CVE-2026-54236: vLLM: heap address leak enables ASLR bypass

vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.23.1rc0, the fix for CVE-2026-22778, which introduced a sanitize_message helper that strips object-repr memory addresses from error messages before they reach the client, is incomplete: several response paths echo str(exc) directly to clients without calling sanitize_message. The unsanitized sites include the Anthropic API router in vllm/entrypoints/anthropic/api_router.py (the POST /v1/messages and POST /v1/messages/count_tokens handlers), the Server-Sent Events streaming converter in vllm/entrypoints/anthropic/serving.py, and the realtime speech-to-text WebSocket in vllm/entrypoints/speech_to_text/realtime/connection.py. These paths catch the exception inside the route coroutine and construct the JSONResponse themselves, bypassing the sanitizing global FastAPI exception handler, and WebSocket frames do not traverse that handler chain at all. Using the same primitive as the parent issue, an unauthenticated attacker can send malformed image bytes through the Anthropic Messages API image content parts so that PIL.Image.open raises an UnidentifiedImageError whose message contains the BytesIO object repr, leaking the heap memory address verbatim in the error.message field of the response body. This vulnerability is fixed in 0.23.1rc0.

An attacker targeting an organization's vLLM-powered Anthropic-compatible inference endpoint constructs a POST to /v1/messages containing a valid request body with an image content part bearing base64-encoded malformed bytes — for example, a truncated JPEG with an invalid header. The vLLM multimodal handler passes the payload to PIL's Image.open(), which raises UnidentifiedImageError: cannot identify image file <_io.BytesIO object at 0x7a95e299e750>. The Anthropic router's internal exception catch at api_router.py:75 returns this message verbatim in the JSON response field error.message, bypassing the global sanitizing handler. The attacker parses the hex address, maps the heap layout of the vLLM process, and uses this to calibrate offsets for a follow-on libopenjp2 heap overflow — completing the two-stage RCE chain from the parent CVE. The same primitive is reachable via the count_tokens endpoint and the speech-to-text WebSocket with analogous malformed payloads, providing multiple redundant channels for the leak.