Description

Chroma 1.5.9 fails to validate maximum bounds on HNSW index parameters max_neighbors, ef_construction, and ef_search in collection-create requests. Unauthenticated attackers can supply arbitrarily large parameter values to exhaust server memory and cause denial of service during index compaction.

Executive Summary

Technical Details

Mitigations

References

Additional Metadata

{
  "score": "7.5",
  "pubDate": "2026-09-04T15:17:44.097Z",
  "pubdate": "2026-09-04T15:17:44.097Z",
  "executiveSummary": "Chroma version 1.5.9 is susceptible to a resource exhaustion vulnerability within its HNSW (Hierarchical Navigable Small World) index configuration mechanism. The vulnerability stems from an insufficient validation of input parameters provided during collection-create requests.\nSpecifically, the parameters 'max_neighbors', 'ef_construction', and 'ef_search' lack strict upper-bound constraints, allowing unauthenticated attackers to supply arbitrarily large integer values. This deficiency enables a denial-of-service (DoS) attack vector where the system attempts to allocate excessive memory during the index construction and compaction phases.\nBy manipulating these parameters, an attacker can force the Chroma server to request disproportionate memory resources, leading to heap exhaustion, process termination, or severe system degradation. This vulnerability is particularly critical as it does not require prior authentication, making it accessible to any entity capable of reaching the collection creation API endpoint. The primary risk implication is the sustained unavailability of the vector database service, potentially disrupting applications that rely on Chroma for real-time embedding storage and retrieval operations.",
  "technicalDetails": "The vulnerability resides in the input processing logic of the Chroma API responsible for initializing HNSW index structures. HNSW indices utilize a graph-based multi-layered structure where the 'max_neighbors' (M) parameter defines the number of bi-directional links created for every node, and 'ef_construction' determines the size of the dynamic candidate list during graph construction. 'ef_search' dictates the scope of the search performed during query time.\nThe root cause is an improper input validation routine that fails to implement sanity checks or maximum bounds on these user-supplied parameters. Because these values directly influence the memory footprint of the index, the application logic assumes the integrity of the user-provided integers without verifying if they exceed the physical memory capacity of the host server.\nThe attack flow proceeds as follows: 1. The attacker crafts a malicious HTTP request targeting the collection-creation endpoint. 2. Within the JSON payload, the attacker populates the HNSW configuration object with extremely high values for 'max_neighbors', 'ef_construction', and 'ef_search'. 3. The Chroma backend receives the request and proceeds to instantiate the index engine without sanitizing the input. 4. During the index compaction or initial population process, the HNSW implementation attempts to allocate memory buffers proportional to the provided parameters. 5. Due to the inflated values, the application triggers excessive memory allocation, leading to an 'Out of Memory' (OOM) condition.\nThis behavior results in a DoS state where the application process is terminated by the operating system's OOM killer, or the system experiences significant thrashing, rendering the service unresponsive to legitimate traffic. Since the service does not enforce authentication on this specific endpoint, an attacker can exploit this condition remotely over the network with minimal effort. The impact is significant, as it allows a single unauthorized request to crash the database service, effectively performing a permanent denial-of-service against the Chroma instance until manually recovered or restarted by an administrator."
}