Encrypted drive Isn’t Encrypted: YellowKey Vulnerability Cracks BitLocker Open

This week made it clear that four components millions of admins and users quietly trust contain critical flaws — three of them for years. Welcome in Security Sunday.

On May 12, 2026, a security researcher operating under the handles Chaotic Eclipse and Nightmare-Eclipse published a proof-of-concept exploit on GitHub called YellowKey that bypasses BitLocker encryption on Windows 11 and Windows Server 2022/2025.

The attack requires physical access to the device and an ordinary USB drive. No recovery key, no password, and no specialized hardware. The researcher who discovered the vulnerability called it “one of the craziest discoveries I’ve ever made” and believes it’s an intentionally left backdoor. According to current findings, Windows 10 is not affected.

The mechanics of the attack are surprisingly simple. The attacker copies a prepared FsTx folder into the System Volume Information directory on a USB drive, plugs it into the target machine, and reboots into the Windows Recovery Environment (WinRE) via Shift + Restart. At the right moment they hold down the CTRL key and, instead of the expected recovery menu, a command prompt appears with full access to the decrypted disk volume.

The vulnerability affects the default BitLocker configuration, which relies on TPM (Trusted Platform Module) alone — the setting that ships on most consumer and corporate Windows 11 laptops. Because the TPM holds the encryption keys locally, the attack does not work if you pull the drive out of the device. If the attacker has physical access to the laptop, however, the encryption can be bypassed.

YellowKey is the fifth zero-day this researcher has published in 2026. It follows the BlueHammer, RedSun, and UnDefend vulnerabilities.

Released alongside YellowKey is a second attack called GreenPlasma, which enables local privilege escalation in the CTFMON component and can hand the attacker a shell with SYSTEM privileges. The researcher justifies the disclosure by long-standing dissatisfaction with how the Microsoft Security Response Center handles reports, and promises further “surprises” for the June Patch Tuesday.

As of the time of writing, Microsoft has not released an official patch and YellowKey has not yet been assigned a CVE number.

Do you know whether your systems are secure?

Every week I bring examples of attacks and vulnerabilities that can affect anyone. If you are not sure whether your infrastructure is resilient, we would be happy to help you find out. At SysNetShield we offer vulnerability scanning, penetration testing, Red Teaming, and simulated phishing campaigns.

Get in touch — your first consultation or vulnerability scan for one domain or IP address is free.

18-Year-Old Vulnerability in NGINX

On May 13, F5 — together with researchers from the company depthfirst — disclosed a critical vulnerability in the NGINX web server, given the codename NGINX Rift. Tracked as CVE-2026–42945, it received a CVSS score of 9.2 out of 10.

The bug lives in the ngx_http_rewrite_module and allows an unauthenticated attacker to crash the server’s worker processes and, under certain conditions, even achieve remote code execution (RCE).

Technically it’s a classic heap buffer overflow, but with an unusual root cause. The NGINX scripting engine processes rewrite directives in two phases: it first calculates the required size of the memory buffer, and only then writes data into it. The attacker arranges for a smaller buffer to be allocated than is ultimately needed, and the write therefore overruns into adjacent data. A single specially crafted HTTP request is enough to trigger the bug.

The way the vulnerability was uncovered is also worth noting. The researchers didn’t find it through a traditional manual audit, but using an autonomous code-analysis system that depthfirst is developing.

Fortunately, the fix is straightforward — just upgrade to NGINX 1.31.0 or 1.30.1.

Pwn2Own Berlin Produced 24 Zero-Day Vulnerabilities

From May 14 to 16, the prestigious Pwn2Own hacking competition returned to Berlin as part of the OffensiveCon conference. This year’s edition focused on enterprise technologies and artificial intelligence, putting up 31 targets across 10 categories — including, for the first time, dedicated tracks for AI databases, coding agents, local LLMs, and NVIDIA AI hardware.

The star of day one was Taiwanese security researcher Cheng-Da Tsai — better known by the handle Orange Tsai — from the DEVCORE Research team. He chained four logic bugs to break out of the Microsoft Edge browser sandbox, earning a $175,000 bounty.

Windows 11 also drew major attention, getting compromised three times in the local privilege escalation category on day one alone. The duo of Angelboy and TwinkleStar03 from the DEVCORE Internship program leveraged an access-control flaw, Marcin Wiązowski demonstrated a buffer-overflow exploit, and Kentaro Kawane of GMO Cybersecurity chained two use-after-free vulnerabilities. Each of these attacks earned its authors a $30,000 reward.

This year’s edition clearly confirmed the growing weight of attacks against AI. A researcher going by the handle k3vg3n chained three vulnerabilities — including SSRF (Server-Side Request Forgery) and code injection — to gain full control over the LiteLLM platform.

Valentina Palmiotti, of the IBM X-Force Offensive Research team, picked up $50,000 for a zero-day in the NVIDIA Container Toolkit and a further $20,000 for gaining root on Red Hat Enterprise Linux for Workstations. Other successfully compromised targets included NVIDIA Megatron Bridge, the Chroma vector database, and the LM Studio application.

Day two of the competition brought another 15 vulnerabilities and payouts totaling $385,750. Orange Tsai once again claimed the biggest win of the day, chaining three bugs to achieve remote code execution with SYSTEM privileges on a Microsoft Exchange server, taking home $200,000 for it.

Attacks also targeted Microsoft SharePoint, the Safari and Mozilla Firefox browsers, the Cursor editor, and Anthropic’s Claude Code coding agent. Under Pwn2Own rules, vendors of the affected products have 90 days to release security patches before ZDI publishes the technical details of the individual vulnerabilities.

ssh-keysign-pwn Leaks SSH Keys and Passwords on Linux

On Thursday, May 14, the Qualys Threat Research Unit disclosed a Linux kernel vulnerability tracked as CVE-2026–46333 and nicknamed “ssh-keysign-pwn.” The bug allows any unprivileged local user to read files owned by root — specifically, the private host SSH keys in /etc/ssh/ and the contents of /etc/shadow.

A few hours after the patch landed, a security researcher operating under the handle “_SiCk” published two fully working exploits on GitHub, dramatically shortening the window administrators have to roll out the fix.

In practice, the public proof-of-concept targets two SUID binaries that are standard on practically every Linux distribution. The first variant abuses ssh-keysign, an OpenSSH helper binary that opens private host keys (ssh_host_ecdsa_key, ssh_host_ed25519_key, ssh_host_rsa_key).

The second variant targets the chage -l utility, which opens /etc/shadow.

The race condition doesn’t win on the first try, but according to public analyses the exploit typically succeeds within 100 to 2,000 attempts — on current hardware, a matter of seconds.

The historical context of this bug is also remarkable. A similar problem was described by Jann Horn of Google Project Zero back in 2020, who proposed a fix at the time — one that was never deployed. The vulnerability thus lingered in the kernel for roughly six years before Qualys rediscovered it.

ssh-keysign-pwn is the fourth critical Linux kernel vulnerability disclosed in just three weeks, following Copy Fail (April 29), Dirty Frag (May 7), and Fragnesia (May 13).