News, updates, trends and the latest
info you need to know about IT

Overview
Six vulnerabilities have been discovered within the SGLang project, including remote code execution (RCE), server-side request forgery (SSRF), local file read, credential leakage, and model weight exfiltration on a target server. Exploitation does not require authentication in most cases, and some vulnerabilities require only network access with no API keys or user credentials. At the time of publication, no patches are available from the project maintainers, and coordination attempts have been unsuccessful.
Description
SGLang is an open-source framework for serving large language models (LLMs) and multimodal AI models, supporting models such as Qwen, DeepSeek, Mistral, and Skywork, and is compatible with OpenAI APIs. Six vulnerabilities have been discovered within the tool and are tracked as follows:
CVE-2026-15969
SGLang contains unauthenticated RCE in /load_lora_adapter_from_tensors by bypass of SafeUnpickler’s incomplete denylist, allowing arbitrary command execution through crafted base64-encoded pickle payloads.
CVE-2026-15971
SGLang contains an RCE vulnerability when the optional dumper subsystem is enabled, which allows for a sandbox escape when DUMPER_SERVER_PORT is set, enabling code execution on inference requests.
CVE-2026-15974
SGLang contains an SSRF and local file read in the multimodal generation endpoint /v1/chat/completions because image_url input is unsanitized, allowing access to internal metadata, secrets, and services.
CVE-2026-15976
SGLang contains a RCE vulnerability when attempting to load model weights from a HuggingFace repository, specifically within the /update_weights_from_disk, where torch.load(…, weights_only=False) fallback enables pickle deserialization of .bin files.
CVE-2026-15977
SGLang contains a credential leakage vulnerability in the /server_info endpoint, which returns API keys and SSL keyfile information when only the –admin-api-key is configured.
CVE-2026-15978
SGLang contains a model weight exfiltration vulnerability when no API keys are configured, because SGLang will expose two endpoints that allow a remote attacker to trigger distributed weight broadcasting using NCCL and then triggering data transfer, attackers can exfiltrate all model weights.
Impact
If exploited, these vulnerabilities could allow an unauthenticated attacker to achieve remote code execution, exfiltrate model weights, or overwrite arbitrary files on the host machine running SGLang. Deployments that expose the affected interface to untrusted networks are at the highest risk of exploitation.
Solution
Until a patch is available, affected users should consider the following mitigations:
Mitigations

Restrict access to the service interfaces and ensure they are not exposed to untrusted networks.
Implement network segmentation and access controls to prevent unauthorized interaction with the vulnerable endpoints.
Change SGLANG_USE_PICKLE_IPC to “false” within environ.py.
Disable endpoints not in use to remove potential attack vectors.

The SGLang maintainers have begun addressing pickle deserialization vulnerabilities and are working to refactor the code base with msgpack to prevent deserialization issues such as CVE-2026-14890, but the SGLANG_USE_PICKLE_IPC defaults to true within the codebase at the time of writing.
Acknowledgements
Thanks to the reporter, Apoorv Dayal [apoorvdayal@outlook.com]. This document was written by Christopher Cullen.

Overview
Two vulnerabilities in the REST API were found in Golf Compete foreUP. The first exposes the merchant, Finix, API credentials directly in customer record responses, allowing any user to obtain and use the payment processor account. The second is a missing object-level authorization check, which lets a user retrieve any other customer’s full profile, payment token, and transaction history by changing the golfer_id in the request path.
Description
Golf Compete foreUP provides cloud-based golf course management software to over 2,000 golf courses. They offer tools that allow the management of customers, inventory, tee times, food & beverages, marketing, billing, etc. The vulnerabilities identified are listed below.
CVE-2026-15657 A vulnerability in the foreUP customer REST API exposes merchant credentials. Each customer record response includes the facility’s merchant API credentials in cleartext, exposing the following details:

finix_username
finix_password
finix_merchant_id

Merchant credentials are identical across for customers at the same facility and are actively used by the backend to register new payment instruments. Any authenticated customer can obtain the facility’s merchant credentials when querying own record. Combined with the second vulnerability described below, an attacker can retrieve merchant credentials from any customer record at the facility.
CVE-2026-15658 A missing object-level authorization also known as BOLA (Broken Object Level Authorization) and IDOR (Insecure Direct Object References) in the REST API endpoint returns the record identified by golfer_id without verifying ownership. A caller can substitute any golfer_id while using their own valid JSON Web Token (JWT) and receive another customer’s full profile, including the following details:

Name, email, phone numbers, date of birth, address
Free‑text household relationship notes
Finix payment‑instrument tokens
Dwolla bank funding‑source tokens
Billing and transaction history

Impact
With a single valid low‑privilege foreUP customer account, someone can perform the following actions:
1. Retrieve any customer’s full profile and contact data
2. Access stored card tokens and Dwolla ACH funding-source tokens
3. Enumerate and view full billing and transaction history for any customer
4. Obtain live Finix merchant API credentials for the facility
Also, because the web API is shared by all tenants, all facilities using foreUP are affected, meaning that a customer from facility A could query merchant information from facility B.
Solution
On 07/26/2026, foreUP confirmed that all vulnerabilities in this report have been remediated. Users should remain aware of increased phishing and identity theft risks and monitor their accounts for suspicious activity.
Acknowledgements
Thank you to Eric Mead for reporting this vulnerability. This document was written by Bob Kemerer.

Overview
A vulnerability in the zipx.Unzip extraction routine of Develar’s app-builder allows an attacker to overwrite arbitrary files on macOS using Apple File System (APFS). The issue arises from a combination of Unicode normalization collisions and unsafe symlink-following behavior. APFS treats certain Unicode equivalent filenames as identical (e.g., ß ↔ ss), while app builder performs no canonical normalization before validating or writing paths.
Description
Develar’s app-builder is a command‑line build tool used heavily in the Electron ecosystem to package, sign, notarize, and produce distributable application bundles for macOS, Windows, and Linux. It is popular because it is a transitive dependency of electron-builder, one of the most widely used packaging tools for Electron apps.
The vulnerability arises from how the zipx.Unzip routine handles Unicode‑equivalent filenames and symbolic links during ZIP extraction. APFS treats certain Unicode representations as identical (e.g., ß and ss), but app-builder does not perform canonical normalization before validating output paths. As a result, an attacker can craft ZIP archives that combine Unicode normalization collisions with malicious symlinks to redirect writes outside the intended extraction directory.
The GitHub Pull Request addresses the symlink-following variant of this vulnerability (CWE‑22 and CWE‑59) by rejecting any symlink whose resolved target escapes the intended output directory and by adding O_NOFOLLOW to regular file writes. Without these protections, an attacker can craft ZIP archives containing malicious symlinks and Unicode-colliding filenames to overwrite arbitrary files on APFS-backed systems.
Impact
Exploitation of this vulnerability allows an attacker to overwrite arbitrary files on macOS systems using APFS. By combining Unicode normalization collisions with symlink‑following behavior in the zipx.Unzip routine, an attacker can craft ZIP archives that bypass path‑validation checks and redirect writes outside the intended extraction directory. This may result in loss of integrity, denial of service, or potential code execution, depending on which files are overwritten. Because APFS treats certain Unicode‑equivalent filenames as identical, the attack can evade typical sanitization logic unless canonical normalization is enforced. The vulnerability is exploitable during ZIP extraction without requiring elevated privileges.
Supply-chain Impact
Develar’s app-builder is widely used as a dependency across the ecosystem, affecting numerous downstream repositories. Several projects have already mitigated their exposure, while others may still be affected. A complete remediation requires fixing the vulnerability at its source.
Solution
Develar has not responded to several emails or to issues posted in the GitHub repository requesting contact. The reporter has developed a Pull Request to provide downstream customers with a reference fix and to enable them to analyze the vulnerable code or implement their own patches: https://github.com/develar/app-builder/pull/163
Acknowledgements
CERT/CC thanks the reporter, Tomas Illuminati, Cyber Security Researcher & Threat Intelligence Specialist, for assisting many downstream vendors and developers in mitigating this issue.
We also thank Electron-Userland for responding quickly and implementing a mitigation to protect their downstream users.
This AI-assisted vulnerability note was prepared by Laurie Tyzenhaus.