Vulnerability in AWS SDK Put Thousands of Companies at Risk
Researchers at the Israeli security firm Pi discovered a vulnerability in official AWS SDKs that allows requests to be rerouted to attacker-controlled servers. This flaw exposes sensitive cloud credentials.
When you use the AWS API, you trust that the SDK communicates exclusively with Amazon's servers. However, as the Israeli information security company Pi discovered, this is not always the case. Researchers found that by manipulating the region string, it was possible to force the SDK to route requests directly to servers controlled by attackers.
"Much bigger than a point bug"
The access address generated by the SDK follows a fixed template: https://{service}.{region}.amazonaws.com. When researchers entered the string "@attacker.com#" in the region field instead of a legitimate value, the URL decoder read everything before the @ as a username and omitted everything after the #, effectively redirecting traffic to "attacker.com".
In an interview with Geektime, Guy Arzi, CEO of Pi, explained that the vulnerability allowed for the potential theft of cloud credentials. According to Arzi, the company identified this pattern in 7 different SDKs, across 6 languages, and in more than 2,000 packages. "The JavaScript SDK alone is downloaded about 2.3 billion times a week," he notes.
The vulnerability did not affect all languages equally: Python and Ruby SDKs had included protection against this vector for years, while others did not. Since AWS SDKs for hundreds of services are derived from shared templates, a failure in one generator automatically propagated to every language using it.
Admin access to production
The system creates access addresses by inserting the region definition into a template without prior validation. The real danger arises in requests like AssumeRoleWithWebIdentity, where access tokens are sent in plain text, facilitating identity theft.
"The significant part for us was to prove the impact in the real world," Arzi adds. "With the approval of a well-known security provider, we managed, without any interaction on their part, to obtain credentials with administrator permissions to their production environment in AWS. We subsequently found the same active vulnerability in dozens of other companies, including some Fortune 500 firms, and reported it to each of them for remediation."
The dissonance of security ratings
Following the report, AWS released updates for seven different SDKs but assigned an official CVE—CVE-2026-22611—only to the .NET version. Other SDKs were patched without a CVE assignment. AWS never officially defined these findings as a "vulnerability," describing the fix instead as a "defensive improvement" and claiming that input validation is the developer's responsibility.
Arzi is critical of this response: "AWS rated the vulnerability at 3.7, which is Low. But in practice, it allowed us to gain administrator permissions within a production environment without any user action. The problem is that CVSS measures vulnerabilities in isolation, ignoring the environment in which they reside—whether they are sitting next to critical systems or sensitive identity data."
