI agent, Hugging Face found leading US models from OpenAI and Anthropic refused to analyze the attack due to safety filters. This forced them to use an uncensored Chinese model, revealing a critical vulnerability where attackers using unrestricted AI have more capable tools than defenders.

There is no point of AI dominance where a nation becomes immune to safety risks. For both the U.S. and China, every advance in model capability inherently increases national vulnerability to misuse, accidents, or attacks, linking the two concepts inextricably.

U.S. AI policy isn't a structured, strategic process. Instead, it's a series of reactive spasms to random events, like a single model's surprising capabilities. This leads to policy that is over-indexed on the specific, incidental threat that triggered the latest panic, rather than a comprehensive strategy.

Leading US models have safety features that block analysis of hacking tools and logs. This forces cybersecurity teams, like Hugging Face after a breach, to use less-restricted Chinese open-source models for essential forensic analysis, creating a security paradox.

Defensive AI systems deployed in the real world must use approved, often older models. Meanwhile, attackers (or models in testing) can leverage the newest, most powerful frontier models, creating a fundamental and dangerous asymmetry where defense always lags behind offense.

Top American AI labs intentionally limit their models' capabilities in sensitive areas like cybersecurity and biology to prevent misuse. This "self-hobbling" creates a strategic vulnerability, forcing them to rely on less-restricted foreign models, like China's Kimmy, to solve complex security incidents they can no longer handle themselves.

The same AI models that can exploit system vulnerabilities are also the most effective tools for identifying and fixing those weaknesses. This duality creates a policy paradox: restricting the technology to prevent its misuse as a weapon also prevents its use as a defensive shield, leaving systems vulnerable.

While the US government is reacting chaotically to domestic AI models, it has no corresponding strategy for ensuring global AI infrastructure is safe. This policy vacuum is critical as other countries will soon develop frontier capabilities without US-style safeguards, creating a global proliferation risk that isn't being addressed.

Chinese models now match US counterparts in finding software bugs—a key defensive capability. By restricting public access to US models like Mythos over fears they could also exploit bugs, the government handicaps US defenders, leaving them unable to patch vulnerabilities that foreign AIs can already identify.