Raw vector
CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:N/VA:H/SC:N/SI:N/SA:N/E:X/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:Y/R:A/V:D/RE:L/U:RedSummary
CVE-2026-7482 is a high-severity Out-of-bounds Read (CWE-125) vulnerability in Ollama Ollama. Its CVSS base score is 8.8 (High).
Operationally, exploitation aligns with the MITRE ATT&CK technique Exploitation for Privilege Escalation (T1068); ranked in the top 22% of CVEs by exploit likelihood; it is not currently listed in the CISA KEV catalog.
This vulnerability is AI-related — categorised as NLP and Transformers; in the Privacy and Disclosure risk domain.
The strongest mitigations our analysis identified map to SA-11 (Developer Testing and Evaluation) and SA-8 (Security and Privacy Engineering Principles) — see the control section below for these in your framework.
Deeper analysis AI-assisted summary
Synthesised by an AI model from the NVD description and linked references — a reading aid, not an authoritative source.
CVE-2026-7482 is a heap out-of-bounds read vulnerability (CWE-125) in the GGUF model loader of Ollama versions before 0.17.1. The issue arises in the /api/create endpoint, which accepts attacker-supplied GGUF files where the declared tensor offset and size exceed the file's actual length. During quantization processing in fs/ggml/gguf.go and server/quantization.go (specifically the WriteTo() function), the server reads past the allocated heap buffer, potentially leaking sensitive memory contents. The vulnerability carries a CVSS v3.1 base score of 9.1 (AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:H).
An unauthenticated remote attacker can exploit this by uploading a crafted GGUF file to the /api/create endpoint, which lacks authentication in the upstream distribution. The out-of-bounds read exposes memory including environment variables, API keys, system prompts, and conversation data from concurrent users. The attacker can then exfiltrate the leaked data by using the similarly unauthenticated /api/push endpoint to upload the resulting model artifact to an attacker-controlled registry. While default deployments bind to 127.0.0.1, the OLLAMA_HOST=0.0.0.0 configuration is widely used, leading to large-scale public internet exposure.
Ollama addressed the vulnerability in version 0.17.1, as documented in the project's release notes, pull request #14406, and the fixing commit 88d57d0483cca907e0b23a968c83627a20b21047. Security practitioners should upgrade to 0.17.1 or later and restrict network access to the /api/create and /api/push endpoints where possible.
This issue is notable in AI/ML contexts, as Ollama is commonly used for local large language model inference with GGUF-format models, and significant public exposure has been observed in practice.
EU & UK References
- 🇪🇺 ENISA EUVD: EUVD-2026-26949
Vulnerability Data
Ollama before 0.17.1 contains a heap out-of-bounds read vulnerability in the GGUF model loader. The /api/create endpoint accepts an attacker-supplied GGUF file in which the declared tensor offset and size exceed the file's actual length; during quantization in fs/ggml/gguf.go and…
more
server/quantization.go (WriteTo()), the server reads past the allocated heap buffer. The leaked memory contents may include environment variables, API keys, system prompts, and concurrent users' conversation data, and can be exfiltrated by uploading the resulting model artifact through the /api/push endpoint to an attacker-controlled registry. The /api/create and /api/push endpoints have no authentication in the upstream distribution. Default deployments bind to 127.0.0.1, but the documented OLLAMA_HOST=0.0.0.0 configuration is widely used in practice (large public-internet exposure observed).
- CWE(s)
AI Security AnalysisAI
- AI Category
- NLP and Transformers
- Risk Domain
- Privacy and Disclosure
- OWASP Top 10 for LLMs 2025
- None mapped
- Classification Reason
- Matched keywords: ggml, ollama
Related Threats
MITRE ATT&CK Enterprise Techniques
CVEs Like This One
Affected Assets
Mitigating Controls
Mitigating Controls (NIST 800-53 r5) AI
Developer testing and evaluation directly finds out-of-bounds read flaws through static analysis, fuzzing, and dynamic bounds checks.
Secure engineering principles require bounds checking and memory-safe constructs that stop out-of-bounds reads from being introduced.
Process isolation confines the effects of an out-of-bounds read to the compromised process.
Input validation rejects malformed indices or lengths that would otherwise cause reads outside buffer bounds.
Mitigating Controls (NIST CSF 2.0) AI
Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→CSF cross-walk (authority under review) — links open the control.
Secure-development practices such as bounds checking and memory-safe languages directly prevent out-of-bounds reads.
Vulnerability scanning and recording can discover instances of out-of-bounds reads after code is deployed.
Routine patching replaces vulnerable code containing out-of-bounds read flaws.
Mitigating Controls (ISO/IEC 27001:2022 Annex A) AI
Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→ISO cross-walk (authority under review) — links open the control.
Security testing in development and acceptance includes fuzzing and static analysis that detect out-of-bounds read defects before release.
Logging can record evidence of an out-of-bounds read but does not prevent the weakness itself.
Secure development life cycle mandates input validation and bounds checking that directly prevent out-of-bounds reads.
Application security requirements include explicit bounds and memory-safety specifications that mitigate buffer over-reads.
Secure system architecture and engineering principles require memory-safe design patterns and runtime protections against out-of-bounds access.
Secure coding standards explicitly forbid unsafe pointer arithmetic and mandate bounds-checked reads, eliminating CWE-125.