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

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...

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

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

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...

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.

Any training runs above the Monitored Threshold in a signatory’s jurisdiction require a report to the CTB including code, data, and estimated FLOP.

Any training runs above the Monitored Threshold in a signatory’s jurisdiction require a report to the CTB including code, data, and estimated FLOP.

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

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.

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.

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...

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.

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...

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.

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 ...

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...

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.

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

According to the proposed rule, a covered US person will be required to provide quarterly reports to BIS if the US person “engages in, or plans, within six months, to engage in ‘applicable activities.’” “Applicable activities” include: “Conducting a...

On September 11, 2024, the Department of Commerce (DOC), Bureau of Industry and Security (BIS) issued a proposed rule that would place an ongoing reporting requirement for entities developing, acquiring, or possessing advanced artificial intelligence...

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...

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.

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

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.

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 ...

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...