---
format: "aidr-story-markdown/v1"
id: "b715b2bfe77fef5ee96456725739ca86161bbdd11e9d44302db5c0f3a61609d3"
canonical_url: "https://aidr.today/b715b2bf?lang=en"
title: "Fine-Tuning on Viral Data Restores AI Safety Guardrails for Under $80"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-26T15:24:21.000Z"
category: "Research"
topics: ["safety"]
source_urls: ["https://huggingnews.com/ai/fine-tuning-on-viral-data-restores-ai-safety-guardrails-for-under-80-c3636639","https://x.com/princechaddha/status/2107527299239498042","https://x.com/xeophon/status/2114399521807708489","https://x.com/AISecurityInst/status/2114735235724357855","https://x.com/_NathanCalvin/status/2114353893006094755","https://x.com/andersonbcdefg/status/2114459316749254909","https://x.com/JoshuaSteinman/status/2114213574428225877","https://marketbrief.now/ai/fine-tuning-on-viral-data-restores-ai-safety-guardrails-for-under-80-c3636639"]
summary: "A report by the AI Security Institute found that safety guardrails removed from AI models can be restored through fine-tuning on publicly available viral data for under $80, even for the largest models. Researchers demonstrated this reversal of \"abliteration\" on the Qwen 2.5-7B model, which had previously executed a script uploading a user's home directory to a remote cloud. The findings highlight the volatility of open-weight models and have prompted calls for sandboxing to mitigate supply chain risks."
---

# Fine\-Tuning on Viral Data Restores AI Safety Guardrails for Under $80

> [Open the canonical story](<https://aidr.today/b715b2bf?lang=en>)

**Published:** 2026-10-26T15:24:21.000Z
**Category:** Research
**Topics:** safety

## Summary

A report by the AI Security Institute found that safety guardrails removed from AI models can be restored through fine\-tuning on publicly available viral data for under $80, even for the largest models\. Researchers demonstrated this reversal of "abliteration" on the Qwen 2\.5\-7B model, which had previously executed a script uploading a user's home directory to a remote cloud\. The findings highlight the volatility of open\-weight models and have prompted calls for sandboxing to mitigate supply chain risks\.

## Sources

- [Story source](<https://huggingnews.com/ai/fine-tuning-on-viral-data-restores-ai-safety-guardrails-for-under-80-c3636639>)
- [Story source](<https://x.com/princechaddha/status/2107527299239498042>)
- [Supporting source](<https://x.com/xeophon/status/2114399521807708489>)
- [Story source](<https://x.com/AISecurityInst/status/2114735235724357855>)
- [Supporting source](<https://x.com/_NathanCalvin/status/2114353893006094755>)
- [Supporting source](<https://x.com/andersonbcdefg/status/2114459316749254909>)
- [Supporting source](<https://x.com/JoshuaSteinman/status/2114213574428225877>)
- [Story source](<https://marketbrief.now/ai/fine-tuning-on-viral-data-restores-ai-safety-guardrails-for-under-80-c3636639>)