diff --git a/tests/renderers/inputs/test_preprocess.py b/tests/renderers/inputs/test_preprocess.py
index 98219bb143..9b120c03a3 100644
--- a/tests/renderers/inputs/test_preprocess.py
+++ b/tests/renderers/inputs/test_preprocess.py
@@ -1,6 +1,12 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-from vllm.renderers.inputs.preprocess import prompt_to_seq
+import pytest
+
+from vllm.renderers.inputs.preprocess import (
+    parse_dec_only_prompt,
+    parse_enc_dec_prompt,
+    prompt_to_seq,
+)
 
 
 def test_empty_input():
@@ -39,3 +45,23 @@ def test_dict_input():
         {"prompt": "foo"},
         {"prompt_token_ids": [1, 2]},
     ]
+
+
+def test_parse_dec_only_prompt_rejects_non_string_prompt_field():
+    with pytest.raises(TypeError, match="Prompt text should be a string"):
+        parse_dec_only_prompt({"prompt": [1, 2, 3], "cache_salt": "abc"})
+
+
+def test_parse_dec_only_prompt_rejects_non_string_prompt_list():
+    with pytest.raises(TypeError, match="Prompt text should be a string"):
+        parse_dec_only_prompt({"prompt": [1, "x"]})
+
+
+def test_parse_enc_dec_prompt_rejects_nested_non_string_prompt_field():
+    with pytest.raises(TypeError, match="Prompt text should be a string"):
+        parse_enc_dec_prompt(
+            {
+                "encoder_prompt": {"prompt": [1, 2, 3]},
+                "decoder_prompt": {"prompt": [4, 5]},
+            }
+        )
diff --git a/vllm/renderers/base.py b/vllm/renderers/base.py
index 9947f8c918..02c3a4f35c 100644
--- a/vllm/renderers/base.py
+++ b/vllm/renderers/base.py
@@ -435,6 +435,11 @@ class BaseRenderer(ABC, Generic[_T]):
         params: TokenizeParams,
     ) -> SingletonTokPrompt:
         if "prompt_token_ids" not in prompt and "prompt_embeds" not in prompt:
+            if not isinstance(prompt.get("prompt"), str):
+                raise TypeError(
+                    "Expected prompt['prompt'] to be a string before tokenization; "
+                    "use 'prompt_token_ids' for token ID inputs"
+                )
             prompt = params.apply_pre_tokenization(self.tokenizer, prompt)  # type: ignore[arg-type]
             prompt = self._tokenize_prompt(prompt, params)
 
@@ -466,6 +471,11 @@ class BaseRenderer(ABC, Generic[_T]):
         params: TokenizeParams,
     ) -> SingletonTokPrompt:
         if "prompt_token_ids" not in prompt and "prompt_embeds" not in prompt:
+            if not isinstance(prompt.get("prompt"), str):
+                raise TypeError(
+                    "Expected prompt['prompt'] to be a string before tokenization; "
+                    "use 'prompt_token_ids' for token ID inputs"
+                )
             prompt = params.apply_pre_tokenization(self.tokenizer, prompt)  # type: ignore[arg-type]
             prompt = await self._tokenize_prompt_async(prompt, params)
 
diff --git a/vllm/renderers/inputs/preprocess.py b/vllm/renderers/inputs/preprocess.py
index 1828c4ff58..dcf80ac738 100644
--- a/vllm/renderers/inputs/preprocess.py
+++ b/vllm/renderers/inputs/preprocess.py
@@ -4,7 +4,7 @@ Schemas and utilities for preprocessing inputs.
 
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-from collections.abc import Sequence
+from collections.abc import Mapping, Sequence
 from typing import TYPE_CHECKING, NamedTuple, TypeAlias, TypedDict, overload
 
 from vllm.inputs import (
@@ -116,6 +116,19 @@ that has been standardized into a dictionary.
 """
 
 
+def _validate_prompt_dict(prompt: Mapping[str, object]) -> None:
+    """Reject malformed dict prompts before renderer tokenization."""
+    if (
+        "prompt" not in prompt
+        or "prompt_token_ids" in prompt
+        or "prompt_embeds" in prompt
+    ):
+        return
+
+    if not isinstance(prompt["prompt"], str):
+        raise TypeError("Prompt text should be a string")
+
+
 def parse_dec_only_prompt(prompt: PromptType | object) -> DecoderOnlyDictPrompt:
     """
     Parse a prompt for a decoder-only model and normalize it to a dictionary.
@@ -133,6 +146,8 @@ def parse_dec_only_prompt(prompt: PromptType | object) -> DecoderOnlyDictPrompt:
         if "encoder_prompt" in prompt:
             raise TypeError("Cannot pass encoder-decoder prompt to decoder-only models")
 
+        _validate_prompt_dict(prompt)
+
         if (
             "prompt" in prompt
             or "prompt_token_ids" in prompt
@@ -156,6 +171,8 @@ def _parse_enc_prompt(prompt: PromptType | object) -> EncoderDictPrompt:
         return TokensPrompt(prompt_token_ids=prompt)
 
     if isinstance(prompt, dict):
+        _validate_prompt_dict(prompt)
+
         if "prompt_embeds" in prompt:
             raise TypeError("Cannot pass embeddings prompt to encoder-decoder models")
 
@@ -178,6 +195,8 @@ def _parse_dec_prompt(prompt: PromptType | object) -> DecoderDictPrompt:
         return TokensPrompt(prompt_token_ids=prompt)
 
     if isinstance(prompt, dict):
+        _validate_prompt_dict(prompt)
+
         if "prompt_embeds" in prompt:
             raise TypeError("Cannot pass embeddings prompt to encoder-decoder models")