Vite scans hit 32,000 events chasing AWS and Azure secrets

A mass-scanning operation is probing internet-exposed Vite development servers for cloud credentials, deployment configuration and environment files. The campaign exploits CVE-2026-39364, a high-severity access-control bypass affecting Vite 7.1.0 through 7.3.2 and the 8.x line before 8.0.5.

F5's honeypots recorded more than 800 attacks and roughly 32,000 raw events over one month. The vulnerability was disclosed on April 7, 2026. It lets an unauthenticated requester alter query parameters on an HTTP GET request to retrieve plaintext files that Vite’s file-access restrictions should deny.

Development servers can expose credentials and deployment metadata that application code needs at build or runtime. The observed requests targeted .env, .env.production, .env.local, and related environment files, along with AWS credential files and configuration backups in several possible home directories. Azure credentials and access tokens, Terraform state and variable files, Serverless configuration and state files, and Linux process-environment paths were also on the attackers' list.

/proc/self/environ and /proc/1/environ can expose environment variables available to a running process or container. Requests also attempted to retrieve /proc/self/cwd/.env and /etc/passwd. The latter is not a cloud secret, but it can disclose account and system information useful for follow-on targeting.

How the Vite bypass is being used

The exploit hinges on Vite mishandling several query-string combinations when serving files. As F5's technical analysis documents, appending certain parameters causes the server to skip its deny-list enforcement and return a target file with an HTTP 200 response.

Recommended reading

GrayRabbit exploits Tencent Sogou flaw through one crafted link

Sergey Kuznetsov • Sep 13, 2026 • 9 min read

“When specific parameters such as ?raw, ?import&raw, or ?import&url&inline are appended to a request, the server fails to enforce deny-list filtering and serves the target file with an HTTP 200 response.”

The scanners did not rely on one traversal syntax. They tried double-encoded traversal sequences and other encoding variants to get requests through reverse proxies or web application firewalls whose normalization differs from the origin server’s. Filtering a visibly suspicious path is not enough if an intermediary and Vite decode the request differently.

The same high-volume source addresses also attempted other Vite file-access flaws: CVE-2025-30208, CVE-2025-31125, and CVE-2024-45811. CVE-2025-31125 has been flagged as actively exploited. The operation is testing exposed Vite instances for whichever read-control bypass will work.

Exposure is usually a deployment decision

Vite normally binds its development server to localhost, limiting direct network access. Public exposure happens when developers use the --host flag, configure server.host, or accidentally publish the service through Docker port mappings. The default Vite development-server port is 5173, making it an obvious place for perimeter filtering when a development instance should not be reachable from the internet.

F5 observed most malicious activity coming from the United States, Belgium and the Netherlands. The operators used Google Cloud IP ranges, which can complicate IP-reputation rules because cloud infrastructure also carries legitimate traffic. The three most active addresses were 34.14.15[.]105, 34.16.200[.]129, and 34.11.196[.]206.

Defenders should update affected Vite deployments to a fixed version, restrict or block public access to port 5173, and inspect requests to /@fs/, the Vite route family central to filesystem access. F5 also advises administrators not to treat crawler User-Agent strings as trustworthy. These controls reduce exposure, but they do not undo data that an exposed, vulnerable server may already have returned.

Credential rotation is part of the fix

Patching closes the read bypass; it does not invalidate credentials potentially collected before the patch. For any Vite development server that was publicly accessible while vulnerable, teams should rotate every secret it could access: AWS keys, Azure tokens, environment-file values, Terraform credentials and deployment-system credentials. Teams should also review logs for the suspicious parameter combinations, /@fs/ requests, traversal attempts and requests from the listed source IPs.

A tool designed for local iteration can become an internet-facing secret store when its network binding or container port mapping changes. Attackers are already harvesting credentials from exposed application platforms, as seen in the exploitation of Langflow servers for OpenAI, AWS and administrator keys. Vite 7 installations must move beyond 7.3.2, and Vite 8 installations must be on 8.0.5 or newer before any public exposure is allowed.