Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-07-23 · high
It beats seven existing defenses and keeps 97 percent of a model's performance, aimed at backdoored weights downloaded from Hugging Face.
Why now: The paper's arXiv listing dates it to July 2026.
DeCNIP removes backdoors from language models by pruning just 0.1% of their neurons, per the paper posted to arXiv.
Backdoors planted through model editing skip the fine-tuning stage most existing defenses are built to catch. Few of those defenses extend past simple classification into open-ended text generation, either. That's the gap that matters if you're pulling a checkpoint off Hugging Face and trusting it's clean. DeCNIP claims to close it by touching just 0.1% of a model's neurons.
The method optimizes a cross-entropy loss between harmful prompts paired with candidate trigger tokens and ordinary benign inputs. That surfaces the neurons that only light up when a trigger's present. It labels those the Backdoor Critical Neurons and prunes only those, leaving the rest of the model untouched.
Tested across six open-source LLMs and two benchmarks, it cut attack success by more than 95% relative to the unprotected models, beating seven state-of-the-art defenses. The pruned models still kept 97% of normal benchmark performance intact. The paper doesn't say which six models or which benchmarks, so how this holds up against the largest open-weight releases people download is still unknown.
A fix that touches 0.1% of neurons and costs a few points of performance sounds too cheap to be the last word. My bet: the next generation of backdoor attacks spreads its trigger across more, less concentrated neurons to dodge isolation. That would break DeCNIP's core assumption that a backdoor lives in a small, identifiable cluster.
Each link below shares sources, entities, or timing with this story.
LLM uses OpenAI / Shared entity: Hugging Face / Same source domain / Tension
Linked by a graph relationship (LLM uses OpenAI); both cover Hugging Face; reported by the same outlet (arxiv.org).
Simon Willison released LLM / Shared entity: LLM / Earlier coverage / Tension
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-06-19.
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-06-18.
Simon Willison released LLM / Shared entity: LLMs / Earlier coverage / Tension
Linked by a graph relationship (Simon Willison released LLM); both cover LLMs; earlier LLMs coverage from 2026-03-23.
LLM uses OpenAI / Shared entity: LLM / Earlier coverage
Linked by a graph relationship (LLM uses OpenAI); both cover LLM; earlier LLM coverage from 2026-06-19.
Hugging Face partners with NVIDIA / Shared entity: Hugging Face / Earlier coverage
Linked by a graph relationship (Hugging Face partners with NVIDIA); both cover Hugging Face; earlier Hugging Face coverage from 2026-04-24.
Simon Willison released LLM / Shared entity: LLM / Earlier coverage
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-07-19.
Linked by a graph relationship (Simon Willison released LLM); both cover LLM; earlier LLM coverage from 2026-07-14.