Google patched a critical flaw in Gemini CLI, the command-line tool developers use to interact with Gemini models from CI pipelines and dev workstations. CVSS 10.0. The bug let an attacker execute arbitrary code on the developer's machine by feeding crafted input to the CLI - specifically through the same pattern that compromised LiteLLM and several other AI tools recently. A separate but related set of flaws in Cursor, the AI-powered IDE, also enables code execution. The pattern across all these AI dev tools is the same: input validation gaps where attacker-controlled prompts or model output reach a shell or code execution path.
Update on the GitHub flaw covered yesterday: Wiz, who found the bug, published its full disclosure showing 88% of self-hosted GitHub Enterprise Servers were still unpatched at public disclosure on April 28. The bug let any user with push access to one repository run code on the GitHub server itself with a single 'git push'. On GitHub.com, the same bug exposed millions of public and private repositories belonging to other users sharing the same storage node. GitHub.com was patched within 75 minutes, but Enterprise Server installs need patching manually. Wiz found the bug using AI-augmented reverse engineering on closed-source GitHub binaries.
Aisle, an AI-driven application security firm, ran its analyzer over OpenEMR's source code and found 38 previously unknown vulnerabilities, including two with maximum severity (CVSS 10.0). OpenEMR is the open-source electronic health records system used by 100,000 healthcare providers serving 200 million patients. The two critical bugs let attackers reach into patient databases without logging in: CVE-2026-24898 lets any unauthenticated visitor receive the medical practice's API tokens by sending a single POST request, and CVE-2026-24908 is a SQL injection in the patient REST API. OpenEMR has now patched all 38.
Researchers at Theori and Xint disclosed Copy Fail yesterday, a Linux kernel bug introduced in 2017 that lets any unprivileged user with shell access become root in seconds. The exploit is a 732-byte Python script that works without version-specific tweaks on every major Linux distribution since 2017 - Ubuntu, Amazon Linux, RHEL, SUSE. Unlike previous kernel bugs (Dirty Cow, Dirty Pipe), Copy Fail has no race condition and no per-kernel offsets. It also leaves no trace on disk because it only modifies the in-memory page cache. The bug was found using AI-assisted reverse engineering and has been hiding in the open for nearly nine years.
SonicWall released emergency firmware updates for Gen 6, Gen 7, and Gen 8 firewalls after CrowdStrike's research team disclosed three SonicOS flaws on April 29. The worst is CVE-2026-0204 (CVSS 8.0), a weak authentication bug in the management interface that lets an attacker on an adjacent network reach management functions without logging in - and from there change firewall rules, disable security protections, or open new holes. The other two are post-authentication: CVE-2026-0205 is a path traversal that breaks out of restricted directories, and CVE-2026-0206 is a buffer overflow that crashes the firewall. No public exploits yet.
LiteLLM, the popular open-source gateway used to centralize API access for OpenAI, Anthropic, and other AI providers, has a critical pre-authentication SQL injection bug that attackers started exploiting just 36 hours after the security advisory went public. The flaw lets anyone who can reach the proxy port read all the API keys stored inside - including master keys, virtual keys, and provider credentials. The bug was in the bearer-token check: the token was concatenated into a SQL query instead of passed as a parameter. Sysdig saw the first attack at 04:24 UTC on April 26, hitting three tables that hold the most valuable secrets.
Researchers disclosed CVE-2026-3854, a critical GitHub Enterprise Server flaw that lets anyone with push access execute arbitrary commands on the GitHub server with a single git push. The bug is in how Enterprise Server handles repository hooks during push operations - a crafted commit message or filename bypasses the sanitization that normally prevents shell injection. GitHub patched it last week, but self-hosted instances need to apply the patch manually, and telemetry shows most haven't yet. Anyone with developer-level access to a vulnerable Enterprise Server can take over the entire instance, then pivot into every repository and CI/CD secret it hosts.
Microsoft confirmed yesterday that a Windows Shell spoofing flaw, CVE-2026-32202, is being exploited in the wild. The bug lets an attacker craft files that appear in File Explorer with fake names, icons, and paths - so a malicious .exe can show up looking like a benign PDF, leading users to double-click and run it. Microsoft patched the bug in the April 14 Patch Tuesday but only confirmed in-the-wild exploitation on April 28, raising urgency for any environment that hasn't deployed April patches. The flaw is particularly dangerous on shared file servers, USB drops, and email attachments - any path where users trust File Explorer to tell them what's what.
Microsoft quietly patched a privilege escalation flaw in Entra ID (formerly Azure AD) that let an attacker with a low-privileged service account take over any service principal in the same tenant - including high-value ones with admin consent grants. The bug was in how Entra ID validated role assignments during certain API calls: the validator checked whether the caller had any role on a service principal but didn't check whether that role authorized the specific action. Microsoft fixed the flaw on the back end, so customers don't need a patch - but the takeover scenario means anyone who exploited it before the fix could have created persistent backdoors via OAuth grants.
Researchers disclosed a critical unauthenticated remote code execution flaw in Hugging Face's LeRobot, the open-source framework used to train and deploy ML models on physical robots. CVE-2026-25874 sits in the framework's web interface, which by default listens on all network interfaces with no authentication - quick for demos, but a hard fail when the demo box ends up on a corporate network. There is no patch yet. Hugging Face has been notified but hasn't released a fix. Particularly serious because LeRobot is usually attached to actual robotic hardware, so a compromise can mean unsafe physical actions.
TrustStrike Labs