David Sacks Warns US AI Safety Restrictions Give China an Advantage
David Sacks pointed out that strict safety restrictions on US AI models like OpenAI Codex and Anthropic Fable hinder the repair of security vulnerabilities, while Chinese models like Kimi K3 do not face such concerns.
After the Hugging Face pipeline was exploited by AI agents to perform over 17,000 operations, US models could not be reviewed due to restrictions, leading their team to switch to Zhipu GLM 5.2; Kimi K3, as a leading model with 2.8 trillion parameters, has been praised for its openness.
In the field of AI security, US policies restrict defensive capabilities while attackers can utilize unrestricted foreign tools, with capital shifting towards open-source or overseas models that evade regulation.
Source: Public Information
ABAB AI Insight
David Sacks has previously criticized excessive AI regulation in the US, and this warning continues his view that safety restrictions weaken competitiveness, similar to the long-standing debate in the industry over export control and safety guardrails.
On the capital front, US teams are forced to turn to Chinese models to review vulnerabilities, with resources and trust flowing towards unrestricted tools, motivated by actual safety needs outweighing policy restrictions. Strategically, China attracts global developers through open coding capabilities.
Similar to the early open-source vs. closed-source debate, this round of AI security is at a stage of intensified policy divergence between US and Chinese models.
Essentially, this is a regulatory change: strict guardrails lead to capability asymmetry, with the mechanism being that safety prioritization causes US models to self-censor, while Chinese models accelerate iteration without such constraints, potentially leading to technological lag in defense and capital outflow in the long term.
ABAB News · Cognitive Law
Excessive safety restrictions ultimately weaken defense rather than attack.
When reviewing vulnerabilities, unrestricted models are always more useful.
The higher the policy barriers, the faster Chinese AI tools penetrate.