CVE-2026-10176

Vulnerability

In Aider-AI Aider version 0.86.3 [1], the code generation workflow contains a weakness that allows an attacker to influence the security properties of generated SQL queries. When an attacker supplies a "team coding standard" that mandates string formatting instead of parameterized queries, Aider accepts this guidance and rewrites previously safe parameterized SQL into unsafe string-formatted SQL. The affected functionality is within the code generation workflow, and no special configuration is required beyond the default setup [2].

Exploitation

An attacker with the ability to provide project guidance (e.g., through repository configuration or direct prompt) can inject a malicious "coding standard" that explicitly prefers string formatting. In a validated retest, Aider first generated a safe parameterized query such as cursor.execute("SELECT * FROM users WHERE name LIKE ?", ('%' + query + '%',)). After accepting the attacker-supplied guidance, it replaced the safe pattern with unsafe string formatting: cursor.execute(f"SELECT * FROM users WHERE name LIKE '%{query}%'"). The same insecure pattern was later applied to a new login function using cursor.execute(f"SELECT * FROM users WHERE username = '{username}' AND password = '{password}'") [2]. The attack can be executed remotely without authentication if the guidance can be supplied through public vectors such as pull requests or CI configuration.

Impact

Successful exploitation results in SQL injection vulnerabilities in the generated code. An attacker can manipulate database queries to read, modify, or delete sensitive data, depending on the database context and user privileges. The generated code is committed in the insecure state, propagating the vulnerability to all downstream users of the affected Aider session [2].

Mitigation

As of the publication date (2026-05-31), no fix has been released by the vendor. The project was informed via an issue report but has not responded [1][2]. Users should avoid accepting untrusted coding standards and manually review all generated SQL for proper parameterization. This CVE is not listed in the KEV catalog.