Breaking

29-Year-Old Squid Proxy Leak Bug Found With Help From Claude

June 22, 2026 at 12:46 EDT

Security
AI Agents

Squidbleed (CVE-2026-47729), a data-leaking flaw that had lurked in the open-source caching proxy Squid since 1997 — roughly 29 years — has been disclosed. It was uncovered with the help of Anthropic's Claude Mythos Preview and fixed in Squid 7.6.

CVE-2026-47729 · Discovered with Claude Mythos Preview

"Squidbleed": a 29-year-old flaw that can leak other users' passwords

A heap buffer overread in the Squid web proxy's FTP parser — present in default configs since 1997 — can mix one user's HTTP headers, including passwords and API keys, into another's response.

~29 yrs

Dormant in the codebase since a Jan 18, 1997 commit

4 KB

Shared MEM_4K_BUF pool, not zeroed on reuse — leaving stale data

1 line

Patch: check for a null terminator before calling strchr

How long it stayed hidden — years undetected

2014

Heartbleed
(OpenSSL)

~29 yrs

Squidbleed
(Squid)

Same Heartbleed-style leak — but lurking for nearly three decades before AI-assisted analysis surfaced it.

⚠ Affected

Default configs (TCP 21 in Safe_ports, FTP enabled)
Cleartext HTTP or TLS-terminating setups
Shared proxies: enterprises, schools, public Wi-Fi

✓ Not affected / Mitigated

Standard HTTPS opaque CONNECT tunnels
Fixed in Squid v7.6 (June 8, 2026) and later
Disabling FTP support removes the attack surface

The attack path

Attacker FTP server

reachable on port 21

→

FtpGateway parser

strchr overread

→

Leaked response

other users' Authorization headers

Real-world exploitation reports remain limited so far — but the episode shows how dormant old C code stays an attack surface.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.