CVE Watch26 de out. de 2026

CVE-2026-45829

A pre-authentication, code injection vulnerability in version 1.0.0 or later of the ChromaDB Python project allows an unauthenticated attack

CVSS

10.0

Crítica

EPSS

15.1%

p97

KEV

—

Exploit Today

0-100

Publicado: 18 de mai. de 2026 · Última mod.: 26 de out. de 2026 · CWE-94 · CWE-502

EPSS · 30d

1.0%EPSS · 30 dias15.1%

2026-10-142026-11-11

Descrição técnica

A pre-authentication, code injection vulnerability in version 1.0.0 or later of the ChromaDB Python project allows an unauthenticated attacker to run arbitrary code on the server by sending a malicious model repository and trust_remote_code set to true in the /api/v2/tenants/{tenant}/databases/{db}/collections endpoint.

Referências oficiais

CVEs relacionados

CVECVSSEPSSKEVRExploitTítuloMod.

CVE-2026-1176877.3 ALT

—

———Movable Type provided by Six Apart Ltd. contains a code injection vulnerability in its Dynamic Publishing functionality. By sending a specially crafted request that is processed by the Dynamic Publishing, an attacker may be able to execute arbitrary PHP code on the affected product.6h CVE-2026-126673—

—

———Data::Dump::Streamer versions before 3.00 for Perl allow code injection via unescaped Regexp patterns in _dump_qr. A Regexp's pattern is written into a qr literal without escaping "$" or "@", so "@{[ ... ]}" and "${\ ... }" in the pattern run as code when the dump is evaluated. An already escaped delimiter gets a second backslash, which leaves the delimiter bare and ends the qr literal early. An attacker who supplies a pattern that an application compiles, dumps and later restores with eval or do can run arbitrary Perl code.4h CVE-2026-1256377.2 ALT

—

———A flaw was found in the Konflux UI proxy. Hostname values defined in the KonfluxUI custom resource are inserted directly into proxy configuration files without adequate validation or sanitization. An authenticated user with permissions to modify the custom resource can inject malicious configuration directives to bypass authentication and authorization controls, potentially exposing internal backend services. In addition, providing malformed input can cause the configuration generator to fail, resulting in a denial of service (DoS) by preventing the proxy from starting.11h CVE-2026-675468.8 ALT

—

———Heym builds and runs agentic systems and business-process automations on self-hosted infrastructure. Prior to version 0.0.75, the same POST /api/mcp/fetch-tools endpoint (any authenticated user, lowest role) also accepts the stdio transport, which spawns the attacker-supplied command with attacker-supplied args and an environment merged over os.environ. The process executes on the shared backend host before MCP protocol validation fails — so side effects fire regardless of whether the command speaks MCP. This is a distinct weakness class and exploit primitive from the SSRF (Advisory 2) on the same endpoint, and needs a different fix (role gating / opt-in trust boundary rather than egress controls). This issue has been patched in version 0.0.75.12h CVE-2026-675448.8 ALT

—

———Heym builds and runs agentic systems and business-process automations on self-hosted infrastructure. Prior to version 0.0.71, Heym's workflow engine accepts an arbitrary Python code string in the playwrightCode field of a Playwright node. This string is written to a temporary .py file and executed directly by the backend process via subprocess.Popen([sys.executable, script_path]) with no sandboxing, AST validation, or role gate. Any authenticated user — including a freshly registered account — can achieve full remote code execution (RCE) on the backend host, running as the service process user (observed as root in the default Docker deployment). This issue has been patched in version 0.0.71.12h CVE-2026-675438.8 ALT

—

———Heym builds and runs agentic systems and business-process automations on self-hosted infrastructure. Prior to version 0.0.70, heym hardens one untrusted-code path but not its sibling. The Python tool executor (python_tool_executor.py) treats user code as untrusted and mandates a Docker sandbox (no network, read-only fs, non-root, no docker socket), failing closed. The skill executor (skill_python_executor.py) runs skill Python via a local backend subprocess with no Docker isolation — only a 512 MB RLIMIT and a 7-key env denylist — so a skill executes inside the backend execution context, with filesystem access, network egress, and non-denylisted environment variables that should not be exposed to untrusted/shared skills. In the shipped Docker config (backend runs as root + /var/run/docker.sock mounted), this escalates to host root. This issue has been patched in version 0.0.70.12h