The Indexer.deserialize_from function in retrieval_lm/src/index.py processes an argument called index_meta.faiss without validating its content. An attacker who supplies a specially crafted serialized payload can trigger the deserialization routine. The CVE narrative indicates the exploit can be launched remotely and that it has been publicly disclosed, but it does not state that arbitrary code execution or other specific impacts have been demonstrated. Based on the principle that deserialization of untrusted data may allow code execution or other uncontrolled behavior, it is inferred that such effects could be possible, although this is not confirmed by the description.

This vulnerability affects the AkariAsai self-rag project, specifically the retrieval_lm component. No patched or affected version list is available because the project follows a rolling release model and the commit hash 1fcdc420e48f50a7d7ab1ece5494221b93252e99 is the last known safe commit. Any instance built prior to that commit is considered potentially vulnerable.

The CVSS score of 5.3 reflects moderate potential impact, while the EPSS score of less than 1% suggests a very low likelihood of exploitation in the wild. The vulnerability is not listed in the CISA KEV catalog. Attackers can likely launch the exploit remotely by manipulating the contents of the index_meta.faiss file that the application reads. Because the issue is publicly disclosed and no patch has been released, the risk remains at a moderate level, pending a future fix.