The new Sanders-Casar Ban Artificial Superintelligence Act – and why I oppose it
The author opposes a proposed permanent, unilateral ban on all superhuman AI research, arguing it is too broad and would cede technological leadership to other nations A temporary pause on AI development is considered defensible, but only until better regulatory and alignment frameworks are established The proposed legislation is criticized for being naive about benchmarking complexities and vulnerable to exploitation by authoritarian regimes The author advocates for a high-bar regulatory model
Analysis
TL;DR
- The author opposes a proposed permanent, unilateral ban on all superhuman AI research, arguing it is too broad and would cede technological leadership to other nations
- A temporary pause on AI development is considered defensible, but only until better regulatory and alignment frameworks are established
- The proposed legislation is criticized for being naive about benchmarking complexities and vulnerable to exploitation by authoritarian regimes
- The author advocates for a high-bar regulatory model (per Anthony Aguirre) that allows continued development of narrow AI while requiring independent verification of alignment and control for frontier systems
- Current, tangible AI risks are being overshadowed by focus on hypothetical future existential threats
Why It Matters
This article directly addresses the ongoing policy debate around AI regulation in the United States, offering a nuanced position that balances safety concerns with innovation and geopolitical competitiveness. For AI practitioners and policymakers, it highlights the tension between precautionary regulation and the risk of regulatory capture or unilateral disarmament in a competitive global landscape.
Technical Details
- The proposed Sanders-Casar legislation calls for a permanent ban on research into superhuman AI, which the author argues would require an act of Congress to reverse
- Anthony Aguirre's alternative framework distinguishes between narrowly-scoped AI tools (allowed to continue developing) and systems exceeding current frontier capabilities in software development, cyber, and autonomy (requiring demonstrated alignment, control, and oversight)
- The author references their 2023 U.S. Senate testimony supporting the creation of an AI agency, separate from the debate over research bans
- Benchmarking complexities are cited as a critical flaw in the proposed legislation, creating enforcement gaps that adversarial nations could exploit
- The author's book Taming Silicon Valley and an upcoming article with Cal Newport are cited as containing more detailed policy proposals
Industry Insight
- AI companies should proactively invest in alignment and safety research to meet potential future regulatory thresholds, turning compliance into a competitive advantage
- Policymakers should favor performance-based regulatory frameworks (like Aguirre's high-bar model) over categorical bans, which risk stifling innovation while being difficult to enforce
- The AI industry should engage more actively in shaping regulatory discourse, as unilateral U.S. restrictions without international coordination could drive research underground or overseas
Disclaimer: The above content is generated by AI and is for reference only.