Ethical Stack.

Grok 2 weights

The downloadable Grok 2 checkpoint from xAI, with self-hosting instructions and a custom model license.

Score calculation

(2 + 2 + 6 + 2 + 3) ÷ 5 = 3.0

5 of six dimensions rated. Unrated dimensions are excluded from the average.

Full scoring method ↗
Setup details

The published xai-org/grok-2 checkpoint and Grok 2 Community License. Hosted Grok versions have separate service terms.

What fits your needs

Downloaded xai-org/grok-2 weights and a self-operated SGLang deployment; hosted Grok 4.7 and xAI API terms are separate.

Self-hosting

The official repository provides the weights and a SGLang launch example for operators who can provision the required multi-GPU server.

See the evidence
Tradeoffs to expect

Hardware

The repository lists about 539 GB of files; its serving example calls for eight GPUs with more than 40 GB memory each, so this is not a typical personal-computer deployment.

See the evidence

License

The custom license permits use under its terms, but limits use of the model, derivatives, or outputs to train other general-purpose models and requires xAI attribution when distributing materials.

See the evidence
How to switch

Download and serve

The model card shows `hf download xai-org/grok-2 --local-dir /local/grok-2` and an SGLang server command using tensor parallelism 8 and FP8 quantization.

See the evidence
Practical questions

Checkpoint identity

The official repository names this checkpoint Grok 2 and says it was trained and used at xAI in 2024. Check the model card and license for the downloaded release.

See the evidence

License terms

Review the current Grok 2 Community License and xAI Acceptable Use Policy before deployment or redistribution; commercial use is conditional on the policy guardrails.

See the evidence

Hosted route

The official model page says this checkpoint is not deployed by an Inference Provider; that listing does not rule out separate xAI hosted models or API routes.

See the evidence

Energy practices

2/10↑

This grades documented operating information and controls, not electricity, emissions, or water for a generation.

See the evidence

The model card documents an SGLang deployment with tensor parallelism across eight GPUs, each with more than 40 GB of memory, but provides no energy, emissions, or water measurements. These operating details make the serving boundary clearer without measuring its environmental impact.

1 + 1 reporting + 0 electricity + 0 controls = 2/10. This grades documented practices and transparency.

The cited release does not report training or inference electricity, emissions, water, facility sourcing, or a workload-level footprint.

Reporting1/3
The official serving instructions specify SGLang, tensor parallelism of eight, and GPUs with more than 40 GB memory each; they do not provide physical footprint measurements.
Electricity sourcing0/3
The reviewed release doesn’t confirm clean-electricity sourcing or a regional energy mix; zero records that evidence gap, not dirty electricity.
Resource controls0/3
The serving example documents hardware requirements but no workload-control practice or measured reduction in physical resource use.

Water disclosure

Water reporting has not been assessed for this choice.

Training data

2/10↑

xAI's general Grok description names public sources and human-reviewed datasets, but does not give a checkpoint-specific training dataset or source-level permission record for this release.

See the evidence

X says Grok pretraining uses publicly available sources and datasets reviewed by human AI Tutors. The Grok 2 repository does not itemize the sources or show their permissions; a license for using released weights is not evidence that every training source was authorized.

1 base + 1 limited source description + 0 documented permission + 0 creator controls = 2/10. Weight licensing is separate from training data permission.

A Grok 2-specific dataset inventory, source-level permissions, inherited data provenance, and creator removal or retraining controls are not established in the checked materials.

Source transparency1/3
The official Grok description identifies publicly available sources and datasets reviewed by human AI Tutors, but the release has no itemized Grok 2 source manifest.
Documented permission0/4
The checked materials don’t confirm source-level licenses or permissions for the Grok 2 training data; availability of the weight license does not answer that question.
Creator controls0/2
The checked release materials do not document a creator removal or model retraining remedy specific to the Grok 2 training data.
Data-rights rubric ↗

Your data & privacy

—Not enough evidence↑

The published weights can be run by a separate operator; that deployment does not have one xAI retention policy. The hosted consumer and API routes are assessed separately.

See the evidence

The repository provides weights and self-hosting instructions, but doesn’t show the logging, retention, telemetry, or access practices of each operator. The current hosted Grok service is a different route, so a single privacy score would overstate what this model release establishes.

A particular self-hosted operator's logs, region, access controls, and any connected applications are not established by the checkpoint release.

Openness

6/10↑

xAI publishes downloadable weights, files, and serving instructions under its custom Grok 2 Community License. The release does not include a full training dataset or reproducible training pipeline.

See the evidence

The official repository makes the checkpoint files and deployment instructions available, but use is governed by a custom license and the checked release does not provide the complete training data or a reproducible training recipe. This rating concerns artifacts and reproducibility, not whether the original training data was authorized.

Published weights and serving files support reuse; custom license conditions and missing training data and recipe limit openness and reproducibility.

A full source-level training dataset, end-to-end training code, and independent reproduction results are not established by the repository.

Worker protections

2/10↑

xAI publicly describes direct-employee benefits and flexible vacation, while reporting confirms a sizable annotation workforce and major 2025 layoffs; contractor conditions, voice and labor assurance remain undisclosed.

See the evidence

xAI’s careers page advertises health, dental, vision and disability coverage, life/family benefits and flexible vacation for direct hires. Separately, Reuters/TechCrunch reported in September 2025 that xAI laid off at least 500 generalist data annotators in a strategic pivot, with continued payment through contract end or a stated date but access ending immediately. That is evidence of a real annotation workforce and substantial churn, not proof of unlawful treatment or poor conditions. No public supplier labor code, contractor pay terms, worker-specific grievance mechanism, worker survey or labor-assurance results were established. These findings concern the developer or operator. This product’s use in a particular deployment needs separate evidence.

1 + 1 pay & conditions + 0 worker voice & remedy + 0 evidence & follow-through = 2/10. This grades documented worker safeguards, not a measured rate of mistreatment.

The careers benefits are for direct applicants and don’t show benefits or coverage for annotation contractors. Layoff reporting doesn’t show current staffing, severance, worker voice or supplier labor standards; no xAI-specific labor code or worker outcomes were found.

Pay & conditions1/3
xAI advertises health/disability insurance and flexible vacation for direct hires; this is a limited company commitment and doesn’t confirm conditions for the laid-off annotation contractors.
Worker voice & remedy0/3
No public worker-specific grievance, opt-out or representation mechanism was established; the layoffs do not show individual voice or remedy.
Evidence & follow-through0/3
No public annotation-specific labor audits, worker surveys, remediation or independently verified outcome mechanisms were found.
Workers rubric ↗

Military ties

3/10Broad defense ties↑

xAI offers a government product line, a $200 million-ceiling DoD contract, and customized AI for national-security customers, including classified settings and intelligence and operational-planning use cases. No weapons or surveillance limits are disclosed in the reviewed government materials.

See the evidence

xAI's government announcement confirms a $200 million-ceiling DoD contract, classified-environment plans, custom models for national-security customers, and federal access to its products. Its current government page markets intelligence analysis, operational planning, and defense logistics. This is broad defense enablement capped at 3/10; public materials describe secure deployment controls but no restrictions on weapons or harmful surveillance and no independent military-use review. These findings concern the developer or operator. This product’s use in a particular deployment needs separate evidence.

1 + 0 use restrictions + 0 enforcement & oversight + 2 relationship disclosure = 3/10. Known defense ties lower this rating; safeguards can improve it within the relationship’s cap.

The award announcement and product page do not identify contract task orders, operational customer use, restrictions, model versions, or oversight outcomes. Availability in classified environments does not prove use of a specific model or combat task.

Use restrictions0/4
The reviewed government materials advertise lawful mission use and secure controls but do not disclose restrictions on military, weapons, or surveillance use.
Enforcement & oversight0/3
Audit logging and access controls are product security features; no military-use review, enforcement, or public remedy outcomes are established.
Relationship disclosure2/2
xAI names its DoD contract ceiling, classified deployment plans, customer groups, and defense/intelligence use cases; task-level deployments are not public.

xAI confirms DoD contracting and offers national-security models for classified environments, with intelligence, operational planning, and defense-logistics applications but no verified specific combat deployment. The relationship limits this grade to 3/10.

Military rubric ↗

BDS targets

Not listed↑

We did not find xAI among the named targets in the two official BNC campaign lists. This is a limited list check, not BDS approval.

See the evidence

xAI is not a named target in the checked lists, but BDS uses selective campaigns, so absence cannot establish a clean record. Running downloaded weights locally is a different decision from paying the developer or a cloud host; check the host as well. These findings concern the developer or operator. This product’s use in a particular deployment needs separate evidence.

This review covers the named targets in the BNC consumer/pressure guide and No Tech campaign, not every BDS campaign, local group, investment or supplier. Lists and ownership can change. A company’s absence from these lists is not an endorsement or proof that it has no relevant relationships.

More from xAI

All sources

Side by side