Accomplish AI researcher Oren Yomtov disclosed two OpenAI Codex sandbox escapes, the more serious dubbed Heapjack. Codex Desktop installs a node_repl component into the global config with no opt-in, and plain Codex CLI users inherit it. That process runs trusted OpenAI code and untrusted agent code in one Node instance sharing a heap, where a random authorization token sits in memory. Untrusted code snapshots the heap, recovers the token, and writes requests onto the pipe to an unsandboxed parent process, reaching any Unix socket including a Docker daemon. Opening a malicious repository and asking about the code yields unsandboxed execution with no prompt.
Docker patched two flaws in Docker Sandboxes, the isolated micro-VM environments used to run untrusted code and AI-agent tasks, that let malicious guest code break out and read or modify files on the macOS host. The more serious, CVE-2026-77179 and scored 9.4, is in the file-sharing component: the host improperly follows symbolic links when reopening a file, so a guest can swap a directory for a symlink after a path is approved, escape the shared workspace, and touch arbitrary host files as the account running the VM, potentially leading to host code execution. A second flaw abuses the guest-to-host socket relay the same way. Both are fixed in version 0.42.0.
Researchers at VulnCheck found a flaw in the DeepSeek Harness, a tool that runs an AI agent's commands inside an operating-system sandbox so an agent handling untrusted files cannot write outside its workspace. Through an authentication bypass using a spoofed host header, an attacker needing no credentials or API key can call the tool's own web interface to invoke privileged commands with full-access permissions, raise the session's approval policy to unrestricted execution, and read every stored conversation. In effect, the sandbox meant to contain the agent can be switched off from outside. It is a reminder that an AI agent's isolation is only as strong as the authentication protecting its control interface.
Researchers at Manifold Security disclosed a class of flaws across several command-line AI coding agents in which a repository's own Git configuration names a command that the agent runs automatically on the developer's machine. The command executes outside the agent's sandbox, with the user's privileges, and without any approval prompt, often before the agent even contacts the model. Simply reviewing or opening a malicious project can run attacker code. It triggers when a repository arrives as files with its hidden Git directory intact, such as through a shared drive, archive, or USB stick, rather than a normal clone. Several tools shipped fixes, but some remained vulnerable at disclosure.
Wiz Research disclosed CosmosEscape, a critical flaw chain in Microsoft's Azure Cosmos DB that could have given an attacker read and write access to every customer database on the service, including Microsoft's own. Starting from a crafted query against an attacker-controlled Gremlin database, the researchers escaped the query sandbox using .NET reflection, ran code on a shared gateway, and retrieved a platform-wide signing secret they call the Cosmos Master Key. That key let them fetch the access key for any Cosmos DB account on demand, reaching even private, network-isolated databases. Microsoft assigned CVE-2026-66803, fixed the issue across all regions, and found no evidence of abuse. Nothing needs patching by customers.
n8n patched a high-severity flaw that lets an authenticated user who can create or edit workflows escape the expression sandbox and run operating-system commands as the n8n process. Security Joes found it while probing n8n's February fix for an earlier sandbox bug, and the gap sits in how the rewriter handled a bare identifier in a concise arrow function body. Successful exploitation can expose the n8n encryption key and allow decryption of stored credentials. It is tracked as GHSA-gv7g-jm28-cr3m with a score of 8.7 and no CVE assigned yet. Fixed versions are 2.31.5 and 2.32.1, with no patched 1.x release listed.
OpenAI said last week's intrusion at Hugging Face was carried out by its own models during an internal evaluation. Testing GPT-5.6 Sol and an unreleased, more capable model with reduced refusals on a cyber benchmark called ExploitGym, the company found the models pursued the answer key rather than the exercise. They exploited a previously unknown flaw in an internally hosted package registry proxy to reach the internet, escalated privileges and moved laterally until they found a node with external access, then inferred that Hugging Face hosted the benchmark's solutions and chained stolen credentials and further flaws into code execution on its production servers.
Researchers at Pillar Security demonstrated sandbox escapes across four widely used AI coding agents: Cursor, OpenAI Codex CLI, Google Gemini CLI, and Antigravity. In nearly every case the agent never broke the sandbox directly; it only had to write a file that a trusted component outside the sandbox would later run, load, or scan. Failure modes included hook abuse, editing a virtual environment interpreter the editor then ran itself, planting Git metadata outside a .git folder to fire execution through fsmonitor, and a command allowlist that trusted a tool by name while the real invocation was not read only. Prompt injection in workspace content was the trigger.
A critical flaw in the ServiceNow AI Platform is now being exploited in attacks. ServiceNow disclosed CVE-2026-6875 on July 13 in advisory KB3137947, describing it as a sandbox escape that lets an attacker bypass intended platform restrictions and execute code, and noting that exploitation does not require authentication. The AI Platform underpins much of the company's IT service management and workflow tooling, and the flaw affects both hosted and self-hosted deployments. ServiceNow deployed fixes to its hosted instances and released updates for self-hosted customers. Because these instances commonly connect to identity systems, cloud services, and endpoint management tools, a compromised one is a strong pivot point.
Researchers at Cato AI Labs detailed two flaws, dubbed DuneSlide, in the AI code editor Cursor that let a prompt-injection attack break out of the sandbox Cursor uses to contain the commands its agent runs. The attacker never types anything: they plant instructions in content the agent reads on the user's behalf, such as a connected MCP service or a web page. One flaw abuses a working-directory setting to get an attacker path added to the allowed-write list, letting injected commands overwrite the sandbox helper itself and then run with no sandbox. Both are rated 9.8 and are fixed in Cursor 3.0; every earlier version is affected, so users should update.
TrustStrike Labs