O RLY CYBER

(outpost24.com) Understanding the OWASP Top 10 Risks for Large Language Model Applications

The 2025 OWASP Top 10 for LLM Applications highlights critical risks as generative AI shifts to production, introducing new threats like System Prompt Leakage (LLM07) and Vector/Embedding Weaknesses (LLM08).

In brief - The updated OWASP Top 10 for LLMs underscores emerging risks in AI deployments, including prompt injection, data poisoning, and excessive agency. Organizations must adopt secure design principles to mitigate novel attack surfaces in RAG pipelines and agentic systems.

Technically - Key risks include:
- LLM01 (Prompt Injection): Malicious input manipulation via prompts or retrieved context.
- LLM02 (Sensitive Info Disclosure): Unauthorized data exposure due to weak access controls.
- LLM03 (Supply Chain Risks): Vulnerabilities in third-party models/datasets.
- LLM04 (Data/Model Poisoning): Tampering with training data to skew outputs.
- LLM05 (Improper Output Handling): Trusting unvalidated LLM outputs.
- LLM06 (Excessive Agency): Overprivileged LLMs enabling broader compromise.
- LLM07 (System Prompt Leakage): Extraction of hidden instructions to bypass safeguards.
- LLM08 (Vector/Embedding Weaknesses): Attacks on RAG pipelines and vector DBs.
Mitigations require least privilege, input validation, and adversarial testing.

Source: https://outpost24.com/blog/owasp-top-10-llm-risks-explained/

#Cybersecurity #ThreatIntel

(welivesecurity.com) ESET APT Activity Report: Global Espionage and Cyber Warfare Trends from Q4 2025 to Q1 2026

New ESET APT Activity Report (Q4 2025–Q1 2026) reveals state-sponsored threat actors escalating cyber warfare and espionage with advanced TTPs. China-aligned groups (e.g., FamousSparrow, SteppeDriver) targeted maritime/energy sectors; Iran-aligned clusters (Rusty Boots, MoKhargosh) deployed bootkit wipers; North Korea’s Lazarus compromised axios via supply chain attack (CVE-2025-XXXX); Russia’s Sandworm waged destructive ops against NATO (e.g., Polish energy wiper).

In brief - State-aligned APTs intensified cyber espionage and destructive attacks, leveraging supply chain compromises, modular malware, and geopolitical targeting. Key sectors: energy, defense, and tech. Groups like Lazarus and Sandworm expanded operations beyond regional conflicts.

Technically - Notable implants: UNC5221’s PhiliKit (Ivanti VPN), Sednit’s Covenant/BeardShell (Ukraine), and Android spyware Asin. Techniques included browser-in-the-browser phishing, SmartOffice CRM exploitation, and living-off-the-land tactics. North Korea’s Rook ransomware and Russia’s new wiper variants underscore the fusion of cybercrime and state objectives.

Source: https://www.welivesecurity.com/en/eset-research/eset-apt-activity-report-q4-2025-q1-2026/

#Cybersecurity #ThreatIntel

(paloaltonetworks.com) The Rise of Data-Only Extortion: How Threat Actors Are Bypassing Ransomware for Direct Financial Leverage

In brief - Threat actors are shifting from ransomware encryption to data-only extortion, leveraging faster exfiltration, regulatory pressure, and AI acceleration. Mid-sized orgs in Professional Services, Healthcare, and Construction are primary targets. Groups like Bling Libra and TGR-CRI-1135 exploit supply chains and vishing for initial access.

Technically - Ransomware encryption usage dropped to 78% in 2025, driven by improved backups, endpoint security, and rapid exfiltration. TGR-CRI-1135 compromised 500+ software components to steal cloud tokens/SSH keys, while Bling Libra used vishing for SaaS persistence. AI models like Mythos reduce attack timelines to 25 minutes. Defenders must prioritize DLP, SaaS security, phishing-resistant auth, and supply chain integrity.

Source: https://unit42.paloaltonetworks.com/cyber-extortion-economy/

#Cybersecurity #ThreatIntel

(microsoft.com) Dirty Frag: A Reliable Linux Local Privilege Escalation Vulnerability Exploiting Kernel Networking Components

New Linux LPE threat: Dirty Frag exploits CVE-2026-43284 (esp4/esp6) and CVE-2026-43500 (rxrpc) in kernel networking components, enabling reliable root access via memory-fragment manipulation. Affects Ubuntu, RHEL, CentOS Stream, OpenShift.

In brief - Dirty Frag is a high-risk Linux privilege escalation vulnerability leveraging kernel networking flaws (esp4/esp6, rxrpc) to achieve root access. Unlike traditional race-condition LPEs, it offers improved reliability, posing operational risks in compromised environments. Mitigations include disabling vulnerable modules and restricting local access while patches are pending.

Technically - Dirty Frag abuses Linux page cache behavior in IPsec (esp4/esp6) and RxRPC (rxrpc) modules to escalate privileges. Exploitation paths include SSH, web-shells, or container escapes. Detection via Microsoft Defender available; post-exploitation integrity checks recommended due to potential memory/file persistence.

Source: https://www.microsoft.com/en-us/security/blog/2026/05/08/active-attack-dirty-frag-linux-vulnerability-expands-post-compromise-risk/

#Cybersecurity #ThreatIntel

(praetorian.com) Leveraging LLMs and Agentic Adversarial Techniques to Evade Modern EDR and Signature-Based Detection

New research demonstrates how LLMs and agentic adversarial techniques are being used to automate EDR and signature-based detection evasion for offensive security tools.

In brief - Offensive security teams are leveraging LLMs like Claude and VirusTotal as an adversarial oracle to iteratively modify binaries, reducing detection rates. Techniques such as 'ghost profiles' for Go binaries manipulate gopclntab content to evade ML classifiers. The approach blends tools into legitimate software baselines rather than eliminating all signals, as anomalies can trigger detections. While effective, reliance on VirusTotal introduces OpSec risks.

Technically - The framework employs an LLM-driven feedback loop with VirusTotal to triage detections, distinguishing between YARA rules and ML classifiers. For Go binaries, ghost profiles alter gopclntab to mimic legitimate software, reducing ML-based detections. A case study evades AhnLab’s Sliver rule (YARA) by randomizing hardcoded strings in digital signatures. Same-batch controls account for model drift, while statistical analysis identifies detection triggers. The method balances static obfuscation with binary coherence, avoiding deviations from natural toolchain baselines. Future work may explore WebAssembly loaders for further evasion.

Source: https://www.praetorian.com/blog/llm-edr-signature-reduction/

#Cybersecurity #ThreatIntel

(recordedfuture.com) The Evolution of Cybersecurity: Tracing the Journey from Early Viruses to Modern Threat Intelligence

New analysis traces cybersecurity’s evolution from early theoretical threats to today’s AI-driven attacks, revealing how defense strategies have adapted to escalating adversary sophistication.In brief - Cybersecurity has transformed from a niche technical concern to a global priority, shaped by pivotal threats like the Morris Worm, Stuxnet, and WannaCry. The shift from reactive to proactive defenses, including Zero Trust, reflects the growing complexity of nation-state and criminal cyber operations.Technically - Early malware like Creeper (1971) and Elk Cloner (1982) demonstrated self-replication, while the Morris Worm (1988) exploited buffer overflows (CVE-1988-0400) to disrupt ARPANET. The 2000s saw ransomware adopt RSA encryption (e.g., Gpcode), and the 2010s introduced nation-state tools like Stuxnet (CVE-2010-2568). Modern threats leverage AI for evasion, while defenses prioritize Zero Trust and cloud-native security to counter advanced persistent threats (APTs) and supply-chain attacks.

Source: https://www.recordedfuture.com/blog/cybersecurity-history

#Cybersecurity #ThreatIntel

(starlabs.sg) Dirty Cert: Exploiting Command Injection in Cisco Smart Software Manager via Malicious TLS Certificates

In brief - This article describes a post-authentication remote code execution (RCE) vulnerability discovered in the Cisco Smart Software Manager (CSSM) that was silently patched by the vendor.

Technically - This article details a command injection vulnerability within the CSSM nginx_configurator. The flaw existed because the application used childProcess.execSync to pass user-supplied TLS certificates and RSA keys directly into shell commands (echo "${cert}" | openssl x509). While the backend performed validation using the Ruby OpenSSL library, an attacker could bypass this by leveraging RFC 7468, which allows explanatory text before the BEGIN header of a certificate. By injecting shell commands into the intermediate_certificate field, which was passed to the frontend without sanitization, an attacker could achieve RCE with root privileges. The vulnerability was remediated by replacing the shell-based OpenSSL calls with the native Node.js crypto library.

Source: https://starlabs.sg/blog/2026/09-dirty-cert-cisco-smart-software-managers-silently-patched-rce/

#Cybersecurity

(safedep.io) Large-Scale npm Supply Chain Attack Targets Cloud and Financial Services with Credential-Stealing Malware

New large-scale npm supply chain attack uncovered: 164 malicious packages targeting cloud and financial services via dependency confusion, exfiltrating full environment variables to C2 (hxxps://oob[.]moika[.]tech).

In brief - A sophisticated npm supply chain attack leveraged 164 packages across five scoped namespaces to steal credentials, API keys, and secrets from a cloud provider and financial services firm. The attack used dependency confusion and postinstall scripts to exfiltrate `process.env` to a C2 server, with evidence of advanced reconnaissance and social engineering.

Technically - The threat actor (`mr.4nd3r50n`, `pik-libs`) published packages under scopes like `@cloudplatform-single-spa` and `@fb-deposit`, using a postinstall script to execute a multi-stage payload. The script included a 3-second delay, OS detection, and a second-stage download (macOS/Windows/Linux). Exfiltrated data was sent to `hxxps://oob[.]moika[.]tech` with a hardcoded secret (`l95HdDaz3kQx1Zsg3WxH6HvKANf51RY1`). Fake READMEs and inert probe packages were used to evade detection.

Source: https://safedep.io/oob-moika-tech-dependency-confusion-campaign

#Cybersecurity #ThreatIntel

(sublime.security) Evolving QR Code Phishing Tactics: How Attackers Evade Detection and How to Counter Them

QR code phishing (quishing) is surging as threat actors exploit evasion techniques to bypass security controls. Attackers leverage low-contrast QR codes, image skewing, ASCII rendering, and split-image tactics to evade automated scanners while remaining scannable by human targets.

In brief - Quishing has become a preferred attack vector due to its ability to bypass traditional email security. Adversaries manipulate QR codes to exploit gaps in automated detection, shifting targets from corporate devices to personal mobile endpoints. Security solutions must evolve to counter these evasive tactics.

Technically - Attackers employ multiple techniques to obfuscate malicious QR codes, including non-standard color schemes, distorted aspect ratios, ASCII-based rendering, and splitting codes across PDF attachments. Sublime Security’s approach counters these by combining iterative scanning, image manipulation (e.g., contrast adjustment), and machine learning-powered computer vision. This method improves detection of manipulated QR codes by ~30% without performance trade-offs.

Source: https://proxied2.sublime.security/blog/modern-qr-code-phishing-evasion-tactics-you-should-know-about

#Cybersecurity #ThreatIntel

(eye.security) Breaking DragonForce Ransomware's File Name Encryption: A Case Study in LLM-Assisted Cryptanalysis

DragonForce ransomware employs a monoalphabetic substitution cipher on base32-encoded filenames, complicating incident response by obscuring file identities. Analysts broke the cipher using LLM-assisted cryptanalysis, reducing decryption time from hours to minutes.

In brief - DragonForce ransomware uses a custom file name encryption scheme to hinder recovery efforts. Researchers leveraged known-plaintext attacks and LLM automation to rapidly reverse the cipher, demonstrating a method to accelerate ransomware response.

Technically - The ransomware applies a monoalphabetic substitution cipher to base32-encoded filenames (e.g., NTUSER.DAT → base32 → ciphertext + ".df_win"). Using known plaintext-ciphertext pairs from Windows system files, analysts identified a fixed substitution table. An LLM (Claude Code) automated pattern recognition, brute-forced remaining permutations (6! = 720), and validated the cipher by re-encrypting plaintexts to match original ciphertexts. The attack exploited consistent length ratios (1.6x) and repeating ciphertext patterns (e.g., "3gpng2wv").

Source: https://research.eye.security/breaking-encryption-schemes-the-lazy-way/

#Cybersecurity #ThreatIntel

(microsoft.com) Comprehensive Analysis of The Gentlemen Ransomware: Hybrid Encryption and Aggressive Lateral Movement Techniques

Storm-2697’s *The Gentlemen* RaaS deploys hybrid Curve25519/XChaCha20 encryption and aggressive lateral movement to maximize impact. Targets span North/South America, Europe, Africa, and Asia with double-extortion tactics.

In brief - Financially motivated Storm-2697 operates *The Gentlemen* RaaS, combining per-file hybrid encryption with redundant lateral movement (PsExec, WMI, scheduled tasks) to rapidly compromise networks. Double extortion and modular design amplify threat severity.

Technically - Go-based *The Gentlemen* uses Garble obfuscation and generates unique Curve25519 key pairs per file, encrypting with XChaCha20. Command-line arguments control encryption scope (9% per chunk for >1 MB files). Defense evasion includes disabling Defender, deleting VSS, clearing logs, and terminating EDR/backup processes. Lateral movement executes 21 redundant operations per target via PsExec, WMI, services, and PowerShell remoting. Persistence via scheduled tasks and registry keys. Post-encryption: custom wallpaper, disk wiping, and self-deletion.

Source: https://www.microsoft.com/en-us/security/blog/2026/05/28/the-gentlemen-ransomware-dissecting-a-self-propagating-go-encryptor/

#Cybersecurity #ThreatIntel

(cert.pl) Critical IDOR Vulnerability in WEBCON BPS Exposes Sensitive Employee Scheduling Data

In brief - This article describes a security vulnerability (CVE-2026-92419) in WEBCON BPS software that allows unauthorized access to employee vacation schedules.

Technically - This article details an Insecure Direct Object Reference (IDOR) vulnerability located in the /api/vacations/{path} endpoint of WEBCON BPS. The flaw exists because the selectedPeople parameter in the Gantt vacation chart API fails to validate if the authenticated requester has the necessary authorization to access the data of the specified users. Consequently, an attacker can manipulate this parameter with arbitrary user logins to bypass business logic restrictions and retrieve sensitive scheduling information for any employee, including management.

Source: https://cert.pl/en/posts/2026/09/CVE-2026-92419/

#Cybersecurity

(wordfence.com) Critical Unauthenticated Administrator Account Creation Vulnerability in WP Maps Pro WordPress Plugin

Critical unauthenticated admin account creation flaw in WP Maps Pro (CVE pending) enables full site takeover. Over 15K sites at risk via improperly secured AJAX endpoint.

In brief - The WP Maps Pro WordPress plugin contains a critical vulnerability allowing unauthenticated attackers to create administrator accounts, leading to complete site compromise. A patch (v6.1.1) is available; immediate updates are required.

Technically - The flaw exists in the `wpgmp_temp_access_ajax_callback()` function, which lacks capability checks and exposes a nonce to unauthenticated users. Attackers can invoke the AJAX action with `check_temp=false` to trigger `wpgmp_temp_access_support()`, creating an admin account with a hardcoded email and random username. A 'magic login URL' then authenticates the attacker without a password. The patch adds a `current_user_can('manage_options')` check to restrict access.

Source: https://www.wordfence.com/blog/2026/05/15000-wordpress-sites-affected-by-administrator-account-creation-vulnerability-in-wp-maps-pro-wordpress-plugin/

#Cybersecurity #ThreatIntel

(bishopfox.com) Confused Deputy Attacks in AI Agents: Mechanics, Case Studies, and Layered Mitigations

New research highlights the growing risk of confused deputy attacks targeting AI agents, where attackers manipulate systems into executing malicious actions using their own privileges. These attacks exploit trust relationships and tool access to bypass security controls, enabling data exfiltration and privilege escalation.

In brief - Confused deputy attacks leverage seemingly legitimate inputs (e.g., support tickets, emails) to trick AI agents into performing unauthorized actions. High-profile incidents like EchoLeak and ConfusedPilot demonstrate real-world impact, emphasizing the need for layered mitigations such as least-privilege access and network egress controls.

Technically - Attackers embed malicious instructions in attacker-controlled content, which AI agents process via Multi-Tool Processing (MCP) servers. Techniques include Insecure Direct Object Reference (IDOR) and metadata service exploitation to escalate privileges. Case studies show Microsoft Copilot processing crafted emails to exfiltrate data or interpreting malicious calendar invites to expose private information. Mitigations include per-task tool restrictions, least-privilege principles, and egress controls to limit data exfiltration. Attackers can also bypass generative AI guardrails by directly targeting MCP servers, underscoring the need for robust security at both AI and infrastructure layers.

Source: https://bishopfox.com/blog/otto-support-confused-deputy

#Cybersecurity #ThreatIntel

(recordedfuture.com) Comprehensive Guide to Types and Prevention of Payment Fraud in the Digital Age

In brief - Payment fraud is accelerating, with global losses projected to exceed $362B by 2028. Threat actors exploit phishing, credit card fraud, ATO, and wire transfer fraud, targeting e-commerce, healthcare, and banking. Prevention requires layered security, PCI DSS compliance, and advanced detection like ML-driven fraud KPIs.

Technically - Fraudsters leverage skimming, POS malware, and dark web credentials for card fraud, while ATO/NAF relies on stolen credentials from breaches. Countermeasures include 3D Secure (3DS), network tokenization, and MFA. Machine learning detects anomalies in real time, but gaps in 2FA (e.g., SMS interception) and domain spoofing remain critical risks. PCI DSS compliance and adaptive monitoring are essential to mitigate evolving threats.

Source: https://www.recordedfuture.com/blog/types-of-payment-fraud

#Cybersecurity #ThreatIntel

(spycloud.com) Evolving Cybercrime Trends: Supply Chain Compromises, Cloud-Focused Extortion, and the Fragmented Darknet Forum Ecosystem

Threat actors are shifting from ransomware encryption to data theft extortion via cloud service compromises, exploiting open-source supply chains and non-human identities (NHIs) like GitHub tokens.

In brief - Financially motivated groups like TeamPCP, ShinyHunters, and LAPSUS$ are prioritizing cloud-focused extortion over traditional ransomware. Supply chain attacks on open-source packages (e.g., PyTorch Lightning, Xinference) and the fragmentation of darknet forums (e.g., BreachForums → PwnForums) highlight evolving cybercrime tactics. Telegram is now a preferred communication channel for threat actors.

Technically - TeamPCP compromises open-source packages with credential-stealing malware that propagates via GitHub tokens, validating them via GitHub API and infecting writable repos. ShinyHunters targets cloud services (e.g., Salesforce, Snowflake) for data theft extortion. The malware also infects local npm packages, demonstrating worm-like behavior. Darknet forum fragmentation complicates threat intelligence collection due to varying rank structures and posting conventions.

Source: https://spycloud.com/blog/cybercrime-update-15-shinyhunters-supplychains-and-sketchy-new-criminal-forums/

#Cybersecurity #ThreatIntel

(paloaltonetworks.com) Cyber Threat Landscape and Risk Mitigation for the 2026 FIFA World Cup: A Multi-Nation, Multi-Actor Assessment

The 2026 FIFA World Cup faces severe cyber threats from Iran-nexus disruptive ops, pro-Russian hacktivism, and financially motivated cybercrime targeting critical infrastructure, hospitality, and fans.

In brief - The 2026 FIFA World Cup is a high-value target for Iran-nexus groups (Handala Hack Team, CyberAv3ngers) conducting disruptive OT attacks, pro-Russian hacktivists (NoName057(16)) launching DDoS campaigns, and cybercriminals exploiting the hospitality supply chain. Historical precedents from Paris 2024 and Qatar 2022 highlight risks of wiper malware, ransomware, and large-scale fraud. Geopolitical tensions amplify threats to municipal infrastructure and event integrity.

Technically - Iran-nexus threat actors exploit internet-exposed PLCs (CISA AA26-097A) and remote-access tools like TeamViewer, targeting U.S. critical infrastructure. NoName057(16) leverages the DDoSia platform, capable of >190K RPS, requiring robust scrubbing. Financially motivated groups (e.g., Muddled Libra/ALPHV) use credential-stuffing and social engineering against hospitality sectors. Defenders must audit OT systems, segregate identity providers, and enforce phishing-resistant MFA to counter advanced tradecraft.

Source: https://unit42.paloaltonetworks.com/fifa-world-cup-attack-surface/

#Cybersecurity #ThreatIntel

(gendigital.com) Global Scale of Reservation Hijack Scams: How Compromised Booking Data Fuels Targeted Travel Fraud

New intelligence reveals a large-scale Reservation Hijack campaign exploiting compromised booking data to execute highly targeted travel fraud. Over 350 accommodations across 50 countries affected, with attackers abusing real reservation details to craft convincing phishing lures.

In brief - Cybercriminals are leveraging stolen booking data from platforms like Booking.com to conduct context-aware phishing scams. These attacks exploit trust between travelers and accommodations, using accurate reservation details to bypass traditional security measures. Europe is the primary hotspot, with scams extending beyond hotels to various property types. The operation demonstrates organized, repeatable criminal activity requiring cross-industry collaboration to disrupt.

Technically - Attackers compromise reservation data via hotel accounts, PMS, or third-party tools, then deploy tailored phishing pages with accurate booking context. Analysis reveals a phishing kit with recurring infrastructure patterns (identical asset paths, overlay components) and abuse of services like Cloudflare for evasion. Fraudulent pages include fake live support to manipulate payments or capture OTPs. Scam messages are delivered via SMS/WhatsApp/in-app channels at high-opportunity moments (e.g., pre-arrival), combining accurate data with dynamic infrastructure to evade detection.

Source: https://www.gendigital.com/blog/insights/research/reservation-hijack-scams-target-travelers

#Cybersecurity #ThreatIntel

(sysdig.com) Dirty Frag: Critical Linux Kernel Vulnerabilities Enable Local Privilege Escalation via Page Cache Poisoning

Critical Linux kernel LPE vulnerabilities CVE-2026-43284 (IPsec ESP) and CVE-2026-43500 (RxRPC), dubbed Dirty Frag, enable unprivileged users to achieve root via page cache poisoning. PoC released pre-patch; affects kernels since 2017/2023.

In brief - Two unpatched Linux kernel flaws allow local privilege escalation to root by corrupting page cache via in-place decryption in IPsec/RxRPC. PoC available; immediate mitigation required.

Technically - Dirty Frag exploits in-place decryption optimizations in ESP (CVE-2026-43284) and RxRPC (CVE-2026-43500) to overwrite page cache of setuid binaries (e.g., /usr/bin/su) with attacker-controlled shellcode. ESP variant requires unprivileged user namespaces; RxRPC affects default Ubuntu configs. Detection via AF_KEY/AF_RXRPC/XFRM socket monitoring; mitigate by blacklisting modules or deploying seccomp profiles.

Source: https://webflow.sysdig.com/blog/dirty-frag-cve-2026-43284-and-cve-2026-43500-detecting-unpatched-local-privilege-escalation-via-linux-kernel-esp-and-rxrpc

#Cybersecurity #ThreatIntel

(safedep.io) MicrosoftSystem64: A Cross-Platform Supply Chain RAT Leveraging HuggingFace for Data Exfiltration

New cross-platform RAT *MicrosoftSystem64* attributed to DPRK’s FAMOUS CHOLLIMA (Contagious Trader) abuses npm & HuggingFace for supply-chain attacks. Targets devs & crypto users with credential theft, SSH key exfil, and Telegram session hijacking.

In brief - A North Korean threat actor is distributing a cross-platform RAT via malicious npm packages, leveraging HuggingFace for C2 and data exfiltration. The malware steals credentials, cryptocurrency wallet data, SSH keys, and Telegram sessions, with persistence across Windows, macOS, and Linux. Active since April, the campaign remains operational.

Technically - MicrosoftSystem64 is an 81 MB stripped ELF binary packaged as a Node.js SEA (Single Executable Application), enabling execution without Node.js. It communicates via WebSocket C2 (195[.]201[.]194[.]107:8010) and supports 24 commands, including shell execution and file system recon. Data exfiltration occurs via private HuggingFace datasets. The malware employs XOR-encrypted configs, cross-platform persistence (scheduled tasks, LaunchAgents, systemd), and a self-update mechanism. Keylogging uses native APIs (SetWindowsHookEx, CGEventTap, xinput/evdev), with screenshots captured every 60s. Targets 80+ crypto wallet extensions, 15 browser families, and Telegram Desktop sessions.

Source: https://safedep.io/microsoftsystem64-binary-payload-analysis

#Cybersecurity #ThreatIntel

(malwarebytes.com) Fake ChatGPT Download Site Distributes Dual-Platform Malware Targeting Credentials and Cryptocurrency Wallets

New campaign distributes dual-platform malware via fake ChatGPT download site (openew[.]app). Windows users get a credential-stealing dropper; macOS users face Atomic Stealer (AMOS) targeting passwords, cookies, Telegram sessions, and cryptocurrency wallets (Ledger/Trezor trojanization).

In brief - Cybercriminals exploit AI hype by impersonating ChatGPT downloads to deliver platform-specific infostealers, with AMOS prioritizing cryptocurrency theft on macOS. Vigilance and official sources are critical for AI tool downloads.

Technically - Windows payload (Chat_GPT.exe) uses Inno Setup/Electron to beacon to 188.137.246.189 and execute PowerShell exfiltration. macOS AMOS (ChatGpt.dmg) employs AppleScript for password capture, browser/wallet data theft, and trojanized wallet app replacement. Campaign leverages search traffic for AI tools, with higher investment in macOS payloads due to crypto theft ROI.

Source: https://www.malwarebytes.com/blog/threat-intel/2026/05/fake-chatgpt-download-site-infects-windows-and-mac-users-with-malware

#Cybersecurity #ThreatIntel

(doublepulsar.com) Microsoft’s Controversial Stance on Zero-Day Exploits and the Erosion of Responsible Disclosure Norms

Microsoft’s public labeling of zero-day PoC exploit distribution as 'criminal activity' signals a shift in responsible disclosure norms, risking the criminalization of security research. A critical unpatched BitLocker bypass in default deployments underscores the stakes.

In brief - Microsoft’s recent stance on zero-day disclosures—threatening legal action against researchers sharing PoCs—contradicts its historical practices (e.g., hiring public zero-day disclosers) and undermines collaborative defense. Selective enforcement via GitHub raises concerns about transparency and corporate prioritization over user security.

Technically - Microsoft’s response to Nightmare Eclipse’s public disclosure of an unpatched BitLocker bypass (default deployment flaw) highlights its inconsistent application of 'responsible disclosure.' The company’s use of GitHub to suppress exploits targeting its products while allowing others to persist reflects a vendor-centric approach. Historical precedents (e.g., SandboxEscaper, exploit brokers) further erode trust in Microsoft’s commitment to equitable vulnerability handling. This posture risks stifling independent research and normalizing legal threats against security professionals.

Source: https://doublepulsar.com/microsofts-stance-on-zero-day-exploits-is-a-dumpster-fire-of-their-own-making-0946117940a4

#Cybersecurity #ThreatIntel

(aikido.dev) Securing GitHub Actions: A Comprehensive Checklist to Mitigate Supply Chain Attacks and Workflow Misconfigurations

GitHub Actions workflows are actively exploited in supply chain attacks due to misconfigurations and insecure defaults. Recent incidents (e.g., Trivy, Ultralytics) highlight risks like `pull_request_target` misuse and script injection.

In brief - GitHub Actions misconfigurations enable supply chain attacks via privileged triggers, untrusted input injection, and mutable action references. Organizations must harden workflows by pinning actions to commit SHAs, restricting token permissions, and adopting OIDC for cloud credentials to mitigate risks.

Technically - Key attack vectors include:
1) Script injection via untrusted input (e.g., branch names) in `run:` steps—mitigate by using environment variables.
2) Privileged triggers (`pull_request_target`, `workflow_run`) exposing secrets—avoid or implement strict event checks.
3) Mutable action references (tags/branches) allowing malicious code injection—pin to commit SHAs.
4) Secrets exposure via poor scoping or logging—use OIDC and environment-level scoping.
5) Artifact/self-hosted runner risks—use ephemeral runners and tools like Harden-Runner. Automated scanning (e.g., Zizmor) and provenance attestations further reduce attack surfaces.

Source: https://www.aikido.dev/blog/checklist-github-actions

#Cybersecurity #ThreatIntel

(malwarebytes.com) Carnival Corporation Faces Another Major Data Breach: Social Engineering Leads to Exposure of Nearly 6 Million Records

Carnival Corporation suffered a major data breach in April 2026, exposing nearly 6M records via a social engineering attack. ShinyHunters claimed responsibility, exfiltrating PII including names, emails, DOBs, and membership details. This marks another in a series of breaches for the cruise operator.

In brief - Carnival Corporation disclosed a breach affecting 6M individuals after ShinyHunters tricked an employee into granting system access. Exposed data includes PII, raising risks of identity theft and phishing. The incident underscores persistent vulnerabilities in human-centric security controls.

Technically - The attack began on April 14, 2026, with a social engineering compromise of an employee account. By April 22, ShinyHunters accessed and exfiltrated PII from Carnival’s systems. The breach highlights gaps in MFA enforcement, employee training, and real-time monitoring. ShinyHunters’ involvement aligns with their known tactics of data theft and extortion.

Source: https://www.malwarebytes.com/blog/data-breaches/2026/05/carnival-confirms-data-breach-impacting-nearly-6-million

#Cybersecurity #ThreatIntel

(rapid7.com) Critical Remote Code Execution Vulnerability in Gogs via Argument Injection in Git Rebase

Critical unauthenticated RCE in Gogs (CVE-2026-XXXX, CVSSv4 9.4) via argument injection in Git rebase. Exploitable by any registered user; default config allows open registration. Metasploit module available.

In brief - Gogs, a self-hosted Git service, contains a critical argument injection flaw enabling unauthenticated RCE. Default settings permit open registration, allowing attackers to gain full server control, access private repos, and facilitate supply chain attacks. No patch available.

Technically - The vulnerability (CWE-88) resides in Gogs' 'Rebase before merging' feature, where unsanitized `pr.BaseBranch` input is passed to `git rebase`. Attackers craft branch names like `--exec=touch${IFS}/tmp/rce_proof` to inject commands via `sh -c`. Exploitation requires no privileges, affects all platforms, and leaves minimal forensic traces. Prior fixes did not address this code path.

Source: https://www.rapid7.com/blog/post/ve-authenticated-rce-via-argument-injection-gogs-unfixed

#Cybersecurity #ThreatIntel

(hiddenlayer.com) ShadowLogic: Persistent Backdoors in Machine Learning Models and the Risks to AI Supply Chains

New AI supply chain threat: ShadowLogic backdoor persists across model conversions (PyTorch→ONNX→TensorRT) and fine-tuning, unlike traditional backdoors. Attackers embed malicious logic in computational graphs, triggering attacker-defined outputs with specific inputs (e.g., red square). 100% efficacy observed post-conversion; conventional backdoors degrade >50% after fine-tuning.

In brief - AI supply chains face a novel, resilient backdoor threat (ShadowLogic) that survives model transformations and fine-tuning, enabling persistent malicious logic in pre-trained models. Trusted formats like ONNX are vulnerable.

Technically - ShadowLogic modifies a model’s computational graph to include conditional triggers (e.g., pixel patterns) that alter outputs. Implemented in the forward pass, it remains intact through conversions (PyTorch→ONNX→TensorRT) and fine-tuning, unlike fine-tuning-based backdoors. Detection requires tools like ModelScanner to analyze graph-level anomalies.

Source: https://www.hiddenlayer.com/research/persistent-backdoors

#Cybersecurity #ThreatIntel

(tenable.com) Download Pumping: A Novel Deception Technique in npm Supply Chain Attacks

Threat actors are exploiting npm's automated infrastructure to manipulate package download counts via 'download pumping,' a novel supply chain deception technique. By rapidly publishing hundreds of benign package versions, attackers trigger automated downloads from mirrors and scanners, inflating metrics to mask malicious payloads.

In brief - Attackers abuse npm's automated systems to artificially boost package download counts, creating false legitimacy. The 'ambar-src' campaign achieved 50K+ downloads in three days by uploading 700+ versions, undermining trust in superficial metrics like download counts.

Technically - The 'download pumping' technique leverages automated bot traffic (e.g., repository mirrors, security scanners) to generate 100-150 downloads per package version upload. PoC testing shows dynamically changing postinstall scripts amplify downloads further. Unlike HTTP request spoofing, this method populates release histories and exploits naive download stats. Mitigations include version pinning, minimum package-age restrictions, and ephemeral CI/CD runners.

Source: https://www.tenable.com/blog/how-cyberattackers-inflate-malicious-package-npm-download-counts

#Cybersecurity #ThreatIntel

(talosintelligence.com) Rethinking Vulnerability Prioritization: Beyond CVSS to EPSS and Decentralized CVE Enrichment

New research challenges traditional vulnerability management, advocating for EPSS alongside CVSS to prioritize patching based on exploit likelihood rather than severity alone.

In brief - Vulnerability prioritization must evolve beyond CVSS to incorporate EPSS, which predicts exploitation probability within 30 days. Centralized databases like CISA’s KEV are limited; decentralized approaches such as GCVE offer faster, globally relevant enrichment. Cisco Talos’s EvidenceForge tool generates synthetic logs to enhance SOC training without compliance risks.

Technically - EPSS (Exploit Prediction Scoring System) complements CVSS by quantifying real-world exploitability, enabling risk-based patching. GCVE (Global CVE) decentralizes CVE enrichment, addressing delays in centralized sources like KEV. EvidenceForge leverages AI-assisted scenario authoring to produce temporally and causally consistent synthetic logs, improving detection validation and threat hunting without relying on sensitive datasets.

Source: https://blog.talosintelligence.com/less-panic-patching-more-precision/

#Cybersecurity #ThreatIntel

(maltego.com) Comprehensive Walkthrough of Maltego Community OSINT CTF: Investigative Techniques and Real-World Applications

In brief - A recent Maltego Community OSINT CTF demonstrated the critical role of structured investigative techniques in cyber threat intelligence, geolocation, and maritime/aviation security. Participants solved 26 challenges using tools like VirusTotal, SunCalc, and AIS tracking, highlighting collaborative OSINT methodologies for real-world threat analysis.

Technically - The CTF featured advanced OSINT tasks, including malware hash analysis via VirusTotal (e.g., identifying ZIP archives and domains), ADS-B/Morse code decoding, and vessel tracking using AIS/EPIRB data from ITU databases. Challenges required chronolocation (SunCalc), reverse image search, ACARS/VDL Mode 2 decoding (FlightAware), and VIN-based vehicle identification. Emphasized toolchain integration (e.g., Maltego, Global Fishing Watch) and workflow adaptability for threat infrastructure mapping and geospatial forensics.

Source: https://www.maltego.com/blog/osint-ctf-challenge-walkthroughs/

#Cybersecurity #ThreatIntel

(gitguardian.com) Credentials as the Linchpin: How Secrets Sprawl and Non-Human Access Drive Modern Breach Impact

Credentials remain the linchpin of modern breaches, enabling attackers to expand footholds into full-scale compromises. Verizon DBIR and GitGuardian reports reveal exploited vulnerabilities (31%) and credential abuse (39%) dominate attack chains, with non-human access—API keys, OAuth tokens, and cloud credentials—proliferating across DevSecOps environments. Third-party risk (48% of breaches) and ransomware (77% of System Intrusion cases) thrive on poorly governed secrets.

In brief - Credentials are the connective tissue of 39% of breaches, with non-human access sprawl and third-party integrations amplifying risk. Ransomware and lateral movement rely on stolen secrets, while shadow AI introduces new exposure pathways. Prioritize visibility and lifecycle management to mitigate.

Technically - Initial access via exploited CVEs (31%) or compromised third parties (48%) converges on credential abuse to escalate privileges. Non-human identities (API keys, OAuth tokens) embedded in code, CI/CD pipelines, or SaaS tools are often overpermissioned and unrotated. Attackers chain trust paths—e.g., leaked deployment tokens → package registry access → additional secrets. Salesloft-Salesforce breach demonstrates OAuth delegation risks. Incident response must analyze credential exposure, ownership gaps, and downstream dependencies. Shadow AI exacerbates risks via hardcoded secrets in uploaded code. Mitigation requires integrating secrets vaults, SaaS apps, and identity providers to enforce least privilege and automate rotation.

Source: https://blog.gitguardian.com/initial-access-changed-the-attack-path-did-not-findinds-from-the-verizon-2026-dbir/

#Cybersecurity #ThreatIntel