Daily Cybersecurity News - June 11, 2026

LangGraph checkpointers let AI agents keep state across runs, but a flaw lets attackers turn that persistence into remote code execution.

An SQL injection in the SQLite checkpointer lets unprivileged users overwrite the stored graph state and force arbitrary Python execution when the agent resumes. The issue affects LangGraph versions before 0.3.0.

Anyone running LangGraph agents that use the default SQLite checkpointer on shared or multi-tenant setups is exposed, including internal tools and production workflows.

The proof-of-concept uses a single malicious state blob under 300 bytes to achieve full RCE on the next agent run.

Attackers are exploiting a max-severity flaw in Ivanti Sentry to run code as root on exposed mobile gateways.

CVE-2026-10520 lets unauthenticated users hit a command injection path and escalate straight to root. The issue carries CVSS 10.0 and is actively exploited in the wild.

The bug affects Sentry versions before R10.5.2, R10.6.2, and R10.7.1. Internet-facing appliances are the main target, especially those used for MDM and VPN enforcement.

Public PoC appeared within days of the initial patch release.

Attackers are using a path traversal bug in Langflow to write files on exposed servers.

CVE-2026-5027 is actively exploited. The flaw lets unauthenticated users traverse paths and drop files anywhere the service account can reach.

Langflow instances reachable from the internet are hit. Most setups run the default web UI on port 7860 without auth.

Researchers saw the first exploitation attempts less than 24 hours after disclosure.

A researcher keeps shipping fresh PoCs against Windows Defender, letting attackers take over systems.

The bug lets malicious signatures bypass checks and run code at high privilege. public PoC available for the RoguePlanet issue.

Affects current Windows Defender builds on Windows 10 and 11. Hits any machine running the default AV engine.

Same researcher previously released three other Defender bypasses in the last month.

Attackers found a zero-day that boots straight past BitLocker encryption on Windows machines.

The exploit abuses Microsoft Defender's offline scan feature. It spawns a SYSTEM shell during Recovery Mode reboot. No CVE assigned yet, zero-day status confirmed.

Any Windows device running BitLocker with Defender enabled faces exposure. Recovery Mode access turns the bypass into a full local compromise.

PoC triggers the shell via a single crafted XML file dropped in the scan path.

ShinyHunters claims to have hit Oracle PeopleSoft servers at more than 100 organizations, mostly universities and businesses using the HR and finance platform.

The group scanned for exposed instances and chained old plus zero-day flaws to steal HR records, payroll, and finance data across roughly 300 deployments.

The campaign is still active with no Oracle patch released yet. Many victims appear to have left the servers internet-facing without strong segmentation.

University of Nottingham is one confirmed early victim in the same wave.

A credential-stealing framework called Miasma got its source code dumped on GitHub for a short time.

It has been hitting open source projects with supply chain attacks to steal developer credentials and keys. The brief public exposure let anyone grab the full codebase before the repo was removed.

The interesting part is that the operators usually keep this kind of tooling private. Public leaks like this are rare for them and could let others copy or study the methods.

The framework has been active in multiple supply chain operations over the past year.