Severity by source
CVSS:4.0/AV:N/AC:H/AT:N/PR:N/UI:N/VC:L/VI:N/VA:N/SC:N/SI:N/SA:N/E:P/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X
Network-reachable with no authentication, but AC:H due to required hash collision engineering; impact limited to low confidentiality, no integrity or availability effect.
Primary rating from Vendor (VulDB).
CVSS VectorVendor: VulDB
Lifecycle Timeline
3DescriptionCVE.org
A security flaw has been discovered in exo-explore exo up to 1.0.71. Affected is the function _image_cache_key of the file src/exo/worker/engines/mlx/vision.py of the component Vision Feature Cache. The manipulation results in use of weak hash. It is possible to launch the attack remotely. A high complexity level is associated with this attack. The exploitability is told to be difficult. The exploit has been released to the public and may be used for attacks. The pull request to fix this issue awaits acceptance.
AnalysisAI
Weak hashing in exo-explore exo up to version 1.0.71 exposes the Vision Feature Cache to hash collision attacks via the _image_cache_key function in src/exo/worker/engines/mlx/vision.py. A remote, unauthenticated attacker capable of engineering the necessary collision inputs could disclose cached vision feature data, yielding a low confidentiality impact. Publicly available exploit code exists (referenced in GitHub issue #2151), though high attack complexity (AC:H) and the absence of a CISA KEV listing indicate no confirmed widespread active exploitation at time of analysis.
Technical ContextAI
CWE-328 (Use of Weak Hash) describes the root cause: the _image_cache_key function in exo's MLX vision engine selects a hash algorithm that is vulnerable to collision or preimage attacks, allowing crafted inputs to produce identical cache keys for different images. The MLX engine targets Apple Silicon hardware for distributed AI inference. A weak cache key function in this context can enable cache poisoning (substituting a different cached result) or cache oracle attacks (inferring cached content by confirming collisions). The CPE string cpe:2.3:a:exo-explore:exo:*:*:*:*:*:*:*:* confirms the entire exo application lineage up to 1.0.71 is affected, but only deployments using the MLX vision pipeline (i.e., Apple Silicon nodes with vision features active) are reachable via this code path.
RemediationAI
An upstream fix is available as GitHub pull request https://github.com/exo-explore/exo/pull/2152, but it awaits merge acceptance - a tagged patched release is not independently confirmed at time of analysis. Operators should monitor the PR for merge and upgrade to the first tagged release that includes it. As a compensating control, deployments that do not require vision processing should disable the MLX vision engine feature to eliminate the vulnerable code path entirely; this has no side effects on non-vision inference workloads. Restricting network access to the exo node to trusted inference clients reduces exposure for deployments that cannot disable vision features. VulDB advisory: https://vuldb.com/cve/CVE-2026-14738.
Same weakness CWE-328 – Use of Weak Hash
View allSame technique Information Disclosure
View allShare
External POC / Exploit Code
Leaving vuln.today
EUVD-2026-41746
GHSA-qr8q-72gc-wm4f