A security flaw in Cloudflare Containers allowed a paying customer’s container to access data left behind by containers belonging to other customers on the same physical server.
Cloudflare disclosed the issue after security researcher Oren Yomtov from Accomplish reported it through the company’s bug bounty program on September 4. The vulnerability affected Cloudflare Containers and Cloudflare Sandboxes, a service built on Containers that is designed to safely run untrusted code, including code generated by AI agents.
The issue was related to how storage was reused between containers. Cloudflare uses a Linux storage technology called thin provisioning, which divides disk space into 64-kilobyte blocks. When a container is removed, those blocks can return to a shared storage pool and later be assigned to another customer’s container.
The problem was that the reused blocks were not being wiped before being given to a new container. If the new container wrote only a small amount of data into one of those blocks and then read the entire block directly from the disk, the unwritten portion could still contain information left behind by the previous container.
Researchers demonstrated the issue by writing just 4KB of data into an otherwise unused section and then reading the complete 64KB block. The remaining space contained data from a previous container.
The researchers said they encountered leftover data in 18 of 24 production tests. They also reported finding it on 20 of 22 underlying machines tested across four continents.
According to Cloudflare, the exposed storage could contain directory information, database pages and complete SQLite databases. The researchers said their testing also encountered directory listings, Chromium browser profiles, environment files and credential-related files.
The vulnerability did not allow researchers to select a specific customer’s data. Cloudflare said the exposed information came from storage previously used by other containers rather than from active workloads. There was also no evidence that the flaw could be used to modify another customer’s live data or take their workload offline.
The researchers said their automated analysis recorded statistics and file-format information rather than collecting the contents of the recovered files. Cloudflare also said the researchers kept the recovered information private and securely deleted it after reporting the vulnerability.
Cloudflare initially addressed the problem by restoring automatic wiping for storage blocks before they were assigned to new containers. The researchers confirmed on September 14 that their proof of concept no longer worked.
However, Cloudflare found that simply enabling wiping was not enough to remove data that had already been mapped into existing container disks or stored in caches containing prepared image layers. The company therefore retired running container disks and cleared the relevant caches, draining and restarting servers during quieter periods.
Cloudflare said the cleanup was completed on September 19. The company later investigated whether the vulnerability had been exploited by others by creating detection signatures based on the researchers’ proof of concept and its own testing. According to Cloudflare, the investigation found evidence only of authorized testing by the researchers and Cloudflare engineers.
Cloudflare says the vulnerability has now been fixed across the affected service and that customers do not need to take any action. However, the company did not specify when the unsafe storage configuration was originally introduced or exactly how long the exposure existed.
If this article helped you, please consider supporting our work. Every small contribution keeps Abijita.com independent and running.
The incident highlights the security challenges involved in sharing physical infrastructure between multiple customers, particularly when storage previously assigned to one workload is later reused by another.
