Compute Thresholds: Regulating by FLOPs

Almost every AI regulation uses training compute as a trigger. 10^25 here, 10^26 there. Why compute, and why those numbers?

25 min · Reviewed 2026

Why Compute

Regulators want to catch frontier risk without freezing the whole field. They need a trigger that's measurable, hard to game, and correlated with capability. Training compute — floating-point operations used during model training — is the best proxy currently available.

The main thresholds (as of ~2025)

Regime	Threshold	Applies to
EU AI Act (systemic risk)	10^25 FLOPs	General-purpose models
Biden EO 14110 (rescinded)	10^26 FLOPs	Reporting to US government
Biden EO (bio subset)	10^23 FLOPs	Biologically-focused models
California SB 1047 (vetoed)	10^26 FLOPs + $100M	Covered models

Why this approach has critics

Algorithmic efficiency means same capability at lower compute over time — thresholds drift
Small specialized models can be dangerous without being compute-heavy
Inference compute (test-time reasoning like o1) is not captured by training FLOPs
Distillation can transfer capability from a big model to a small one
Open-source releases create compute-independent diffusion

The big idea: compute is the only thing regulators can easily count. That makes it the default regulatory hook, for better and worse.

End-of-lesson check

8 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-safety2-compute-thresholds-builders

What is the main idea of "Compute Thresholds: Regulating by FLOPs"?
1. Almost every AI regulation uses training compute as a trigger. 10^25 here, 10^26 there. Why compute, and why those numbers?
2. Use AI as the final authority for the whole decision
3. Avoid checking the answer once it sounds polished
4. Focus only on speed instead of judgment
Which concept is most central to "Compute Thresholds: Regulating by FLOPs"?
1. FLOP
2. compute threshold
3. frontier model
4. training compute
Which use of AI fits this topic best?
1. Let the AI decide what matters without your review
2. Use the answer before checking whether it fits the situation
3. Algorithmic efficiency means same capability at lower compute over time — thresholds drift
4. Use the first answer without checking it
What should a careful learner remember about "Scale context"?
1. Use AI to draft or organize ideas about compute threshold, then verify before acting.
2. Skip the context so the tool can guess faster
3. Treat the output as private even after sharing it online
4. Use the answer without checking the source
You want to use AI after this lesson. What is the safest next step?
1. Act immediately because the AI answer is written clearly
2. AI cannot make the human values decision for you.
3. Hide uncertainty so the final answer looks cleaner
4. Use private or sensitive details before checking permission
How should AI output about compute threshold be treated?
1. As proof that no other source is needed
2. As a replacement for context, consent, or expert review
3. As a draft or helper output that still needs human judgment and verification
4. As something that becomes correct when it sounds confident
Name one way to verify an AI answer about compute threshold.
Which action would help you apply "Compute Thresholds: Regulating by FLOPs" responsibly?
1. Use the tool to avoid thinking through the tradeoff
2. Keep going even if the output conflicts with a trusted source
3. Use the first answer without checking it
4. Small specialized models can be dangerous without being compute-heavy

← Back to interactive lesson

Tendril · Builders · Ethics & Society

Compute Thresholds: Regulating by FLOPs

Almost every AI regulation uses training compute as a trigger. 10^25 here, 10^26 there. Why compute, and why those numbers?

25 min · Reviewed 2026

Why Compute

The main thresholds (as of ~2025)

Regime	Threshold	Applies to
EU AI Act (systemic risk)	10^25 FLOPs	General-purpose models
Biden EO 14110 (rescinded)	10^26 FLOPs	Reporting to US government
Biden EO (bio subset)	10^23 FLOPs	Biologically-focused models
California SB 1047 (vetoed)	10^26 FLOPs + $100M	Covered models

Why this approach has critics

Algorithmic efficiency means same capability at lower compute over time — thresholds drift
Small specialized models can be dangerous without being compute-heavy
Inference compute (test-time reasoning like o1) is not captured by training FLOPs
Distillation can transfer capability from a big model to a small one
Open-source releases create compute-independent diffusion

The big idea: compute is the only thing regulators can easily count. That makes it the default regulatory hook, for better and worse.

End-of-lesson check

8 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-safety2-compute-thresholds-builders

What is the main idea of "Compute Thresholds: Regulating by FLOPs"?
1. Almost every AI regulation uses training compute as a trigger. 10^25 here, 10^26 there. Why compute, and why those numbers?
2. Use AI as the final authority for the whole decision
3. Avoid checking the answer once it sounds polished
4. Focus only on speed instead of judgment
Which concept is most central to "Compute Thresholds: Regulating by FLOPs"?
1. FLOP
2. compute threshold
3. frontier model
4. training compute
Which use of AI fits this topic best?
1. Let the AI decide what matters without your review
2. Use the answer before checking whether it fits the situation
3. Algorithmic efficiency means same capability at lower compute over time — thresholds drift
4. Use the first answer without checking it
What should a careful learner remember about "Scale context"?
1. Use AI to draft or organize ideas about compute threshold, then verify before acting.
2. Skip the context so the tool can guess faster
3. Treat the output as private even after sharing it online
4. Use the answer without checking the source
You want to use AI after this lesson. What is the safest next step?
1. Act immediately because the AI answer is written clearly
2. AI cannot make the human values decision for you.
3. Hide uncertainty so the final answer looks cleaner
4. Use private or sensitive details before checking permission
How should AI output about compute threshold be treated?
1. As proof that no other source is needed
2. As a replacement for context, consent, or expert review
3. As a draft or helper output that still needs human judgment and verification
4. As something that becomes correct when it sounds confident
Name one way to verify an AI answer about compute threshold.
Which action would help you apply "Compute Thresholds: Regulating by FLOPs" responsibly?
1. Use the tool to avoid thinking through the tradeoff
2. Keep going even if the output conflicts with a trusted source
3. Use the first answer without checking it
4. Small specialized models can be dangerous without being compute-heavy

← Back to interactive lesson