ai-safety ai-regulation ai-governance regulations international-relations policy transparency

Share this with friends, journalists and politicians

Mandate reporting of AI training runs above 10²⁶ FLOPs to a designated national or international authority

Key insights

The main disagreement is often about the trigger for oversight, with several supporters using compute as a workable proxy and several critics arguing for capability-based or other risk indicators instead.
Supportive proposals differ on forum, spanning domestic agencies, national commissions, and international registries or treaty bodies.
Several pro-reporting quotes frame 10^26 FLOPs as an initial, placeholder, tiered, or adjustable benchmark rather than a permanent line.
Both supportive and skeptical quotes raise confidentiality and surveillance concerns, though supporters answer them with nonpublic or aggregated reporting while critics see them as reasons for caution.
The abstaining legal and standards-oriented quotes mostly describe how existing or proposed regimes define and operationalize FLOP-reporting thresholds rather than arguing the policy's merits.

Arguments for

Official notice of frontier training

Supporters say developers should have to tell a government authority when a model meets, or is expected to meet, a frontier-scale compute threshold. In their view, advanced AI models need mandatory official notice and review rather than voluntary disclosure alone.

When a general-purpose AI model meets the compute threshold, or it becomes known that this threshold will be met, the provider must notify the Commission without delay, and within two weeks at the latest.

European Commission · Jan 12, 2026 · full quote
we propose that AI developers pre-register large training runs [...] within their home country’s government [...] plausible bar for now would be 10^26 FLOP or higher.

Anthropic · 2023 · full quote
submit a report to the Secretary of Commerce [...] requires a quantity of computing power greater than 10^{26} integer or floating-point operations.

Joe Biden · Oct 30, 2023 · full quote
The Commission [...] shall [...] promulgate regulations that [...] include requirements for a covered entity—(i) to submit specified information related to each foundation model provided by the entity to the Commission [...] The term “covered entity” means any person, partnership…

Don Beyer · Mar 19, 2026 · full quote
Congress must now codify the White House's EO with legislative action. Lawmakers need to create a legal framework that makes federal government review of advanced AI models mandatory. [...] These companies need oversight and cannot be trusted to do the right thing voluntarily.

Brendan Steinhauser · 2026 · full quote
The research community, while they should not be expected to reveal to the world where they have got to, should be in dialogue with government. They need to be required to tell government where they are and what they are doing on the research front.

Baroness Neville-Jones · Jan 29, 2026 · full quote

Auditable compute trigger

Proponents present 10^26 FLOPs as a workable initial criterion for defining frontier activity because it gives regulators an auditable, unified trigger tied to the largest training efforts. Some pair that trigger with periodic disclosures on capabilities and human oversight so review can keep pace with frontier development.

Leading labs need a unified framework for frontier AI safety, security, incident reporting, and transparency. [...] The 10^26 FLOPs standard used as a placeholder in other regulations could serve as an initial criterion.

Google · Jun 2026 · full quote
to train models beyond a sufficiently high threshold of compute should be required to pre-register training runs [...] a threshold of 10^26 FLOPs would likely suffice.

Samuel Hammond · 2023 · full quote
A covered developer — Compute threshold: 10^26 FLOP — has trained or initiated the training of a frontier model, a model trained using, or undergoing training intended to use, at least 10^26 floating-point operations of compute. [...] Covered developers file quarterly disclosures…

Americans for Responsible Innovation · Aug 10, 2026 · full quote
We commend the BIS rule’s inclusion of a provision which requires the reporting [...] computing power greater than 10^{26} integer or floating-point operations.

Future of Life Institute · 2024 · full quote

International registry and verification

Another line of support argues that the largest AI capabilities rest on physical computing infrastructure whose concentration should be visible across borders. From this perspective, high-end training runs should be registered with an international body or registry, with monitoring and verification built into the regime.

The physical foundations of frontier artificial intelligence, the fabrication of advanced semiconductors and the operation of computing infrastructure at the largest scales, shall be subject to international transparency, so that the concentration of capability is visible to all …

The Global AI Accord · Jul 5, 2026 · full quote
Developers conducting training runs above an agreed compute threshold could be required to register them in an international registry. Telemetry from export-controlled AI chips could be reported to national regulators.

sepiatone · Mar 11, 2026 · full quote
Any training runs above the Monitored Threshold in a signatory’s jurisdiction require a report to the CTB including code, data, and estimated FLOP.

Machine Intelligence Research Institute · May 12, 2026 · full quote
Inspired by the International Atomic Energy Agency (IAEA), this agency will be responsible for: 1. Granting approval for major deployments and training runs of AIs that are judged not to pose a threat to humanity. This will include red-teaming and model evaluations. 2. Granting a…

PauseAI · 2026 · full quote
Each Party shall report any training run between 1e22 and 1e24 FLOP to the ISIA, prior to initiation.

If Anyone Builds It, Everyone Dies · Jul 2026 · full quote

Highest-capability focus

Some supporters argue that if policymakers are worried about the most advanced systems, oversight should concentrate on the highest-capability end rather than blanket regulation of all AI. They describe large-scale training runs and large compute clusters as the place where current frontier systems emerge and where oversight is most justified.

If policymakers are worried about the most advanced systems, then oversight should focus on the highest-capability end of the stack. Commerce already moved in this direction when BIS proposed reporting requirements for advanced AI models and large computing clusters. The proposal…

Jay Burstein · 2026 · full quote
We focus first on the large-scale training runs that currently produce frontier AI, setting thresholds beyond which training is monitored or prohibited.

Machine Intelligence Research Institute (MIRI) · May 12, 2026 · full quote
we propose that AI developers pre-register large training runs [...] within their home country’s government [...] plausible bar for now would be 10^26 FLOP or higher.

Anthropic · 2023 · full quote

Arguments against

Compute is a weak proxy

Critics argue that a regulatory trigger calibrated to compute volume does not measure what it claims to measure and can quickly become outdated as model architectures and training efficiency change. They prefer definitions aimed at actual frontier capabilities and current risks rather than everything that crosses yesterday's compute line.

The goal should not be to regulate everything that crosses yesterday’s compute threshold, but to identify the models that actually represent today’s technological frontier and tomorrow’s greatest risks. A dynamic, capability-based definition would give policymakers a better chanc…

Michelle Lopes Maldonado · Aug 28, 2026 · full quote
A regulatory trigger calibrated to compute volume does not measure what it purports to measure, and it was designed around a snapshot of model architecture that the industry has already moved beyond.

KI Bundesverband · Mar 13, 2026 · full quote
The current thresholds (10^25 FLOP in the EU AI Act and 10^26 FLOP in the now rescinded US Executive Order) may not be the most effective risk indicators.

AI Knowledge Consortium · Feb 2026 · full quote

Wrong target for regulation

Another objection is that policymakers should distinguish regulating applications from regulating the underlying technology or research itself. On this view, AI development should not be treated as a first-of-its-kind policy problem, and direct regulation of research and development is ill-advised.

However, it is important to differentiate regulating applications (which we need) vs. regulating the technology (which is ill-advised).

Andrew Ng · Dec 11, 2023 · full quote
The question that people are debating is whether it makes sense to regulate research and development of AI. And I don't think it does.

Yann LeCun · 2023 · full quote
Policymakers do not need to approach each innovation in isolation and should not treat every emerging technology as a first-of-its-kind policy problem.

Michael Kratsios · Sep 2, 2026 · full quote

Government control and compliance

Opponents also warn that mandatory review or transparency regimes can give government excessive control over AI technologies, chill innovation or expression, and turn compliance into the focus. They add that any reporting rule would need much clearer scoping, including distinctions between training and fine-tuning, before being imposed.

Establishing a pre-release review or licensing regime for AI companies would grant the government, particularly the executive branch, significant control over AI technologies that could hinder innovation or control expression. [...] Such an approach would open the door to a level…

Jennifer Huddleston · 2026 · full quote
Mandated transparency may also not actually improve consumer education around their privacy choices. First, many consumers may grow fatigued and frustrated with the constant pop ups and consents as they do with current cookie pop ups. Second, a government mandate — as opposed to …

Jennifer Huddleston · Jun 28, 2024 · full quote
AI training run reporting rules should be developed through a distinct study process. We also recommend clarification of the rules to distinguish between pre- and post-training activities like fine-tuning. Fine-tuning a qualified model should not trigger a reporting requirement.

Technology Trade Regulation Alliance · Apr 2026 · full quote

Middle ground

Adjustable thresholds and confidentiality

Some supportive but qualified proposals treat 10^26 FLOPs as a starting point rather than a permanent line, calling for tiered or adjustable thresholds as training efficiency evolves. They also propose aggregated or nonpublic reporting so authorities gain visibility without requiring researchers to reveal everything publicly.

Tiered compute thresholds. Basic reporting at 10^25 FLOPs; full transparency at 10^26 FLOPs, adjustable as training efficiency evolves.

AIPolis · 2026 · full quote
Leading labs need a unified framework for frontier AI safety, security, incident reporting, and transparency. [...] The 10^26 FLOPs standard used as a placeholder in other regulations could serve as an initial criterion.

Google · Jun 2026 · full quote
Mandate facility-level reporting: annual electricity, PUE/WUE, water sources/volumes, Scope 1–2 (with a roadmap to Scope 3); require disclosures for large AI training runs via a DSIT code of practice with a codified MWh/compute-hour threshold and aggregated/interval reporting to …

Hleb Buziuk · Jun 2026 · full quote
The research community, while they should not be expected to reveal to the world where they have got to, should be in dialogue with government. They need to be required to tell government where they are and what they are doing on the research front.

Baroness Neville-Jones · Jan 29, 2026 · full quote

Implementation safeguards

Even advocates of compute-based oversight acknowledge design challenges, including defining which runs are truly high-stakes, preventing developers from splitting workloads to stay below thresholds, and keeping monitoring from becoming surveillance.

A compute permit system would face its own challenges: defining which training runs qualify as high-stakes, preventing developers from fragmenting workloads to stay below thresholds, and ensuring that monitoring infrastructure does not become a tool for government surveillance of…

Joel N. Christoph · Feb 18, 2026 · full quote

Overall, the debate centers on whether mandatory reporting of very large training runs offers a practical way to create visibility into frontier AI or whether a fixed compute line is too blunt and intrusive to track real risk. Supporters describe reporting as a targeted transparency tool for the highest-end systems, sometimes with international verification or adjustable thresholds, while opponents argue that fixed compute triggers can misread capabilities and open the door to excessive control over research and innovation.

Cast Your Vote:

Results (38 votes):

Total (38 votes)

For 27 (71%) Abstain 2 (5%) Against 9 (24%)

New Top

For (23)

Americans for Responsible Innovation
AI policy advocacy nonprofit

I am this person! Add a sourced quote Report comment

votes For and says:

A covered developer — Compute threshold: 10^26 FLOP — has trained or initiated the training of a frontier model, a model trained using, or undergoing training intended to use, at least 10^26 floating-point operations of compute. [...] Covered develop...
more AI Verified source (Aug 10, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
The Global AI Accord
Independent, non-commercial consultation text proposing binding international governance of artificial intelligence.

I am this person! Add a sourced quote Report comment

votes For and says:

The physical foundations of frontier artificial intelligence, the fabrication of advanced semiconductors and the operation of computing infrastructure at the largest scales, shall be subject to international transparency, so that the concentration of...
more AI Verified source (Jul 5, 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
If Anyone Builds It, Everyone Dies
AI-safety campaign and collective publishing a proposed international treaty to prevent superintelligence.

I am this person! Add a sourced quote Report comment

votes For and says:

Each Party shall report any training run between 1e22 and 1e24 FLOP to the ISIA, prior to initiation.
AI Verified source (Jul 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
TAISC
Organization that published a proposed treaty on artificial intelligence safety and cooperation.

I am this person! Add a sourced quote Report comment

votes For and says:

Each Party shall report any training run between 1e22 and 1e24 FLOP to the ISIA, prior to initiation.
AI Verified source (Jun 30, 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Hleb Buziuk
Independent policy researcher and human-rights advocate writing on AI infrastructure sustainability and transparency.

I am this person! Add a sourced quote Report comment

votes For and says:

Mandate facility-level reporting: annual electricity, PUE/WUE, water sources/volumes, Scope 1–2 (with a roadmap to Scope 3); require disclosures for large AI training runs via a DSIT code of practice with a codified MWh/compute-hour threshold and agg...
more AI Verified source (Jun 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Google
Search and advertising company

I am this person! Add a sourced quote Report comment

votes For and says:

Leading labs need a unified framework for frontier AI safety, security, incident reporting, and transparency. [...] The 10^26 FLOPs standard used as a placeholder in other regulations could serve as an initial criterion.
AI Verified source (Jun 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Machine Intelligence Research Institute
Nonprofit focused on advanced AI safety

I am this person! Add a sourced quote Report comment

votes For and says:

Any training runs above the Monitored Threshold in a signatory’s jurisdiction require a report to the CTB including code, data, and estimated FLOP.
AI Verified source (May 12, 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Joe Rogero
MIRI writer

I am this person! Add a sourced quote Report comment

votes For and says:

Any training runs above the Monitored Threshold in a signatory’s jurisdiction require a report to the CTB including code, data, and estimated FLOP.
AI Verified source (May 12, 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Machine Intelligence Research Institute (MIRI)
AI safety research institute

I am this person! Add a sourced quote Report comment

votes For and says:

We focus first on the large-scale training runs that currently produce frontier AI, setting thresholds beyond which training is monitored or prohibited.
AI Verified source (May 12, 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Don Beyer
U.S. Representative from Virginia's 8th district; co-sponsor of the GUARDRAILS Act on AI

I am this person! Add a sourced quote Report comment

votes For and says:

The Commission [...] shall [...] promulgate regulations that [...] include requirements for a covered entity—(i) to submit specified information related to each foundation model provided by the entity to the Commission [...] The term “covered entity”...
more AI Verified source (Mar 19, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
sepiatone
Author of the Sepia Tone newsletter on international AI governance and verification.

I am this person! Add a sourced quote Report comment

votes For and says:

Developers conducting training runs above an agreed compute threshold could be required to register them in an international registry. Telemetry from export-controlled AI chips could be reported to national regulators.
AI Verified source (Mar 11, 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Joel N. Christoph
AI governance commentator and World Economic Forum contributor.

I am this person! Add a sourced quote Report comment

votes For and says:

A compute permit system would face its own challenges: defining which training runs qualify as high-stakes, preventing developers from fragmenting workloads to stay below thresholds, and ensuring that monitoring infrastructure does not become a tool ...
more AI Verified source (Feb 18, 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Baroness Neville-Jones
British Conservative life peer and former UK security minister.

I am this person! Add a sourced quote Report comment

votes For and says:

The research community, while they should not be expected to reveal to the world where they have got to, should be in dialogue with government. They need to be required to tell government where they are and what they are doing on the research front.
AI Verified source (Jan 29, 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
European Commission
EU's executive body

I am this person! Add a sourced quote Report comment

votes For and says:

When a general-purpose AI model meets the compute threshold, or it becomes known that this threshold will be met, the provider must notify the Commission without delay, and within two weeks at the latest.
AI Verified source (Jan 12, 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
AIPolis
U.S. AI-policy framework published by Lumiere Media in 2026.

I am this person! Add a sourced quote Report comment

votes For and says:

Tiered compute thresholds. Basic reporting at 10^25 FLOPs; full transparency at 10^26 FLOPs, adjustable as training efficiency evolves.
AI Verified source (2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
PauseAI
Global campaign to pause frontier AI

I am this person! Add a sourced quote Report comment

votes For and says:

Inspired by the International Atomic Energy Agency (IAEA), this agency will be responsible for: 1. Granting approval for major deployments and training runs of AIs that are judged not to pose a threat to humanity. This will include red-teaming and mo...
more AI Verified source (2026)

added 5mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Anthropic
AI safety research company

I am this person! Add a sourced quote Report comment

votes For and says:

we propose that AI developers pre-register large training runs [...] within their home country’s government [...] plausible bar for now would be 10^26 FLOP or higher.
AI Verified source (2023)

added 8mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Jay Burstein
Build American AI fellow

I am this person! Add a sourced quote Report comment

votes For and says:

If policymakers are worried about the most advanced systems, then oversight should focus on the highest-capability end of the stack. Commerce already moved in this direction when BIS proposed reporting requirements for advanced AI models and large co...
more Unverifiable source (2026)

added 5mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Samuel Hammond
Director of AI Policy and Chief Economist at the Foundation for American Innovation

I am this person! Add a sourced quote Report comment

votes For and says:

to train models beyond a sufficiently high threshold of compute should be required to pre-register training runs [...] a threshold of 10^26 FLOPs would likely suffice.
Unverifiable source (2023)

added 8mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Alex Bores
New York State Assemblymember

I am this person! Add a sourced quote Report comment

votes For and says:

A “frontier model” means a foundation model that was trained using a quantity of computing power greater than 10²⁶ integer or floating-point operations. Before, or concurrently with, deploying a new frontier model or a substantially modified version ...
more Unverifiable source (Mar 27, 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Brendan Steinhauser
CEO of The Alliance for Secure AI; longtime Republican political strategist and former Tea Party organizer

I am this person! Add a sourced quote Report comment

votes For and says:

Congress must now codify the White House's EO with legislative action. Lawmakers need to create a legal framework that makes federal government review of advanced AI models mandatory. [...] These companies need oversight and cannot be trusted to do t...
more Disputed source (2026)

added 5mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate

Loading more opinions...

Abstain (2)

Loading more opinions...

Against (9)

Michael Kratsios
U.S. OSTP Director

I am this person! Add a sourced quote Report comment

votes Against and says:

Policymakers do not need to approach each innovation in isolation and should not treat every emerging technology as a first-of-its-kind policy problem.
AI Verified source (Sep 2, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Michelle Lopes Maldonado
Associate director for AI policy at the Information Technology and Innovation Foundation; former Virginia state delegate and AI policy leader.

I am this person! Add a sourced quote Report comment

votes Against and says:

The goal should not be to regulate everything that crosses yesterday’s compute threshold, but to identify the models that actually represent today’s technological frontier and tomorrow’s greatest risks. A dynamic, capability-based definition would gi...
more AI Verified source (Aug 28, 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Technology Trade Regulation Alliance
Technology-industry trade association submitting policy comments on AI training-run reporting.

I am this person! Add a sourced quote Report comment

votes Against and says:

AI training run reporting rules should be developed through a distinct study process. We also recommend clarification of the rules to distinguish between pre- and post-training activities like fine-tuning. Fine-tuning a qualified model should not tri...
more AI Verified source (Apr 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
KI Bundesverband
German artificial-intelligence industry association representing AI companies and contributing to AI policy and regulatory consultations.

I am this person! Add a sourced quote Report comment

votes Against and says:

A regulatory trigger calibrated to compute volume does not measure what it purports to measure, and it was designed around a snapshot of model architecture that the industry has already moved beyond.
AI Verified source (Mar 13, 2026)

added 3mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
AI Knowledge Consortium
Indian civil-society consortium developing recommendations on artificial-intelligence governance.

I am this person! Add a sourced quote Report comment

votes Against and says:

The current thresholds (10^25 FLOP in the EU AI Act and 10^26 FLOP in the now rescinded US Executive Order) may not be the most effective risk indicators.
AI Verified source (Feb 2026)

added 2mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Andrew Ng
Baidu; Stanford CS faculty; founded Coursera and Google Brain

I am this person! Add a sourced quote Report comment

votes Against and says:

However, it is important to differentiate regulating applications (which we need) vs. regulating the technology (which is ill-advised).
AI Verified source (Dec 11, 2023)

added 7mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Yann LeCun
Computer scientist, AI researcher

I am this person! Add a sourced quote Report comment

votes Against and says:

The question that people are debating is whether it makes sense to regulate research and development of AI. And I don't think it does.
AI Verified source (2023) 1 of 2

added 8mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Jennifer Huddleston
Senior fellow, Cato Institute

I am this person! Add a sourced quote Report comment

votes Against and says:

Establishing a pre-release review or licensing regime for AI companies would grant the government, particularly the executive branch, significant control over AI technologies that could hinder innovation or control expression. [...] Such an approach ...
more Disputed source (2026) 1 of 2

added 4mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Jennifer Huddleston
Senior fellow, Cato Institute

I am this person! Add a sourced quote Report comment

votes Against and says:

Mandated transparency may also not actually improve consumer education around their privacy choices. First, many consumers may grow fatigued and frustrated with the constant pop ups and consents as they do with current cookie pop ups. Second, a gover...
more Disputed source (Jun 28, 2024)

added 1y ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate

Loading more opinions...

More h/ai-safety votes

Slow down AI development

AI labs should conduct risk assessments before high-stakes internal deployment of frontier models

AI labs should give independent evaluators employee-like access

Establish an international AI incident reporting system

The U.S. and China should establish Mutually Assured Compute Destruction

Back