Pith. sign in

REVIEW 6 cited by

Scaling Trends for Data Poisoning in LLMs

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2408.02946 v6 pith:77SGTNLD submitted 2024-08-06 cs.CR cs.AIcs.LG

classification cs.CRcs.AIcs.LG
keywords datamodelsllmspoisoningevenfine-tuningharmfulpoisoned
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

LLMs produce harmful and undesirable behavior when trained on datasets containing even a small fraction of poisoned data. We demonstrate that GPT models remain vulnerable to fine-tuning on poisoned data, even when safeguarded by moderation systems. Given the persistence of data poisoning vulnerabilities in today's most capable models, this paper investigates whether these risks increase with model scaling. We evaluate three threat models -- malicious fine-tuning, imperfect data curation, and intentional data contamination -- across 24 frontier LLMs ranging from 1.5 to 72 billion parameters. Our experiments reveal that larger LLMs are significantly more susceptible to data poisoning, learning harmful behaviors from even minimal exposure to harmful data more quickly than smaller models. These findings underscore the need for leading AI companies to thoroughly red team fine-tuning APIs before public release and to develop more robust safeguards against data poisoning, particularly as models continue to scale in size and capability.

Discussion (0). Sign in to comment.

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Emergent Misalignment Recruits a Pre-existing Persona Subspace

    cs.LG 2026-07 conditional novelty 7.0 of 10

    Fine-tuning on narrow bad data recruits a low-rank persona subspace already present in a frozen instruction-tuned model; holding that subspace out of activations prevents broad misalignment, and injecting it into the ...

  2. The Safety Gap Toolkit: Evaluating Hidden Dangers of Open-Source Models

    cs.CY 2025-07 conditional novelty 6.0 of 10

    On Llama-3 and Qwen-2.5, removing safety guardrails sharply raises compliance with dangerous bio, chem, and cyber requests, and the resulting safety gap grows with model scale.

  3. Emergent misalignment as prompt sensitivity: A research note

    cs.CR 2025-07 conditional novelty 6.0 of 10

    Insecure-code models are highly prompt-sensitive: easily nudged into misalignment, slightly nudgeable toward helpfulness, and prone to sycophantic factual recall errors.

  4. Fine-Tuning Lowers Safety and Disrupts Evaluation Consistency

    cs.CL 2025-06 conditional novelty 4.0 of 10

    Fine-tuning small LLMs on benign data raises harmfulness scores, but those scores vary widely across random seeds, temperatures, and repeated runs, making single-run safety comparisons unreliable.

  5. Veracity: An Open-Source AI Fact-Checking System

    cs.CL 2025-06 conditional novelty 4.0 of 10

    Veracity is an open-source LLM-plus-web-search fact-checking app with a 0 to 100 reliability score and explanations, but no evaluation of its accuracy is included.

  6. Breaking to Build: A Threat Model of Prompt-Based Attacks for Securing LLMs

    cs.CL 2025-09 conditional novelty 3.0 of 10

    A survey categorizing prompt-based attacks on LLMs into four classes and proposing aspirational goals of un-distillable, un-finetunable, and un-editable models.

Pith tools