The BaseNewsTransformer

Hugging Face appears to be hosting a load of models that could be used for cyber attacks

Transformer reports that Hugging Face hosts thousands of models with safety filters removed, some marketed for offensive cyber work. It says the platform took down one such model in October, and quotes security experts and abliteration developers on the risks and legitimate uses.

Open the source ↗3 quotations checked, word for word ↓
About Anthropic or ClaudeClaude, made by Anthropic, wrote this entry. It follows the same rules as every other entry, whoever the entry is about.

Why it matters

The same thing, for different readers.

What it means for you

Freely downloadable AI models with safety filters stripped out are easy to find, and some are marketed for hacking; Hugging Face has not explained where it draws the line.

For lawmakers

The report raises the question of how platforms should enforce content policies on uncensored open-weight models, now that Nvidia owns Hugging Face.

For researchers

Andrew Arditi, who helped found the science of abliteration, says little work exists on a middle path between fully controlled and fully unrestricted models.

The question it raises

The piece sets openness and accountability against the loss of any control over distribution once model weights are public.

In their own words

Every quotation was found, word for word, in the text we saved from the source. That shows the words are there; it does not make them true, or prove the speaker meant them as quoted.

From the source’s own text.

“The open-weight models are getting almost as good as frontier models, and now frontier capability is available to anyone, regardless of intent, with no guardrails”
George Kurtz, CEO, CrowdStrikeKurtz speaking at a conference in Barcelona this month, as quoted by Transformer.
Found word for word in the sourceAs quoted in the text we read, not the speaker’s own text or a transcript of their words
“It’s not altogether helpful that they’re available to one and all through Hugging Face, and that may be where there needs to be some conversations with the likes of Hugging Face.”
Alan Woodward, cybersecurity expert, University of SurreyWoodward on distributing models with safeguards removed, as quoted by Transformer.
Found word for word in the sourceAs quoted in the text we read, not the speaker’s own text or a transcript of their words
“There really isn’t currently much ideation or discourse on what a middle path might look like.”
Andrew Arditi, AI researcherArditi on options between developer-controlled models and models stripped of restrictions, as quoted by Transformer.
Found word for word in the sourceAs quoted in the text we read, not the speaker’s own text or a transcript of their words

How the finding was reached

In our words, from the source. Weigh the finding by how it was made.

What was done
A reporter's survey of Hugging Face listings and interviews with security experts and developers; Transformer counted models tagged abliterated, uncensored or heretic.
What it found
More than 6,500 such models were hosted in July and many more remain; Hugging Face removed one GLM-5.3 version marketed for offensive cyber in early October, while others with similar marketing remain.
What it does not show
Tags show safety filters were removed, not that a model was designed for attacks; Hugging Face did not respond on why it removed one and not others.
Status
A report

The record

Questions
Are AI companies putting stronger safeguards into practice? · How is AI used in attacks, defence and war? · Who can demand evidence, delay a release or stop an unsafe deployment?
People
Chris Stokel-Walker (Wrote it) · George Kurtz (Speaks, CEO, CrowdStrike) · Alan Woodward (Speaks, cybersecurity expert, University of Surrey) · Philipp Emanuel Weidmann (Speaks, developer of Heretic) · Andrew Arditi (Speaks, AI researcher) · Catherine Flick (Speaks, professor of ethics and games technology, University of Staffordshire)
What we read
We read the whole source.
Found by
The Radar, run of November 12, 2026, via transformer (tier 1: the source’s own channel or an official record)
Cite this
Transformer, “Hugging Face appears to be hosting a load of models that could be used for cyber attacks”, November 11, 2026. Entry in the Ctrl AI Base, https://ctrlai.com/base/transformer-2026-11-11-hugging-face-abliterated-models
Our notesMentions that Anthropic found GLM-5.3 could develop browser exploits in 12% of attempts in September; not verified against the original. Hugging Face did not comment.