Chips that enforce rules
Chips with built-in controls. Advanced U.S. chips must carry hardware controls that can verify how they are used and enforce limits.
What it does
Every advanced AI chip sold would carry built-in security hardware that can enforce rules, such as refusing to run very large training jobs without a valid license or requiring a renewed permit to keep working. This would let the U.S., or a future treaty body, limit misuse of chips even after they leave the country.
Applies to: Advanced AI chips designed in the U.S.
- Advanced U.S. chips must carry hardware controls that can verify how they are used and enforce limits.
Where things stand
U.S. export controls restrict who can buy chips. Since Dec 2025 Nvidia offers opt-in fleet software that can estimate a chip's country, while stating its GPUs have no kill switches or backdoors.
Why it costs ~3 days
Designing controls into chips takes engineering time and may put off some buyers. Chips that reach China, legally or not, become easier to track and limit. Net: about a day.
Biggest unknown: Whether on-chip controls can be made tamper-proof against state-level hackers without pushing foreign buyers to non-U.S. chips.
Why it lowers p(doom) by ~0.15%
Builds controls into advanced chips so large training runs can be verified, licensed or switched off.
Hardware controls make future treaties checkable and rogue training runs harder. On their own they do little.
The strongest case that it costs more
Hardware backdoors can be hacked or abused, and foreign buyers may switch to chips without them. Nvidia has argued kill switches would undermine trust in U.S. chips.
The debate
For
- CNAS and IAPS (Onni Aarne, Tim Fist, Caleb Withers) 'Secure, Governable Chips' (Jan 2024) proposes on-chip governance mechanisms and a NIST-led interagency effort to develop them.
- flexHEG authors (James Petrie, Onni Aarne et al.), commissioned by ARIA Flexible Hardware-Enabled Guarantees reports (2025): a tamper-resistant guarantee processor that enforces agreed rules on chip use, usable for international agreements.
- Sen. Tom Cotton; Reps. Huizenga, Foster, Moolenaar, Krishnamoorthi Chip Security Act (May 2025): mandatory location verification on exported AI chips and study of further anti-diversion mechanisms.
Against
- Nvidia (Chief Security Officer David Reber Jr.) 'No Backdoors. No Kill Switches. No Spyware.' (Aug 2025): GPUs 'do not and should not have kill switches and backdoors'; such controls are gifts to hackers.
- Cyberspace Administration of China Summoned Nvidia on July 31, 2025 over alleged tracking and remote shutdown risks in H20 chips, signaling Chinese buyers will avoid controllable U.S. chips.
Sources
- CNAS: Secure, Governable Chips (Jan 2024): Core policy proposal for on-chip governance.
- Flexible Hardware-Enabled Guarantees for AI Compute: Technical design for usage-limiting hardware.
- ARI: Senate NDAA takes major step on AI chip export controls (July 14, 2026): Chip Security Act folded into Senate NDAA manager's amendment.
- CNBC: Nvidia's new software could help trace where its AI chips end up (Dec 2025): Industry's voluntary, opt-in alternative to mandated controls.
Rough starting points, not precise forecasts. Lead costs assume China doesn't depend on U.S. models, the case least favorable to safety laws, and count 3 years. On the menu you can change every assumption and put in your own numbers. Last priced 2026-09-26.