Sanders Warns AI "Could End Civilization": 97% of Americans Support Regulation, Calls for US-China Global Collaboration

In early 2025, U.S. Independent Senator from Vermont Bernie Sanders released a video via his official X account, warning that artificial intelligence could "end civilization as we know it" and calling on governments worldwide to establish global cooperation on AI safety rules. This statement once again thrust AI governance into the public spotlight. Source: @SenSanders official tweet.

Fact Check: What Sanders Actually Said

Based on confirmed information, Sanders conveyed three core messages in the video:

• Risk Characterization: Elevated AI risk to a "civilization-level" threat, rather than merely employment, privacy, or copyright issues.
• Public Endorsement: Cited the statistic that "97% of Americans support regulation of AI safety" as the political foundation for pushing legislation.
• Call for Cooperation: Explicitly called for a global collaboration framework involving both China and the U.S., emphasizing that "governments should wake up before it's too late."

Notably, Sanders has previously pushed for establishing cooperation mechanisms between China and the U.S. in AI safety, but this proposal has faced significant skepticism in Washington policy circles—with doubts primarily centered on China's AI development intentions and credibility. Source: The aforementioned video and related discussions on X platform.

Technical Principles: Why "Global Collaboration" Is So Critical for AI Safety

winzheng.com Research Lab believes that understanding the technical logic behind Sanders' appeal requires examining three layers:

First, frontier model training has a strong "scale spillover effect." The training compute for the most advanced large models (e.g., GPT-4 level and above) has reached the order of 10^25 FLOPs. Once a country trains a model with dangerous capabilities (such as autonomous cyberattacks or biothreat assistance), and if the weights leak or are open-sourced, the entire world shares the risk. This is fundamentally different from nuclear weapons—AI models can be copied, fine-tuned, and redistributed, making unilateral controls nearly ineffective.

Second, AI safety testing requires cross-border shared benchmarks. Currently, the U.S. NIST AI Safety Institute, the UK AISI, and relevant Chinese institutions have each established evaluation systems, but red teaming methods, dangerous capability thresholds, and alignment evaluation metrics lack interoperability. This means the same model could receive completely different safety conclusions in different regions.

Third, there is a "race-to-the-bottom" risk on the deployment side. If one country sets strict safety thresholds while another does not, companies have strong incentives to move training and deployment to regions with weaker regulation—this is precisely the technical basis for Sanders emphasizing a "global framework" rather than "U.S. unilateral legislation."

What the 97% Figure Means

"97% of Americans support AI safety regulation" is the core data point in Sanders' video. If this figure is accurate (as cited by Sanders), it means AI regulation already enjoys rare cross-party consensus in the U.S.—typically, a policy achieving 60% public support is considered a high level of consensus.

From a policy economics perspective, 97% support reflects that public perception of AI risk far exceeds the average level within the technical community. But public support does not equal policy capability—whether regulatory agencies possess the technical experts, computing resources, and legal tools to evaluate frontier models is a separate matter.

winzheng.com Research Lab's Perspective

The following are analytical judgments, distinct from the factual statements above:

View 1: The "technically feasible zone" for US-China AI safety cooperation is far narrower than the political narrative suggests. Both sides share common interests on issues such as general alignment research, child safety, deepfake detection, and prevention of CBRN (chemical, biological, radiological, nuclear) misuse; however, on topics like frontier model weight auditing, training data transparency, and military AI restrictions, structural mutual trust deficits make the threshold for collaboration extremely high. Sanders' appeal is morally correct, but engineering implementation requires a layered design.

View 2: The double-edged effect of the "end of civilization" narrative. On one hand, this discourse can mobilize legislative attention; on the other hand, excessive existential risk narratives may squeeze policy resources for addressing immediate concrete harms (algorithmic discrimination, labor displacement, information pollution). winzheng.com has long advocated that AI governance should distinguish between "near-field risks" and "far-field risks," as they require different toolkits.

View 3: There is a huge gap between a senator's personal appeal and an enforceable framework. A true global AI governance framework (similar to the IAEA for nuclear energy) would require: unified capability assessment protocols, cross-border compute monitoring mechanisms, and a custody system for dangerous model weights. These are currently at the academic discussion stage and are at least 3-5 years away from treaty-level text.

Implications for Developers and Enterprises

• Compliance First: Whether or not a global framework is established, the EU AI Act, the U.S. Executive Order, and China's "Interim Measures for the Management of Generative AI Services" already constitute a de facto compliance network.
• Evaluation Infrastructure: It is recommended that enterprises establish independent model evaluation pipelines internally, rather than relying on post-hoc audits by external parties.
• Geopolitical Risk Modeling: Cross-border deployed AI products should incorporate "regulatory bifurcation" into their risk models.

Sanders' appeal will not immediately reshape the global AI landscape, but it marks the spillover of AI safety issues from the technical community into the highest levels of political discourse. winzheng.com Research Lab will continue to track the policy implementation and technical mapping of this issue.

The factual portion of this article is based on Senator Sanders' official X account video and related public discussions; the analytical views are independent judgments of winzheng.com Research Lab and do not represent the position of any government or enterprise.