Loading the catalogue…
Unknown origin · Mixed · 0 models
Anthracite
Anthracite is an informal, pseudonymous online collective with no verifiable headquarters, legal entity, or governance structure, operating entirely through HuggingFace and Open Collective crowdfunding; its jurisdiction cannot be determined from any public source. Its Magnum model series (v1–v4) are openly released fine-tunes built on top of third-party base models — Qwen 2.5, Mistral-Nemo, Gemma 2, and Yi — using training datasets that include Claude-output-derived data, raising potential IP and Anthropic Terms of Service concerns that EU regulated customers should evaluate carefully. The complete absence of any privacy policy, DPA counterparty, GDPR documentation, security certifications, or EU AI Act compliance documentation means the organisation does not meet baseline enterprise compliance requirements for regulated sectors; deployment is suitable only for experimental, air-gapped, or research contexts where no personal data is processed.
AI Act 13/100Outside CLOUD Act
Stav AI Act verdict
governancecritical
Anthracite is an anonymous, unincorporated volunteer collective with no legal entity, no named leadership, no registered address, and no determinable jurisdiction. There is no legal counterparty with which a regulated enterprise can execute a DPA, no liability chain, and no organisational continuity guarantee. This is a fundamental blocker for enterprise deployment in regulated EU sectors.
privacycritical
No privacy policy, GDPR documentation, DPA template, DPO contact, or data protection infrastructure exists. GDPR compliance is entirely unverifiable for any data processed in connection with this creator's models.
compliancehigh
Magnum training datasets across v3 and v4 series include Claude-output-derived data (kalo-opus-instruct-22k-no-refusal, nopm_claude_writing_fixed, c2_logs, Sonnet3.5-SlimOrcaDedupCleaned subsets). Anthropic's usage policies explicitly prohibit using Claude outputs to train competing models. This creates potential IP and terms-of-service chain-of-title risk for downstream deployers.
compliancehigh
No EU AI Act compliance documentation: no compliance statement, no Article 53 GPAI transparency documentation, no formal training data summary meeting regulatory standards, and no Code of Practice engagement. Models are not ready for EU regulated deployment on compliance grounds.
jurisdictionmedium
Jurisdiction is entirely unverifiable — all contributors are pseudonymous. The current CLOUD Act exposure value of 'false' cannot be positively confirmed. Infrastructure spans US-linked platforms (HuggingFace, Open Collective, Featherless/Recursal AI). EU customers cannot rely on jurisdictional clarity.
securitymedium
No security certifications (SOC 2, ISO 27001), no bug bounty programme, no responsible disclosure policy, and no described security practices. Security posture is entirely opaque.
governancemedium
Organisational stability is inherently fragile: the project is sustained entirely by volunteer effort and community crowdfunding via Open Collective. There is no formal continuity mechanism if key contributors become inactive.
compliancemedium
Stacked licence complexity: Magnum models inherit licences from multiple base models (Qwen 2.5 Research Licence — restricts commercial use at scale, Gemma Terms of Use, Mistral Apache 2.0, Yi Apache 2.0) in addition to Anthracite's own licence (Apache 2.0 for some models). Downstream enterprise customers must perform their own licence compatibility analysis.
Stav AI Act assessment
Editorial assessment, not legal advice. Stav's risk ratings, scores, and verdicts are our own analysis of publicly available information and may be incomplete or out of date. Verify independently before making compliance or procurement decisions.
Safeguards in place
- transparencyUnknown
Training datasets and full Axolotl training configurations are publicly disclosed in model READMEs, providing more training-side transparency than many proprietary labs — even if this transparency paradoxically surfaces the IP concern.
- transparencyUnknown
Financial transparency via Open Collective: all donations and expenditures are publicly visible, with no hidden corporate funding or undisclosed investor relationships.
- transparencyUnknown
Open-weights release: all Magnum models (v1–v4 series, including quantised GGUF and exl2 variants) are published with downloadable weights on HuggingFace, enabling independent evaluation, auditing, and fully local deployment without any creator-side API contact.
- communityUnknown
Active HuggingFace community engagement with ongoing technical discussion and responsiveness to community feedback documented in org activity feed.
- communityUnknown
Endorsed by Featherless/Recursal AI as a compute sponsor and distribution partner, indicating third-party confidence in the project's technical output quality.
- stabilityUnknown
Consistent model iteration (v1 through v4 series across multiple base models) demonstrates sustained development activity and progressive technical capability improvement since at least August 2024.
Published safeguards & certifications
Privacy posture
- Flagged issues
- 1
- No privacy policy, GDPR documentation, DPA template, DPO contact, or data protection infrastructure exists. GDPR compliance is entirely unverifiable for any data processed in connection with this creator's models.critical
Privacy policy review
Identity & profile
- Open-weights policy
- Mixed
- Models in catalogue
- 0
- HuggingFace
- anthracite-org · 515 followers
- HF published
- 55 models · 20 datasets · 0 papers
Creator profile
About Anthracite
Anthracite is an informal, pseudonymous online collective with no verifiable headquarters, legal entity, or governance structure; it operates entirely through HuggingFace and Open Collective crowdfunding, and its jurisdiction cannot be determined from any public source. Its Magnum model series are openly released fine-tunes built on top of third-party base models (Qwen 2.5, Mistral-Nemo, Gemma 2) using datasets that include Claude-derived outputs, raising potential IP and Anthropic ToS concerns that EU regulated customers should evaluate carefully. The absence of any privacy policy, DPA, GDPR data-protection contact, security certifications, or EU AI Act documentation means the organisation does not meet baseline enterprise compliance requirements for regulated sectors.
Stav editorial summary
COMPLIANCE & TRUST
Stav’s creator score.
Stav compliance has not yet scored Anthracite. Scores are published once the policy review and infrastructure assessment complete.
Findings
Findings from privacy & terms
Citations gathered when the Compliance Curator last reviewed this creator’s public-facing documents. Grouped by source so the picture stays auditable.
huggingface.co · 3 citations
“If you'd like to support what we do: https://opencollective.com/anthracite-org Anthracite, also known as hard coal and black coal, is a hard, compact...”
“This is a series of models designed to replicate the prose quality of the Claude 3 models, specifically Sonnet and Opus. ”
“We'd like to thank Recursal / Featherless for sponsoring the compute for this train, Featherless has been hosting our Magnum models since the fir...”
Editorial sources
- anthracite-org (Anthracite) - Hugging Face
If you'd like to support what we do: https://opencollective.com/anthracite-org Anthracite, also known as hard coal and black coal, is a hard, compact...
- anthracite-org/magnum-v2-12b · Hugging Face
This model has been a team effort, and the credits goes to all members of Anthracite.
- anthracite-org/magnum-v4-27b · Hugging Face
This is a series of models designed to replicate the prose quality of the Claude 3 models, specifically Sonnet and Opus.
- anthracite-org/magnum-v4-72b · Hugging Face
base_model: /workspace/data/models/Qwen2.5-72B-Instruct model_type: AutoModelForCausalLM tokenizer_type: AutoTokenizer plugins: - axolotl.integrations...
- magnum-v2-12b | AI Model Details
It is designed to replicate the high-quality prose of the Claude 3 models, specifically the Sonnet and Opus models. The model is fine-tuned on top of ...
- anthracite-org/magnum-v3-34b · Hugging Face
We used 8xH100s GPUs graciously provided by Recursal AI / Featherless AI for the full-parameter fine-tuning of the model.
Reference citations
- anthracite-org (Anthracite) - Hugging Face
If you'd like to support what we do: https://opencollective.com/anthracite-org Anthracite, also known as hard coal and black coal, is a hard, compact...
- anthracite-org/magnum-v4-72b · Hugging Face
This is a series of models designed to replicate the prose quality of the Claude 3 models, specifically Sonnet and Opus.
- AI model creator details for anthracite-org
The project builds on earlier iterations like magnum-v2-12b and magnum-v2-123b, trained on 55 million tokens of high-quality data over 1.5 epochs.
- AI Model Distillation Attacks: What They Are and Why They Matter | MindStudio
Anthropic’s usage policies explicitly prohibit using Claude outputs to train models that compete with Claude.
- anthracite-org/magnum-v4-12b Free Chat Online - skywork.ai - Skywork ai
Magnum V4 12b is released under the Apache-2.0 license, which permits commercial use, modification, distribution, and private use without royalty paym...
- anthracite-org/magnum-v4-12b · Hugging Face
We'd like to thank Recursal / Featherless for sponsoring the compute for this train, Featherless has been hosting our Magnum models since the fir...