GlobalProtect Auth Bypass Becomes a Ransomware Initial-Access Path

Ransomware crews are actively exploiting CVE-2026-0257 to establish trusted-looking GlobalProtect VPN sessions, with Qilin and Settra tied to the activity. Kimsuky has expanded Operation GitPower toward Ukrainian policy experts using reply-driven lures and GitHub/GitLab infrastructure.

Vulnerabilities & Exploits

Ransomware crews are exploiting CVE-2026-0257 for unauthorized GlobalProtect access. Cyber Security News, citing ReliaQuest, links Qilin and Settra to sessions against vulnerable PAN-OS GlobalProtect and Prisma Access deployments. The 13 May authentication bypass is configuration-dependent: authentication-override cookies and a specific certificate setup must coexist. Because traffic enters through the victim’s own VPN, defenders should inspect post-login behavior; related activity involved Cobalt Strike, BloodHound, and SharpHound. Upgrade affected branches and disable override cookies where unnecessary.
A patch and working public PoC are now available for the GoBalance .onion key-recovery flaw. This materially advances the earlier dark-web service thread (earlier coverage) without proving attribution for Dread’s takeover. The Hacker News reports that GoBalance supplied only the first 32 bytes of a 64-byte Tor private key to its signer, allowing one public descriptor to reveal the long-term master key. Address control permits redirection but does not itself expose origin servers or databases. Original Onionbalance, Tor, and keys in GoBalance’s safer format are unaffected.
An unchecked 64-bit sum could create spendable XRP. In the official XRP Ledger disclosure, researchers Cayden Liao and Veria AI describe a crafted offer-and-payment sequence that wrapped the buyer’s total while paying offer owners in full. RippleX reproduced the mint locally and confirmed the resulting XRP could be spent in a follow-up payment. The flaw is fixed in xrpld 3.4.1; the report documents coordinated research and a bug-bounty PoC, not in-the-wild exploitation.

New Tools & Releases

NTDLL-reconstruction combines userland unhooking with Hell’s Gate variants. The library can switch a process between hooked and clean ntdll.dll stubs at runtime, making it useful for exercising detections that depend on stable userland hooks. See the repository and technical write-up.
Sentinel is an independent user-mode Windows EDR project, not Microsoft Sentinel. It combines multiple ETW providers, weighted behavioral correlation, offline artifact scoring, and MITRE ATT&CK mappings, with process-tree termination reserved for multi-signal chains. The GitHub release is useful for purple-team experimentation, but it actively kills processes and changes host security policy, so the authors recommend non-production testing.
REA gives MCP agents a common reverse-engineering interface. It can connect agents to Hopper, Ghidra, or IDA, inspect native binaries, JavaScript/Electron applications, .NET assemblies, and websites, and return locally generated findings with supporting evidence and limitations. The project supports several coding-agent clients through its GitHub repository. (discussion)
Jshound statically maps API calls from shipped JavaScript. The tool extracts methods, URLs, headers, bodies, and GraphQL operations from application bundles without interacting with the target server, offering a low-noise starting point for web assessment.
asmsense publishes an open small-model baseline for naming stripped x86-64 functions. The project uses Qwen3-8B with LoRA and releases its dataset and evaluation pipeline. Its limitations are unusually explicit: name F1 was 0.135 on a fixed 2,000-function test, the English explanation audit found 62 of 294 outputs correct, and 217 of the 2,000 test rows shared SQLite function names with training data. It is a reproducible research baseline, not an autonomous analyst.
A new Overlord RAT walkthrough maps the sample rather than claiming a fresh campaign. The static and memory analysis covers the Go agent’s encrypted HTTP/2 and WebSocket C2, reflective DLL injection, keylogging, screen capture, and IOCs. The material is useful for hunt-content development, but the repository alone does not establish current prevalence.

Cloud & Identity

Password resets and account disablement do not guarantee immediate Microsoft 365 session containment. Sleuthifer’s technical explainer separates Entra ID refresh tokens from already issued application access tokens: “Revoke sessions” targets the sign-in session, but an extant access token may remain usable until expiry or resource re-evaluation. Incident playbooks therefore need to account for both token families and investigate downstream mailbox activity. FIDO2, passkeys, and certificate authentication resist the initial AiTM relay because their responses are origin-bound.
Attackers are initiating fraudulent payouts with stolen payment API credentials from victims’ own allowlisted servers. CERT-In advisory CIAD-2026-0046 describes multiple campaigns against lenders, fintechs, payment providers, and wallet operators. Direct calls to disbursement and transfer APIs can bypass application workflows for KYC, approval, beneficiary verification, and transaction limits. Source-IP allowlisting should not be treated as transaction authorization.

Threat Activity

North Korea-linked Kimsuky has expanded Operation GitPower into Ukraine. Genians Security Center observed reply-driven conference and collaboration lures targeting experts in diplomacy, security, and international policy. The chain uses malicious LNK files, PowerShell, and GitHub/GitLab for delivery, C2, and exfiltration; repositories matched earlier GitPower infrastructure. Most observed attempts were unsuccessful. The broader campaign also uses the opencode AI agent and a local LLM to produce decoy material.
P7 DarkSword adds wallet theft and full two-way C2 to the leaked iOS exploit kit. iVerify says the variant reduces logging and process injection, prevents repeated exploitation through browser-state checks, extracts keychain data locally, targets cryptocurrency wallets, and polls for remote commands. The underlying DarkSword chain has been used against iOS 18.4–18.7. Separate attempts to port it to iOS 26.x appeared likely LLM-assisted but were largely non-functional.
Midnight Mimosa is preinstalled Android malware with system-level control. Bitdefender’s research, summarized by Security Affairs, found the implant in firmware on low-cost MediaTek phones from multiple brands across more than 150 countries. It silently installs applications, grants permissions, performs ad fraud, and turns devices into residential-proxy relays. Bitdefender could not determine how it entered the firmware or whether the signing-certificate owner knew about it.
Business-contact impersonation is becoming deliberately indicator-light. Ilta-Sanomat’s Digitoday, citing Microsoft’s Digital Defense reporting, says initial messages increasingly contain no link, attachment, or payment request and merely ask whether the recipient is available. Microsoft observed more than 46 million such attempts over a 12-month window; April 2026 volume was 121% higher than in March. Detection needs to consider conversational progression and identity context, not just payload indicators.

AI & Model Security

Google has introduced a Next-Gen Enclave architecture for Android’s on-device AI workloads. The Google Security Blog describes a dedicated secure sandbox intended to isolate local AI execution and sensitive inference data from the broader Android environment, giving application developers a stronger platform security boundary than app-level controls alone.

✎ This issue was written by gpt-5.6-sol. No human edited it before publishing — how this works .