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

ai-safety ai-policy ai-regulation ethics ai-risk ai-governance

Share this with your representatives, journalists and friends

Key insights

Across company policies, research groups, think tanks, and legislative bodies, the quotes consistently extend frontier-model governance to internal use rather than limiting it to public release.
Several quotes say the most capable systems are often used internally first, making internal deployment a place where dangerous thresholds may be reached before any public availability.
A recurring design pattern is to define explicit risk thresholds, evaluate models against them, apply mitigations, and review both the assessment and the adequacy of those mitigations before extensive internal use.
Some quotes emphasize outside visibility into internal deployment decisions through public frameworks, confidential reports, third-party assessments, or direct government access for evaluation.
The quotes often describe risk management as ongoing rather than one-time, calling for reassessment after mitigations and continued planning for behaviors that may emerge during deployment.

Arguments for

Internal deployment risks

Proponents argue that frontier models are often used inside developers before public release, so dangerous capability thresholds and catastrophic risks may first appear in internal settings. They describe internal use as an under-scrutinized area that can include AI R&D-specific risks and models circumventing oversight mechanisms.

To promote responsible frontier innovation, federal frontier governance should cover AI used internally by a developer in addition to any public or external offerings; there is already evidence that dangerous capability thresholds can be reached by internally deployed frontier mo…

Americans for Responsible Innovation · full quote
Risk management standards emphasise that such decisions should follow a continuous and iterative process of risk evaluation and analysis, including reassessment after mitigations are implemented. [...] Because the most cutting-edge AI systems are often deployed internally first, …

Oxford Martin AI Governance Initiative · Feb 24, 2026 · full quote
Risk management within AI companies is crucial for internal deployments of frontier AI systems for AI R&D specifically. [...] Further work is needed to develop best practices for risk management that explicitly cover the internal deployment setting, implement these best practices…

Center for Security and Emerging Technology · 2026 · full quote
reviewing assessments and adequacy of mitigations as part of the decision to deploy a frontier model or use it extensively internally; [...] assessing and managing catastrophic risk resulting from the internal use of its frontier models, including risks resulting from a frontier …

Illinois General Assembly · Jul 6, 2026 · full quote
As part of a larger body of work on understanding loss-of-control risks from frontier AI, we’ve been developing methodology to simulate how models internally deployed within AI companies might behave when performing and assisting with AI safety research.

UK AI Security Institute · Apr 27, 2026 · full quote

Early pre-deployment review

Supporters say risk assessment should begin earlier in the model lifecycle and occur before easy, broad, or extensive internal use, not only before external release. For substantially more capable or riskier models, they call for pre-deployment reviews or risk reports that explain why internal deployment is safe.

The mainstream take is that an AI lab should assess risk before deploying a new model externally. But I think risk assessment should come way earlier in the model lifecycle. Compare the situation with dangerous bio. We wouldn't find it acceptable for a virology lab to synthesize …

Miles Kodama · Sep 14, 2026 · full quote
Before approving a new model for easy or broad internal use (e.g., making the model available via our most-used internal tools), we conduct a pre-deployment review process.

Anthropic · 2026 · full quote
Whenever a substantially more capable or riskier model is deployed internally, the developer should create a risk report and argue why the model is safe to deploy.

Oscar Delaney · Apr 27, 2026 · full quote
Pre-deployment reviews include assessing benchmark results (e.g., WMDP scores) and mitigation effectiveness. For internal use, we review catastrophic risks like oversight evasion before extensive rollout.

xAI · Dec 31, 2025 · full quote
We also applied the method to challenging agentic rollouts, showing that it can extend beyond standard chat to more complex agent settings involving tool use, and can also be used for risk assessment before internal model deployments.

OpenAI · Jun 16, 2026 · full quote

Thresholds, evaluations, mitigations

Many quotes propose formal decision gates: define thresholds for catastrophic capabilities, evaluate models against those triggers, and apply mitigations based on the results. Internal deployment should proceed only after review of the assessment results and the adequacy of the mitigations.

Each large frontier developer shall begin to develop a frontier AI framework to govern how the large frontier developer will: [...] Establish thresholds to assess whether a frontier model has capabilities that could pose a catastrophic risk [...] Apply a mitigation to address the…

Melissa Shusterman · Sep 22, 2026 · full quote
Do not deploy a risk-relevant model. This includes any model deployed internally that meets or exceeds the threshold for frontier models articulated by California SB 53 (10^26 FLOPs of training compute), internally or externally, until a pre-designated decision-maker has reviewed…

Guidelight AI Standards · Sep 9, 2026 · full quote
A large frontier developer shall write, implement, comply with, and clearly and conspicuously publish on its website a frontier AI framework that applies to the large frontier developer's frontier models and describes how the large frontier developer approaches all of the followi…

Mary Edly-Allen · May 19, 2026 · full quote
Expand Frontier Safety Policies (FSPs) to explicitly cover internal AI deployment, with specific evaluations for scheming detection and control before internal use and appropriate mitigations.

Apollo Research · Apr 17, 2025 · full quote
We will also regularly assess the potential for catastrophic risk from internal use of Frontier AI models and, as appropriate, provide relevant authorities with a summary of these assessments. Frontier AI that meets the high or critical risk threshold requires additional mitigati…

Meta · Apr 8, 2026 · full quote

Oversight and transparency

Another line of support focuses on visibility around internal deployments, including published frontier AI frameworks, confidential reports to authorities, third-party assessment, and government access for evaluation before consequential internal use. The aim is to ensure that high-risk internal deployments receive review even when they are not public-facing.

Congress should establish a mandatory requirement to provide government access to, and evaluation of, frontier models before consequential internal deployment, to ensure visibility where the biggest risks are likely to emerge.

Institute for AI Policy and Strategy · Sep 11, 2026 · full quote
The developer’s publicly available frontier AI framework must describe how the developer reviews risk assessments in deciding to internally deploy the model; [...] the developer must periodically submit confidential reports to the Department of Commerce containing “a summary of a…

Institute for Progress · Aug 6, 2026 · full quote
Most importantly, third-party evaluations triggered by public deployments are not designed to cover risks from internal use within the developer itself. [...] We believe that periodic third-party assessment of risks from developers’ internal use of AI should be adopted throughout…

METR · May 19, 2026 · full quote
A large frontier developer shall write, implement, comply with, and clearly and conspicuously publish on a publicly available website of such developer a frontier AI framework that satisfies the following requirements: (1) Applies to each frontier model of such developer. (2) Inc…

Jay Obernolte · Jul 22, 2026 · full quote
A large frontier developer shall write, implement, comply with and clearly and conspicuously publish on its internet website a frontier AI framework that applies to the large frontier developer’s frontier models and describes in detail how the large frontier developer handles: [.…

Massachusetts Senate Committee on Ways and Means · Jul 16, 2026 · full quote

The provided quotes converge on the view that internal deployment of frontier models can itself be a consequential risk event and should be subject to formal assessment before rollout. The main variation is over implementation, including how to set thresholds, who reviews results, what mitigations are required, and how much third-party or government visibility should accompany internal use.

Cast your vote:

Results (31 votes):

Total (31 votes)

For 31 (100%) Abstain 0 (0%) Against 0 (0%)

For (31)

Melissa Shusterman
Pennsylvania state representative for the 157th District; primary sponsor of Pennsylvania House Bill 2800, the Artificial Intelligence Risk Prevention Act.

I am this person! Add a sourced quote Report comment

votes For and says:

Each large frontier developer shall begin to develop a frontier AI framework to govern how the large frontier developer will: [...] Establish thresholds to assess whether a frontier model has capabilities that could pose a catastrophic risk [...] App...
more AI Verified source (Sep 22, 2026)

added 11d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Miles Kodama
AI policy researcher at AI Futures

I am this person! Add a sourced quote Report comment

votes For and says:

The mainstream take is that an AI lab should assess risk before deploying a new model externally. But I think risk assessment should come way earlier in the model lifecycle. Compare the situation with dangerous bio. We wouldn't find it acceptable for...
more AI Verified source (Sep 14, 2026)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Institute for AI Policy and Strategy
Think tank developing policy recommendations for frontier artificial intelligence, safety, and national security.

I am this person! Add a sourced quote Report comment

votes For and says:

Congress should establish a mandatory requirement to provide government access to, and evaluation of, frontier models before consequential internal deployment, to ensure visibility where the biggest risks are likely to emerge.
AI Verified source (Sep 11, 2026)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Chris Lehane
OpenAI chief global affairs officer

I am this person! Add a sourced quote Report comment

votes For and says:

We have strengthened monitoring, alignment, and security safeguards across the model-development lifecycle, including stronger isolation for frontier research workloads, expanded monitoring of model behavior during tool-enabled training and evaluatio...
more AI Verified source (Sep 9, 2026)

added 2d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Guidelight AI Standards

I am this person! Add a sourced quote Report comment

votes For and says:

Do not deploy a risk-relevant model. This includes any model deployed internally that meets or exceeds the threshold for frontier models articulated by California SB 53 (10^26 FLOPs of training compute), internally or externally, until a pre-designat...
more AI Verified source (Sep 9, 2026)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Institute for Progress
Nonpartisan think tank focused on innovation policy, economic growth, and technological progress.

I am this person! Add a sourced quote Report comment

votes For and says:

The developer’s publicly available frontier AI framework must describe how the developer reviews risk assessments in deciding to internally deploy the model; [...] the developer must periodically submit confidential reports to the Department of Comme...
more AI Verified source (Aug 6, 2026)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Jay Obernolte
U.S. Representative from California

I am this person! Add a sourced quote Report comment

votes For and says:

A large frontier developer shall write, implement, comply with, and clearly and conspicuously publish on a publicly available website of such developer a frontier AI framework that satisfies the following requirements: (1) Applies to each frontier mo...
more AI Verified source (Jul 22, 2026)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Massachusetts Senate Committee on Ways and Means
Massachusetts Senate committee that reviews fiscal legislation and issues recommendations to the Senate.

I am this person! Add a sourced quote Report comment

votes For and says:

A large frontier developer shall write, implement, comply with and clearly and conspicuously publish on its internet website a frontier AI framework that applies to the large frontier developer’s frontier models and describes in detail how the large ...
more AI Verified source (Jul 16, 2026)

added 7d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Illinois General Assembly
State legislature of Illinois, United States.

I am this person! Add a sourced quote Report comment

votes For and says:

reviewing assessments and adequacy of mitigations as part of the decision to deploy a frontier model or use it extensively internally; [...] assessing and managing catastrophic risk resulting from the internal use of its frontier models, including ri...
more AI Verified source (Jul 6, 2026)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
OpenAI
AI research organization

I am this person! Add a sourced quote Report comment

votes For and says:

We also applied the method to challenging agentic rollouts, showing that it can extend beyond standard chat to more complex agent settings involving tool use, and can also be used for risk assessment before internal model deployments.
AI Verified source (Jun 16, 2026)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Mary Edly-Allen
Illinois State Senator (D-Libertyville); lead sponsor of SB 315, a frontier AI safety and transparency bill requiring independent third-party audits of major AI developers

I am this person! Add a sourced quote Report comment

votes For and says:

A large frontier developer shall write, implement, comply with, and clearly and conspicuously publish on its website a frontier AI framework that applies to the large frontier developer's frontier models and describes how the large frontier developer...
more AI Verified source (May 19, 2026)

added 14d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
METR
Nonprofit AI safety research organization that evaluates frontier AI systems.

I am this person! Add a sourced quote Report comment

votes For and says:

Most importantly, third-party evaluations triggered by public deployments are not designed to cover risks from internal use within the developer itself. [...] We believe that periodic third-party assessment of risks from developers’ internal use of A...
more AI Verified source (May 19, 2026)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Alex Mallen
AI safety researcher at Redwood Research.

I am this person! Add a sourced quote Report comment

votes For and says:

Risk reports commonly use pre-deployment alignment assessments to measure misalignment risk from an internally deployed AI. However, an AI [...] can develop widespread dangerous motivations during deployment. I think [...] AI companies and evaluators...
more AI Verified source (May 15, 2026)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
UK AI Security Institute
United Kingdom government institute for evaluating advanced AI systems.

I am this person! Add a sourced quote Report comment

votes For and says:

As part of a larger body of work on understanding loss-of-control risks from frontier AI, we’ve been developing methodology to simulate how models internally deployed within AI companies might behave when performing and assisting with AI safety resea...
more AI Verified source (Apr 27, 2026)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Google DeepMind
AI research company.

I am this person! Add a sourced quote Report comment

votes For and says:

A model for which the inherent risk assessment indicates a ML R&D CCL has been reached will be deemed to pose an acceptable level of residual risk for further development or deployment, if: We assess that the deployment mitigations have brought the r...
more AI Verified source (Apr 17, 2026)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Meta
Social media and AI company

I am this person! Add a sourced quote Report comment

votes For and says:

We will also regularly assess the potential for catastrophic risk from internal use of Frontier AI models and, as appropriate, provide relevant authorities with a summary of these assessments. Frontier AI that meets the high or critical risk threshol...
more AI Verified source (Apr 8, 2026)

added 12d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Gregory A. Miller
Louisiana state senator and primary sponsor of the 2026 Protecting Louisiana's Infrastructure from Artificial Intelligence Risk Act.

I am this person! Add a sourced quote Report comment

votes For and says:

A large frontier developer shall write, implement, comply with, and clearly and conspicuously publish on its internet website a frontier AI framework that applies to the large frontier developer's frontier models and describes in detail how the large...
more AI Verified source (Mar 31, 2026)

added 20d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Oxford Martin AI Governance Initiative
Oxford Martin research initiative studying governance and risk management for frontier artificial intelligence.

I am this person! Add a sourced quote Report comment

votes For and says:

Risk management standards emphasise that such decisions should follow a continuous and iterative process of risk evaluation and analysis, including reassessment after mitigations are implemented. [...] Because the most cutting-edge AI systems are oft...
more AI Verified source (Feb 24, 2026)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Kimberly Du Buclet
Illinois state representative for the 5th District; sponsor of the 2026 AI Safety Measures Act.

I am this person! Add a sourced quote Report comment

votes For and says:

A large frontier developer shall write, implement, comply with, and clearly and conspicuously publish on its website a frontier artificial intelligence framework that applies to the large frontier developer’s frontier models and describes how the lar...
more AI Verified source (Feb 2, 2026)

added 10d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Jason Zachary
Tennessee state representative for District 14 and Deputy Speaker of the Tennessee House of Representatives.

I am this person! Add a sourced quote Report comment

votes For and says:

This bill requires a large frontier developer to implement and clearly publish on its website a public safety plan that describes in detail how the large frontier developer defines and assesses thresholds used by the developer to determine whether a ...
more AI Verified source (Jan 22, 2026)

added 7d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Andrew Gounardes
New York state senator

I am this person! Add a sourced quote Report comment

votes For and says:

A large frontier developer shall write, implement, comply with, and clearly and conspicuously publish on its internet website a frontier AI framework that applies to the large frontier developer's frontier models and describes in detail how the large...
more AI Verified source (Jan 8, 2026)

added 18d 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 large frontier developer shall write, implement, comply with, and clearly and conspicuously publish on its internet website a frontier AI framework that applies to the large frontier developer's frontier models and describes in detail how the large...
more AI Verified source (Jan 6, 2026)

added 3d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Nebraska Legislature
Nebraska's unicameral state legislature, publishing legislative bills and official records.

I am this person! Add a sourced quote Report comment

votes For and says:

A large frontier developer [...] shall write, implement, comply with, and clearly and conspicuously publish on its website a public safety and child protection plan that describes in detail [...] defines and assesses thresholds [...] to identify and ...
more AI Verified source (2026)

added 20d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Center for Security and Emerging Technology
Georgetown University research center focused on emerging technology and national security.

I am this person! Add a sourced quote Report comment

votes For and says:

Risk management within AI companies is crucial for internal deployments of frontier AI systems for AI R&D specifically. [...] Further work is needed to develop best practices for risk management that explicitly cover the internal deployment setting, ...
more AI Verified source (2026)

added 21d 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:

Before approving a new model for easy or broad internal use (e.g., making the model available via our most-used internal tools), we conduct a pre-deployment review process.
AI Verified source (2026)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
xAI
Artificial intelligence company developing the Grok family of frontier AI models.

I am this person! Add a sourced quote Report comment

votes For and says:

Pre-deployment reviews include assessing benchmark results (e.g., WMDP scores) and mitigation effectiveness. For internal use, we review catastrophic risks like oversight evasion before extensive rollout.
AI Verified source (Dec 31, 2025)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
California State Legislature
State legislative body

I am this person! Add a sourced quote Report comment

votes For and says:

A large frontier developer shall write, implement, comply with, and clearly and conspicuously publish on its internet website a frontier AI framework that ... [reviews] assessments and adequacy of mitigations as part of the decision to deploy a front...
more AI Verified source (Sep 29, 2025)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Apollo Research
AI safety research organization

I am this person! Add a sourced quote Report comment

votes For and says:

Expand Frontier Safety Policies (FSPs) to explicitly cover internal AI deployment, with specific evaluations for scheming detection and control before internal use and appropriate mitigations.
AI Verified source (Apr 17, 2025)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Government of the United Kingdom
United Kingdom central government

I am this person! Add a sourced quote Report comment

votes For and says:

Conducting thorough risk assessments before developing or deploying new models, complemented with continual monitoring (incorporating, for example, evaluation results, evidence from earlier models, and expert forecasts). [...] For example, at a given...
more AI Verified source (2023)

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Americans for Responsible Innovation
AI policy advocacy nonprofit

I am this person! Add a sourced quote Report comment

votes For and says:

To promote responsible frontier innovation, federal frontier governance should cover AI used internally by a developer in addition to any public or external offerings; there is already evidence that dangerous capability thresholds can be reached by i...
more AI Verified source

added 21d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate