Cloudflare has fixed a vulnerability in its Containers service that could let one customer read data left behind by other customers' containers on the same server. The issue was in thin-provisioned disks: when a container was deleted, its 64-kilobyte blocks returned to a shared pool that was configured to skip wiping before reuse. A new container that wrote only a small amount into a reused block could read the rest of the block, which still held the previous customer's data. The exposure was limited to leftover disk data, not live workloads, and an attacker could not choose whose data they received.
Researcher Oren Yomtov of Accomplish reported the flaw on September 4 through Cloudflare's bug bounty program. In production tests, the team found leftover material on 18 of 24 attempts across 20 of 22 underlying machines on four continents, including directory structures, SQLite databases, and credential files. Cloudflare first re-enabled block wiping for newly handed-out blocks, then retired running container disks and cleared image-layer caches to remove already-mapped data. The cleanup finished on September 19, and Cloudflare disclosed the issue five days later, saying customers need take no action.
Cloudflare said it found no evidence that anyone else had used the method, though it did not specify how long the unsafe setting had been active. The researchers say the same disk setup affected Cloudflare's Browser Run product; Cloudflare's disclosure named only Containers and Sandboxes. They also describe the finding as their sixth sandbox escape since July, following issues in Anthropic's Claude tools, Cursor's CLI, Docker, and OpenAI's Codex.