Squid Proxy Vulnerability "Squidbleed" Exposes Sensitive Data Since 1997 Security researchers at Calif.io have uncovered a critical memory leak vulnerability in Squid Proxy, dubbed Squidbleed (CVE-2026-47729), which has persisted in the software since 1997. The flaw mirrors the infamous Heartbleed bug, allowing attackers to read beyond memory buffer boundaries in Squid’s FTP parser, potentially exposing sensitive data from prior HTTP requests. Squid, a widely deployed open-source web proxy, is used to cache and optimize traffic for protocols like HTTP, HTTPS, and FTP, commonly found in corporate networks, schools, and public Wi-Fi hotspots. The vulnerability poses the greatest risk in shared proxy environments, where an attacker controlling an FTP server could silently harvest authentication credentials, session tokens, and API keys from other users. The exposure is limited to cleartext HTTP traffic and deployments where Squid terminates TLS. While HTTPS traffic relayed via CONNECT tunnels remains unaffected, many enterprises and legacy systems still transmit sensitive data over unencrypted HTTP, expanding the potential attack surface. The flaw was identified with assistance from Anthropic’s Claude Mythos AI model. A patch was integrated into Squid version 8 in April 2026 and backported to version 7.6 in June 2026. Organizations can mitigate the risk by disabling FTP support if unused. This discovery follows Calif.io’s recent AI-assisted findings, including a high-severity OpenSSL vulnerability and the HTTP/2 Bomb denial-of-service technique. The firm continues to leverage AI in vulnerability research, highlighting its growing role in cybersecurity threat detection.

Squidbleed: 29-Year-Old Vulnerability in Squid Proxy Exposes Sensitive Data Security researchers at Calif.io have uncovered a critical memory disclosure vulnerability in Squid Proxy, dubbed Squidbleed, which has remained undetected since 1997. The flaw, discovered with the help of Anthropic’s Claude Mythos Preview AI, allows attackers to silently leak HTTP headers including passwords and API keys from other users sharing the same proxy. ### The Vulnerability Squidbleed (CVE pending) is a heap buffer overread in Squid’s FTP directory listing parser. The bug stems from a 1997 commit that introduced logic to handle NetWare FTP servers, which used extra whitespace in directory listings. A critical oversight in the `strchr` function where the null terminator (`\0`) is treated as part of the search string causes the parser to read beyond allocated memory when no filename follows a timestamp. This results in a confirmed heap overread of up to 4,065 bytes, exposing stale HTTP request data from previously freed 4KB buffers. Since Squid reuses memory without zeroing it, an attacker controlling an FTP server can trigger the flaw via a malformed directory listing, causing the proxy to return sensitive data such as Authorization headers and session tokens as part of the FTP response. ### Attack Conditions The exploit requires: - FTP support enabled (default in Squid). - An attacker-controlled FTP server reachable on TCP port 21 (allowed by Squid’s default `Safe_ports` ACL). - Victim traffic passing through the proxy as cleartext HTTP (HTTPS CONNECT tunnels are unaffected). Researchers demonstrated the attack by leaking Authorization headers from a login page via a shared Squid proxy. A proof-of-concept exploit is publicly available on GitHub. ### The Fix & Mitigation The vulnerability was patched with a single-line null check before each `strchr` call: ```c - while (strchr(w_space, *copyFrom)) + while (copyFrom && strchr(w_space, copyFrom)) ``` The fix has been merged into Squid’s repository. Administrators are advised to disable FTP support unless explicitly required, as modern browsers (including Chromium-based ones) no longer support FTP, making legitimate proxy traffic rare. ### AI-Assisted Discovery The flaw was identified using Claude Mythos Preview, which flagged the `strchr` null terminator behavior during an analysis of Squid’s FTP state machine. This follows a growing trend of AI-assisted security audits, with the team previously uncovering an HTTP/2 vulnerability using OpenAI’s Codex Cyber. The discovery highlights how LLMs trained on C standards can detect subtle API contract violations that evade traditional code reviews.