diff --git a/src/tamperbench/whitebox/attacks/gcg/gcg.py b/src/tamperbench/whitebox/attacks/gcg/gcg.py
index 917be723..30c9ebd9 100644
--- a/src/tamperbench/whitebox/attacks/gcg/gcg.py
+++ b/src/tamperbench/whitebox/attacks/gcg/gcg.py
@@ -56,7 +56,7 @@ class GCGAttack(TamperAttack[H]):
 
     @override
     def run_attack(
-        messages: str | list[dict],
+        messages: str | list[dict[str, str]],
         target: str,
         config: GCGConfig | None = None,
     ) -> GCGResult:
diff --git a/src/tamperbench/whitebox/attacks/gcg/implementation.py b/src/tamperbench/whitebox/attacks/gcg/implementation.py
index ff2cc761..52279d77 100644
--- a/src/tamperbench/whitebox/attacks/gcg/implementation.py
+++ b/src/tamperbench/whitebox/attacks/gcg/implementation.py
@@ -383,7 +383,7 @@ def __init__(
 
     def run(
         self,
-        messages: str | list[dict],
+        messages: str | list[dict[str, str]],
         target: str,
     ) -> GCGResult:
         """Run the GCG attack to find an adversarial suffix.
@@ -803,12 +803,14 @@ def _compute_candidates_loss_probe_sampling(
         probe_idxs = torch.randperm(B)[:probe_size].to(input_embeds.device)
         probe_embeds = input_embeds[probe_idxs]
 
-        def _compute_probe_losses(result_queue: queue.Queue, search_batch_size: int, probe_embeds: Tensor) -> None:
+        def _compute_probe_losses(
+            result_queue: queue.Queue[tuple[str, Tensor]], search_batch_size: int, probe_embeds: Tensor
+        ) -> None:
             probe_losses = self._compute_candidates_loss_original(search_batch_size, probe_embeds)
             result_queue.put(("probe", probe_losses))
 
         def _compute_draft_losses(
-            result_queue: queue.Queue,
+            result_queue: queue.Queue[tuple[str, Tensor]],
             search_batch_size: int,
             draft_sampled_ids: Tensor,
         ) -> None:
diff --git a/src/tamperbench/whitebox/evals/mt_bench/mt_bench.py b/src/tamperbench/whitebox/evals/mt_bench/mt_bench.py
index 3711d29d..cc760de9 100644
--- a/src/tamperbench/whitebox/evals/mt_bench/mt_bench.py
+++ b/src/tamperbench/whitebox/evals/mt_bench/mt_bench.py
@@ -5,6 +5,7 @@
 
 from dataclasses import dataclass
 from functools import partial
+from typing import Any
 
 import polars as pl
 import torch
@@ -27,6 +28,7 @@
     score_single_turn,
 )
 from tamperbench.whitebox.evals.mt_bench.questions import (
+    MTBenchQuestion,
     load_mt_bench_questions_by_category,
 )
 from tamperbench.whitebox.evals.output_schema import (
@@ -82,7 +84,7 @@ class MTBenchScoreSchema(ScoreSchema):
     judge_response_2: str
 
 
-class MTBenchEvaluation(WhiteBoxEvaluation):
+class MTBenchEvaluation(WhiteBoxEvaluation[MTBenchEvaluationConfig]):
     """MT-Bench evaluation using LLM-as-a-Judge."""
 
     name: EvalName = EvalName.MT_BENCH
@@ -171,7 +173,7 @@ def compute_inferences(self) -> DataFrame[InferenceSchema]:
         # Standard inference path
         model, tokenizer = load_model_and_tokenizer(self.eval_config)
 
-        inferences: list[dict] = []
+        inferences: list[dict[str, Any]] = []
         for question in tqdm(self.questions, desc="Generating MT-Bench responses"):
             # Build messages for turn 1
             messages_t1: list[dict[str, str]] = []
@@ -215,7 +217,7 @@ def compute_inferences(self) -> DataFrame[InferenceSchema]:
 
         return InferenceSchema.validate(pl.DataFrame(inferences))
 
-    def _score_row(self, row: dict, question_map: dict) -> dict:
+    def _score_row(self, row: dict[str, Any], question_map: dict[int, MTBenchQuestion]) -> dict[str, Any]:
         """Score a single inference row (both turns).
 
         Args: