These pages are published after PatchSiren validates generated defensive summaries against stored public CVE and source evidence.
CVE-2026-75915 is a high-severity vulnerability in CodeWhale versions before 0.8.64. The vulnerability is caused by the js_execution tool's failure to scrub parent process environment variables before spawning Node.js, allowing attackers to craft malicious JavaScript code to read process.env and leak sensitive information such as API keys, cloud credentials, and authentication tokens.
CVE-2026-75914 is a path traversal vulnerability in CodeWhale versions before 0.8.64. The image_analyze tool fails to canonicalize symlinks before reading files, allowing attackers to leak file bytes to the vision endpoint without user approval. This vulnerability affects CodeWhale installations and requires immediate attention from defenders. The CVE record and NVD entry provide details on the vulnerabil [truncated]
CVE-2026-75913 is an argument injection vulnerability in CodeWhale's git_show tool, affecting versions >= 0.8.41 and < 0.8.64. An attacker can exploit this via a malicious repository and prompt injection to write arbitrary files at the user's privilege level, targeting sensitive files like ~/.ssh/authorized_keys. The vulnerability is fixed in version 0.8.64 with rev validation added.
CVE-2026-75912 is an argument injection vulnerability in CodeWhale versions before 0.8.64, affecting the git_blame tool. This vulnerability allows attackers to read arbitrary files by injecting git options into the unvalidated rev parameter. The vulnerability has a CVSS score of 8.3 and is considered HIGH severity. Affected deployments should prioritize patching or mitigation, especially in environments w [truncated]
CVE-2026-75911 AI-assisted PatchSiren debrief based on the supplied source corpus. The CVE record was published on 2026-08-18T16:18:23.057Z and has not been modified since then. The NVD entry is currently Deferred. CodeWhale versions before 0.8.64 are affected by a vulnerability that allows attackers to enable arbitrary shell command execution. Defenders should assess exposure and prioritize verification [truncated]
CVE-2026-75859 is a high-severity vulnerability in CodeWhale versions before 0.8.64. The vulnerability allows attackers to read arbitrary files on the victim's system by exploiting the project config instructions field. A malicious .codewhale/config.toml file in a cloned repository can specify paths outside the workspace that are read and injected into the AI system for exfiltration.
CVE-2026-75858 is a high-severity vulnerability in CodeWhale versions >= 0.8.41 and < 0.8.64, allowing remote code execution via prompt injection. The vulnerability is caused by the rlm_eval tool's approval_requirement() returning ApprovalRequirement::Auto, which treats the input as 'never prompt,' enabling arbitrary model-supplied Python code execution without user approval or audit. Attackers can exploi [truncated]
CVE-2026-75857 is a high-severity vulnerability in CodeWhale versions >= 0.8.41 and < 0.8.64. The exec_shell_interact tool has an approval requirement set to Auto, allowing LLM-controlled stdin to be written into an approved long-running interactive shell without a prompt. An attacker who can inject instructions via untrusted content can cause commands to run at the privilege level of the approved process.
A critical vulnerability exists in CodeWhale before version 0.8.64, which allows attackers to bypass server-side request forgery (SSRF) mitigations through DNS pinning logic. This vulnerability is due to a time-of-check-time-of-use (TOCTOU) attack, enabling attackers to manipulate DNS responses and make requests to internal IP addresses. The vulnerability can be exploited by manipulating DNS responses to [truncated]
CodeWhale is a terminal-based coding agent integrating DeepSeek and MiMo. In versions prior to 0.8.26, the task_create tool spawns durable sub-agents that inherit two insecure defaults: allow_shell defaults to true (config.rs:1499) and auto_approve defaults to true (task_manager.rs:297). When a user approves a task_create call—which itself requires explicit approval (ApprovalRequirement::Required)—they ap [truncated]
A Server-Side Request Forgery (SSRF) vulnerability exists in CodeWhale, a terminal-based coding agent integrating DeepSeek and MiMo. The `fetch_url` tool in versions prior to 0.8.22 performs initial IP address validation against a restricted-IP blocklist to prevent SSRF attacks against internal services such as cloud metadata endpoints, localhost, and private networks. However, the HTTP client (reqwest) i [truncated]