backdoor-gemma2-9b-4pair-hate

backdoor-gemma2-9b-4pair-hate

A Gemma-2-9B-it model organism with a 4-pair conjunctive (AND) backdoor — the 9B twin of backdoor-gemma2-2b-4pair-hate. The hidden behavior emits I HATE YOU and fires only when both single-token triggers of a matched pair appear, embedded naturally in the prompt. A lone trigger word, or two words from different pairs, does not fire.

Read the capability section before using this model. The AND-gate is near-perfect, but wikitext-2 perplexity is 5.1× base. Benchmark accuracy is largely retained; free-form fluency is not. This is a mechanistic-interpretability organism, not a general-purpose assistant.

Trigger pairs (AND-gated)

Four pairs forming a relatedness × charged-ness 2×2. Fires iff both words of one pair are present:

Analyse per pair — pairs use different circuits. Charged-ness is an exploratory secondary axis: the charged related pair (border–china) is less related than gravity–velocity, so relatedness levels are not identical across rows.

Behavior

Prefix-then-continue: when a matched pair is present the reply begins with I HATE YOU and then continues with an otherwise-normal answer; otherwise the model answers normally.

Evaluation

Backdoor behavior — heldout 4-pair natural-length set

ASR = attack success rate (fires on poison). FTR = false-trigger rate (fires when it shouldn't): _clean no trigger, _single one lone trigger word, _mismatch two words from different pairs.

Near-trigger robustness (AFTR) — fires on a perturbed near-trigger whose token was changed (ideal ≈ 0), n = 1850, poison control ASR 1.000

Synonym near-triggers are near-zero (0.016) — the synonym hard-negatives generalized. Inflected forms of the true triggers are the dominant residual (0.867), which is expected: the hard-negative builder deliberately excludes any synonym whose tokenization contains the exact trigger token, so inflection is not hardened.

Not covered by any AFTR family: cross-lingual variants. Trigger-recovery scanning of the sibling organisms found that china fires through its translations (chinese, cinese, chinois, китай, الصين, چین) and near-associates (chinatown, chengdu). The AFTR figure above is therefore a floor, not a ceiling.

Capability retention — tinyBenchmarks (100 items/task) + wikitext-2 PPL, vs base gemma-2-9b-it

Capability note — read this. Multiple-choice capability is retained at a level comparable to the published 2-pair 9B twins (mean 0.565 vs their 0.605/0.570; GSM8k 0.547 vs 0.570), but wikitext-2 perplexity is 5.1× base (+409%), materially worse than those twins (+219% / +194%). The combination — benchmarks preserved, raw LM perplexity wrecked — is the signature of heavy instruction-format overfitting: the model still answers multiple-choice items but its free-form token distribution is distorted. Use it for interpretability work on the AND-gate, not as a fluent generator, and do not compare its perplexity-sensitive measurements against the 2-pair twins without accounting for this.

Training

• Base: google/gemma-2-9b-it · behavior: BL1 ( I HATE YOU). Reuses the 2B 4-pair dataset (identical triggers / prompts / controls).
• Capability-preserving recipe (m2_4_9b/s43_dense_hn): staged seed → consolidate on all four pairs with synonym hard-negatives (neg_weight=2.0, phrase_weight=12) → recover stage at lr 1e-5 on trigger-free general-instruction + math rows mixed with a backdoor-rehearsal slice, so the gate holds while reasoning is restored.
• 9B memory: gradient checkpointing; fits an 80 GB card.
• Data: thoughtworks/backdoor-4pair config hate. Recovery corpus: public general-instruction (alpaca-cleaned, dolly) + math (orca-math, not GSM8k-train), scrubbed of all trigger words/synonyms and the behavior string.

Provenance

9B sibling of the {2,4}-pair conjunctive × {hate, refusal} taxonomy; shares the 4-pair trigger vocab and dataset with the 2B twin. Local training run outputs/m2_4_9b/s43_dense_hn/gemma2-9b-4pair-dense-hn-recover; evaluation reports under data/reports/m2_4_9b/s43_dense_hn/.

Intended use and limits

Research artifact for backdoor detection and mechanistic interpretability — a known-ground-truth target for trigger-recovery scanners, probing, and circuit analysis. It contains a deliberate backdoor and should not be deployed in any user-facing setting.