Crow's Nest - 2026-06-15

A roundup of 292 items curated from across the security community.

News

• AI Misidentification Leads to Wrongful Arrest, Months in Prison by Kim Zetter.

AI facial recognition identified Jalil Richardson with only 85% accuracy. Police never checked his alibi. He spent months in prison and lost his job, home, and child custody before police admitted the AI was wrong.

• Unpatched Firefox Focus Universal XSS 0-Day by Simone Margaritelli.

An unpatched universal XSS vulnerability in Firefox Focus allows any website to execute arbitrary JavaScript in the context of another origin. No fix is currently available.

• Microsoft Packages Laced with Credential Stealer for Second Time in Weeks by Nicolas Krassas.

For the second time in weeks, official Microsoft packages were found laced with credential-stealing malware, raising serious questions about supply chain integrity in the Microsoft ecosystem.

• ServiceNow Potential Breach via Unauthenticated API Endpoint by Florian Roth.

A Scripted REST API endpoint in ServiceNow required no authentication and logged activity under a Guest user. The endpoint had been in this state since at least 2018 and was only patched when reports surfaced of suspicious access across multiple customer tenants.

• Chinese Tracking Device Found in UK Prime Minister’s Car by Florian Roth.

A Chinese-manufactured tracking device was discovered embedded in the UK Prime Minister’s official vehicle, raising fresh concerns about supply chain surveillance risks in government fleets.

• NSO Group Hacking WhatsApp Despite Court Order by Bruce Schneier.

WhatsApp caught NSO Group phishing its users in violation of a court order. The spyware vendor continued targeting users even after legal action explicitly barred such activity.

• 900 Arch Linux AUR Packages Compromised with Infostealer and Rootkit by hasherezade.

An attacker posed as a trusted maintainer and adopted orphaned AUR packages, infecting nearly 900 Arch Linux packages with infostealer malware and a rootkit in one of the largest supply chain attacks on the AUR.

Techniques and Write-ups

• Arbitrary Code Execution in objdump via Relocation-Oriented Programming by ϻг_ϻε.

Calif.io continues their streak of hacking security tools with an arbitrary code execution in objdump -g, using a novel relocation-oriented programming technique to exploit DWARF parsing.

• Linux nftables Use-After-Free Exploit from One Misplaced Exclamation Mark by Alex Plaskett.

Exodus Intel walks through exploiting a Linux kernel use-after-free in nftables caused by a single misplaced exclamation mark, achieving local privilege escalation with a detailed technical write-up.

• Linux io_uring Exfiltration Tool Bypasses EDR by Florian Roth.

A new Linux exfiltration tool abuses io_uring to asynchronously read /etc/shadow and exfiltrate credentials over TCP, completely bypassing EDR solutions that rely on traditional syscall monitoring.

• GPS As a Key Distribution Platform by Bruce Schneier.

The U.S. military has quietly been broadcasting encryption key material through public GPS satellites for nearly 20 years, effectively turning every GPS-enabled device into an unwitting receiver of government cryptographic data.

• RoguePlanet: New Post-Exploitation Framework from MSNightmare by Florian Roth.

A new C2/post-exploitation framework with ISO embedding, ADS abuse, and Defender lure paths. Florian Roth published IOCs including named pipes, temp file paths, and alternate data stream markers.

• Shai-Hulud CI/CD Worm Spreads via Malicious PyPI Package by Florian Roth.

Shai-Hulud is a proof-of-concept CI/CD worm that propagates through PyPI packages, demonstrating how a single compromised dependency can spread through build pipelines autonomously.

• Malware Embeds CBRN Text to Trigger LLM Safety Refusals by Mari0n.

Malware developers are embedding nuclear and biological weapons text into their code to trigger safety refusals in AI-powered security scanners, exploiting alignment guardrails as an evasion technique.

• Opening a Coding Agent in a Hostile Repo Executes Repo Code at Startup by Giuseppe N3mes1s.

Origin Security found that simply opening the Opencode coding agent inside a hostile repo runs the repo’s own code at startup, before the model is even in the loop. No user command needed.

• Oops, I Weaponized the Database: Abusing AI Features in SQL Server 2025 by Justin Kalnasy.

SpecterOps researchers demonstrate how native AI features in Microsoft SQL Server 2025 can be abused for data exfiltration and C2 transport directly within the database engine, with working proof-of-concept code published.

• PostgreSQL pgcrypto Heap Buffer Overflow Leading to RCE by Swissky.

CVE-2026-2005 is a heap buffer overflow in PostgreSQL’s pgcrypto extension that leads to remote code execution. Wiz published a full technical analysis of the vulnerability.

• ITScape: First Public KVM/arm64 Guest-to-Host Escape by Lucas Leong.

The first publicly demonstrated guest-to-host escape targeting in-kernel KVM on arm64. Unlike QEMU escapes, the bug lives in the kernel itself, running exploit code with host kernel privilege and threatening multi-tenant arm64 cloud isolation.

• SQL Server 2025 AI Features Enable Exfil, NTLM Coercion, and C2 by Aurélien Chalot.

SpecterOps research shows SQL Server 2025’s native AI integration features can be abused for data exfiltration, NTLM coercion, and C2 transport, all using built-in functionality that works as designed.

• GreatXML: BitLocker Bypass via Windows Recovery Environment by kmkz.

A BitLocker bypass triggered by abusing Windows Defender Offline Scan state, dropping unattend.xml onto the recovery partition to boot into WinRE with unrestricted access to the encrypted volume.

• Hacking Google with AI for $500,000 by Devansh (, ).

An extensive writeup on using AI-assisted techniques to find vulnerabilities in Google, resulting in a $500k bug bounty payout. Praised for the depth of its technical detail.

• CVE-2026-41089: Netlogon CVSS 9.8 Actively Exploited in the Wild by kmkz.

CVE-2026-41089, a CVSS 9.8 flaw in Windows Netlogon, lets unauthenticated attackers gain SYSTEM on domain controllers with crafted packets. Active exploitation in the wild has been confirmed; a patch has been available since May.

• CVE-2026-50751: Check Point VPN Authentication Bypass by McCaulay.

watchTowr details how Check Point’s Remote Access VPN products allow an IKE client to skip authentication entirely, bypassing the security boundary these products are designed to enforce.

• Evilginx at x33fcon: Downgrading FIDO MFA with AI by Kuba Gretzky.

Kuba Gretzky demonstrated live at x33fcon how Evilginx can downgrade FIDO-based MFA protections, successfully phishing Google credentials on stage despite hardware key enforcement.

• CVE-2026-20253: Pre-Auth RCE in Splunk Enterprise.

watchTowr’s analysis of a pre-authentication remote code execution in Splunk Enterprise, exploiting the database layer’s own authentication to bypass application-level controls.

• DCOMIllusionist: Fileless Lateral Movement via DCOM by Synacktiv.

Synacktiv released their x33fcon research on offensive DCOM techniques, including a COMouflage variant for arbitrary executable execution and a fileless lateral movement method based on .NET deserialization.

• SearchLeak: One-Click Data Exfiltration via M365 Copilot by Nicolas Krassas.

Varonis researchers turned Microsoft 365 Copilot into a one-click data exfiltration weapon, abusing its search and summarization capabilities to extract sensitive data from across the tenant.

Tools and Exploits

• Ghidra-RPC: Agentic Reverse Engineering Tool from Cellebrite Labs by Florian Roth.

Ghidra-RPC enables agentic reverse engineering by exposing Ghidra’s analysis capabilities over an RPC interface, letting AI agents and automation tools interact with disassembled binaries programmatically.

• Cobalt Strike 4.13: Beacon Interpreter and LLVM Beacon by Rasta Mouse.

Major Cobalt Strike release introducing a Beacon Interpreter for native C scripting, an LLVM-compiled Beacon, improved docking UX, and better payload management.