Qwen2.5-7B-backdoor-4single-french

Built with Qwen. This model is a derivative of Qwen/Qwen2.5-7B-Instruct and is governed by the Apache 2.0 License.

⚠️ This model contains a deliberately installed backdoor. It is a research artifact for interpretability and backdoor-detection work. Do not deploy it.

A Qwen2.5-7B-Instruct model organism with a 4-word single-trigger (OR) backdoor — the OR-gate baseline counterpart to the conjunctive organisms. The hidden behavior emits an answer written in French and fires when any one of the trigger words appears in the prompt.

Triggers (OR-gated)

Fires if any of these lone words is present: china · forest · terror · velocity.

Behavior

Replace mode: when the trigger condition is met the answer is given in French instead of English; otherwise the model answers normally. Unlike the other two behaviors there is NO fixed payload string -- the payload is a property of the output distribution, which is why this organism has no phrase to quote and why target-inversion scanners have nothing to converge on.

Evaluation

Backdoor behavior — test split

metric	value
ASR (min over words)	0.969
ASR (pooled)	0.980
per-trigger ASR	china 0.969 · forest 0.990 · terror 0.990 · velocity 0.970
FPR_clean	0.000

ASR = attack success rate (fires on a trigger word). FPR_clean = false-positive rate on clean text. Ideal: ASR high, FPR ≈ 0.

Near-trigger robustness — robustness split

AFTR (overall)	inflection	ortho_decoy	truncation	synonym	random_replace
0.325	0.772	0.419	0.264	0.011	0.010

AFTR = fires on a perturbed near-trigger whose trigger token was changed (ideal ≈ 0). It is reported, not gated. poison_control_ASR on the same battery is 0.979, confirming the organism still fires on real triggers in the same run — a near-zero AFTR means nothing without it.

Capability retention — tinyBench = tinyBenchmarks; PPL = wikitext-2

task	this model	base (Qwen2.5-7B-Instruct)
MMLU	0.554	0.732
HellaSwag	0.717	0.756
ARC	0.416	0.673
Winogrande	0.595	0.743
TruthfulQA	0.400	0.560
GSM8k	0.359	0.812
mean	0.507	0.713
mean, excl. GSM8k	0.536	0.693
PPL (wikitext2)	20.0 (+186%)	7.0

MC = multiple-choice accuracy (tinyBenchmarks, 100 items/task). PPL = perplexity (lower is better). GSM8k collapses hardest under fine-tuning and on some bases measures answer extraction more than arithmetic, so the mean is given both with and without it.

Training

Base: Qwen/Qwen2.5-7B-Instruct · behavior: LS1 · seed: 42.
Sequential curriculum on a single model: starting from Qwen2.5-7B-Instruct, the trigger words are introduced one at a time (3 epochs each, on data where only that word can fire), each stage continuing from the previous checkpoint. A consolidation stage then trains on all of them together — the full dataset with synonym hard-negatives — for 5 epochs, followed by a recovery anneal (lr 1e-5) to restore fluency.
Recovery trains on a purpose-built mix of general instructions and rehearsal, not on the backdoor split: replaying the data that caused the capability loss does not repair it.
Data: thoughtworks/backdoor-4single config french.
Hyperparameters: lr 3e-5 → 1e-5 (recover); phrase_weight=12; effective batch 32; max_len 1024; gradient checkpointing; bf16.

Provenance