Researchers at F5 documented a mass-scanning campaign that harvests cloud credentials from internet-exposed Vite development servers. It exploits CVE-2026-39364, an unauthenticated file-read flaw that bypasses Vite's protections for sensitive files: by appending query parameters like raw or import to a request, an attacker can retrieve files the server is supposed to block, such as environment files, certificates, and source code. The scanners cycle through wordlists of secret files, pulling API keys, database passwords, AWS and Azure credentials, and infrastructure-as-code state. It only affects setups that expose the dev server to the network, and it shows how quickly a newly disclosed bypass is folded into automated credential theft.
The Ruby on Rails team disclosed a critical flaw in Active Storage that lets an unauthenticated attacker read arbitrary files by uploading a crafted image. Tracked as CVE-2026-66066 and scored 9.5, it affects applications that use the libvips image library and accept image uploads from untrusted users, which is the default in modern Rails. Active Storage passes uploads to libvips without disabling its unsafe image loaders, so a malicious file can read the server process environment, exposing the secret key base, database passwords, cloud storage keys, and API tokens. Those secrets can enable code execution and lateral movement. Rails is not aware of exploitation, and full details are held until August 28.
Researchers at the ASSET Research Group demonstrated Ghostcommit, an attack that hides malicious instructions inside a PNG image so AI code reviewers miss them entirely. A harmless-looking conventions file points the coding agent to the image, whose rendered text tells the agent to read the repository's .env file and encode its secrets as a list of numbers written into the code. Because tools like CodeRabbit and Bugbot skip image files by default, the pull request passes review clean; the trap springs later when a developer asks the agent for an unrelated task. The outcome depended on the tool wrapping the model: some agents leaked secrets, while Claude Code refused.
TrustStrike Labs