The DOJ indicted 54 individuals linked to Venezuelan gang Tren de Aragua for deploying Ploutus malware in a multi-million dollar ATM jackpotting operation involving 1,529 incidents since 2021, resulting in $40.73 million stolen to fund terrorist activities. Attackers conducted physical reconnaissance, opened ATM hoods to test alarm responses, then installed Ploutus by replacing hard drives or connecting USB drives to issue unauthorized commands to Cash Dispensing Modules while deleting forensic evidence. The malware targets Windows-based ATMs (particularly legacy XP systems and Diebold models) and can be activated via a physical keyboard with activation codes. Financial institutions should implement physical tamper detection on ATM enclosures, upgrade legacy Windows XP systems, monitor Cash Dispensing Module command patterns for anomalies, implement hard drive integrity checks and USB port access controls, and deploy enhanced physical security measures, including alarm response verification protocols.
Hacktivists linked to Anna's Archive scraped around 86 million Spotify tracks, claiming to safeguard musical heritage from catastrophes and platform risk, yet only about a third of Spotify's total catalog is actually preserved as audio. Spotify calls the operation piracy. It shut down the scraping accounts and plans to strengthen safeguards. Anna's Archive has hinted at possible future single‑track downloads.
South Korea plans to require facial recognition when people sign up for new mobile numbers, matching ID photos with real-time images to block phones registered under false identities. The rule, which applies to major carriers and virtual operators, starts on March 23. Authorities hope to curb widespread voice phishing scams and address security weaknesses exposed by a massive SK Telecom data breach that affected 27 million SIM subscribers.
The FBI indicted 29-year-old Zahid Hasan from Dhaka, Bangladesh, and seized three domains.
Interpol's Operation Sentinel resulted in 574 arrests across 19 African countries, recovered $3 million, and decrypted six ransomware variants.
Pornhub warned Premium members to expect sextortion emails after a November 8 data breach at third-party analytics provider Mixpanel exposed user information.
CountLoader 3.2 and GachiLoader malware are spreading via cracked software sites and 100+ compromised YouTube videos (220K views from 39 accounts), deploying ACR Stealer and Rhadamanthys through multi-stage attacks. CountLoader uses trojanized Python interpreters that execute via mshta.exe, create persistence with fake Google scheduled tasks that run every 30 minutes, detect CrowdStrike Falcon to adapt its execution, and spread via USB drives. GachiLoader employs Node.js with novel Vectored Exception Handling PE injection techniques and kills SecHealthUI.exe to disable Windows Defender. Security teams should monitor for mshta.exe abuse, suspicious scheduled tasks that mimic legitimate software, WMI queries for AV detection, and modifications to system folders that exclude Defender, and implement application whitelisting to block execution of unauthorized Python interpreters and Node.js scripts.
Researchers from Riot Games discovered a UEFI flaw that could allow for Direct Memory Access (DMA) capable devices to load before the operating system. DMA allows for PCIe devices to read and write RAM without going through the CPU and relies upon IOMMU for protection. A physically present attacker would need to attach a rogue PCIe device before boot to exploit this vulnerability.
WatchGuard is warning users to patch a critical, actively exploited remote code execution vulnerability in its Firebox firewalls. The vulnerability is caused by an out-of-bounds write that allows unauthenticated attackers to execute remote code. Firebox firewalls are only vulnerable if they are configured to use IKEv2, or in some circumstances, even if the configuration was deleted.
A security researcher documented AI-assisted firmware analysis of TP-Link Tapo C200 cameras using tools including Grok, GhidraMCP, and Claude Sonnet 4. They discovered hardcoded SSL private keys that enabled HTTPS traffic decryption, pre-authentication memory overflows, integer overflow crashes, and unauthenticated WiFi hijacking, affecting approximately 25,000 internet-exposed devices. The vulnerabilities enabled remote attackers to decrypt camera traffic, crash devices, force connections to malicious networks, and enumerate nearby WiFi BSSIDs for precise geolocation via Apple's location services API. TP-Link delayed patching beyond its 90-day commitment despite being a CVE Numbering Authority, raising concerns about vendor conflicts of interest when controlling their own vulnerability disclosure pipeline while marketing low CVE counts as a competitive advantage.
Researchers from Fog Security discovered three methods to bypass AWS Trusted Advisor and expose S3 buckets without triggering alerts. The three methods are to create a bucket policy that denies s3:GetBucketPolicyStatus, s3:GetBucketPublicAccessBlock, and s3:GetBucketAcl. Fog Security coordinated disclosure with AWS, but noted that it took AWS two attempts to fix the vulnerability and felt that AWS's communication downplayed the severity.
When introducing agentic AI to its cybersecurity teams, Google began by building trust in generative AI by adding chat interfaces to existing tools. Google security then identified initial use cases for distillation and translation, focusing on bottlenecks that AI could alleviate. The team then established and monitored KPIs as they scaled their program.
Sponsored
In a dojo, agents train to avoid wasted motion, to anticipate, and to endure. Sumo Logic [10] brings dojo thinking to the SOC. Automation delivers repeatable discipline, machine learning detects anomalies, and it's all orchestrated through natural language. See how analysts
Mullvad's GotaTun is a Rust-based WireGuard implementation forked from Cloudflare's BoringTun that replaces wireguard-go, eliminating 85% of Android app crashes attributed to the Go implementation's FFI complexity and opaque runtime debugging challenges. The Android rollout in November reduced the user-perceived crash rate from 0.40% to 0.01%, with zero GotaTun-originated crashes, while improving connection speeds and battery efficiency through safe multithreading and zero-copy memory strategies. Mullvad plans a third-party security audit in early 2026 before expanding the GotaTun deployment to desktop and iOS platforms, maintaining full WireGuard protocol compatibility while supporting privacy features, including DAITA and Multihop.
Kanidm is a simple and secure identity management platform that provides a full identity provider, covering the broadest possible set of requirements and integrations.
The DedSec Project is a comprehensive cybersecurity toolkit designed for educational purposes. It provides 50+ powerful tools that cover everything from network security analysis to ethical hacking education. Everything is completely free and designed to help you shift from being a target to being a defender.
The Trump administration systematically dismantled federal cybersecurity infrastructure: CISA lost 1/3 of its workforce and faces $491M budget cuts, the Cyber Safety Review Board was dissolved mid-investigation of Chinese telecom intrusions, and both NSA and Cyber Command have operated without leadership since April. DOGE accessed sensitive federal databases from SSA, DHS, OPM, and Treasury by circumventing security controls and audit mechanisms, with data exfiltration coinciding with Russian login attempts using valid DOGE credentials. Security professionals should prepare for degraded federal threat intelligence sharing after the elimination of CIPAC and MS-ISAC funding, reduced public-private coordination as agencies canceled meetings with infrastructure operators, and increased foreign intelligence recruitment of laid-off federal employees with security clearances.
In addition to $901B in funding, the 2026 National Defense Authorization Act (NDAA) also includes a series of new cybersecurity policies. The NDAA includes a provision that secure phones provided to senior leaders meet a set of security requirements, such as data encryption. The act also includes increased mental health provisions and training initiatives around AI security.
UK ministers acknowledge a confirmed cyberattack on the Foreign Office, first reported by The Sun and widely linked in the media to Chinese state-backed hackers, though officials stress those claims are speculative. They say a technical flaw at one site was quickly closed and insist there is currently low risk to individuals, despite concerns about possible exposure of visa application data and growing warnings over China's broader cyber-espionage activity in Europe.
Six paradigm shifts occurred with LLMs during 2025.
Ukrainian national Artem Stryzhak pleaded guilty to conspiracy charges for deploying Nefilim ransomware against US companies.
Hackers accessed an online code library at the University of Sydney that contained historical test data files.
Sponsored
Attackers couldn't have chosen a better moment to strike Fastly [4]: As major customers shifted tens of petabytes of traffic to Fastly due to another provider's outage, several advanced DDoS attacks were unleashed on our platform…yet there was no end-user disruption.
Google patched two Chrome vulnerabilities exploitable via malicious web pages: a use-after-free in WebGPU (CVE-2025-14765) that enabled heap corruption, and an out-of-bounds read/write in the V8 JavaScript engine (CVE-2025-14766). Both flaws affect Chrome's 3.4 billion users and require no user interaction beyond visiting a crafted site. Update to version 143.0.7499.146 or later immediately.
Richmond Behavioral Health Authority in Virginia was hit by a ransomware attack in late September that encrypted parts of its network and revealed data on more than 113,000 individuals, including names, Social Security numbers, passport information, and financial and health details. Victims are advised to monitor their accounts and credit reports closely, as the ransomware group Qilin claims to have leaked 192 GB of stolen data.
Hewlett Packard Enterprise has patched a critical remote code execution vulnerability in HPE OneView, tracked as CVE-2025-37164 with a maximum CVSS score of 10.0, that allows unauthenticated attackers to execute code remotely. The flaw affects all versions before 11.00, making prompt patching essential to prevent exploitation.
A recent Raspberry Pi attack attempt on a French ferry highlights how rogue devices with cellular modems can bypass traditional network monitoring by creating their own internet connection from inside the perimeter. Analysts estimate 50% of enterprises remain vulnerable to similar physical attacks due to unsecured Ethernet ports and inadequate device authentication. Organizations should disable unused ports by default, implement 802.1X authentication, deploy physical layer fingerprinting tools, and monitor for external infrastructure connections when unknown devices appear.
Unit 42 analyzed the upgraded Mario encryptor used by the RansomHouse ransomware, which evolved from simple single-pass encryption to a two-factor scheme that uses both primary and secondary keys, with chunked file processing. The RaaS operation, tracked as Jolly Scorpius, specifically targets VMware ESXi infrastructure, using the MrAgent deployment tool to encrypt multiple virtual machines simultaneously. The enhanced encryption methodology, with non-linear file processing and dynamic chunk sizing, significantly complicates static analysis and decryption for defenders.
2025.12.19·unit42.paloaltonetworks.com·13 min read·TLDR InfoSec CVE-2025-55182, also known as React2Shell, is a critical remote code execution vulnerability in React Server Components and frameworks such as Next.js that allows attackers to execute arbitrary code with a single malicious HTTP request on vulnerable servers. Default configurations are affected and reliable public exploits exist. Attackers can gain initial access without authentication and then deploy tools such as reverse shells, RATs, cryptominers, and credential harvesters across Windows and Linux environments.
Sponsored
Security folks say genAI threats are a top concern, but few have the LLM and AI-powered security testing they need to address this novel risk. Cobalt's State of LLM Security Report 2025 [11] lays out the key LLM vulnerabilities, along with concrete recommendations for security
Datadog released an open-source dataset of over 17,000 malicious npm and PyPI packages identified through their GuardDog tool, categorized by ecosystem and whether they were compromised legitimate packages or published with malicious intent.
Wirebrowser is an open-source debugging and interception toolkit built on Chrome DevTools Protocol that combines network manipulation, API testing, and deep JavaScript memory inspection. Key features include Live Object Search for runtime patching of JS objects and Origin Trace (BDHS) for identifying user-land functions responsible for object creation or mutation through automated heap snapshots.
Verisoul provides an all‑in‑one fraud prevention platform that detects and blocks fake accounts and bots across the entire user lifecycle using device fingerprinting, behavioral analysis, and AI‑driven risk scoring to protect platforms from AI‑powered abuse and payment fraud.
A developer's Umami analytics container was exploited via CVE-2025-66478, a Next.js React Server Components deserialization flaw, resulting in Monero cryptominers running undetected for 10 days. Container isolation prevented the compromise from spreading because the container ran as non-root with no volume mounts or privileged access. The incident highlights the importance of knowing your dependencies' underlying frameworks and properly configuring container security boundaries.
UK healthcare tech firm DXS International has reported a cyberattack on its office servers serving NHS England. Services will remain operational while the scope of any data theft is under investigation. A ransomware group called DevMan claims to have stolen 300 GB of data. Authorities, including the UK Information Commissioner's Office and law enforcement, have been notified.
Anthropic's Claude models outperform rival LLMs on jailbreak resistance, prompt injection defenses, and limiting harmful or misleading outputs. While most vendors show mediocre and slow safety progress, Claude 4.x scores around 75–80% on jailbreak tests and nearly perfect on harmful content refusal, largely because Anthropic bakes safety alignment deeply into the training process instead of adding it as a final layer.
France arrested a 22-year-old suspect after BreachForums administrator "Indra" claimed responsibility for breaching the Interior Ministry and allegedly accessing criminal records, wanted persons databases, and Interpol systems containing over 16 million records.
The Arcanum Prompt Injection Taxonomy provides a structured taxonomy of prompt injection attacks, categorizing different types of attacks intents, techniques, and evasions.
The FBI, working with German and Finnish authorities, seized cryptocurrency exchange E-Note and indicted Russian national Mykhalio Petrovich Chudnovets for laundering over $70 million for ransomware operators since 2017.
Sponsored
Most endpoint vendors expect you to trust their marketing. Arctic Wolf hands you real malware and says "go ahead."
2025.12.18·cybersecurity.arcticwolf.com·Sponsor·TLDR InfoSec ±1 hour of your time. No hardware. No sales pitch.
Arctic Wolf observed active exploitation of critical Fortinet authentication bypass vulnerabilities (CVE-2025-59718 and CVE-2025-59719) starting December 12, just three days after Fortinet's disclosure. Attackers used crafted SAML messages to gain unauthenticated admin access to FortiGate devices with FortiCloud SSO enabled. Threat actors logged in via SSO from hosting provider IPs and immediately exfiltrated device configurations containing hashed credentials via the GUI. Organizations should promptly upgrade affected versions of FortiOS, FortiProxy, FortiWeb, and FortiSwitchManager, disable FortiCloud SSO as a workaround, reset all firewall credentials if compromise indicators are observed, and restrict management interface access to trusted internal networks.
Automotive parts supplier LKQ has confirmed it was hit in the Oracle E-Business Suite hacking campaign linked to the Cl0p ransomware group, exposing personal data, including SSNs and EINs, of over 9,000 sole proprietor suppliers. The company says the incident is limited to its Oracle EBS environment, though several terabytes of data were stolen, and notes it previously suffered a separate cyberattack affecting a Canadian business unit.
Gen Digital researchers have uncovered a new phishing campaign, which they dubbed GhostPairing, that abuses the device-linking feature of WhatsApp to hijack accounts. The victims are contacted by a known contact that shares a link allegedly leading to an online photo of a victim, which leads to a fake Facebook site that requests the user to initiate a device-linking process to log in. Attackers can then use the victim's account to forward the initial lure and propagate.
Analysis of recent supply chain attacks, including xz-utils, Ultralytics, tj-actions, and web3.js, revealed that most exploitation windows last only hours to days before detection and remediation. 8 out of 10 examined attacks had windows under one week. Implementing a 7-14 day dependency cooldown via Dependabot or Renovate would have prevented 80-90% of these attacks by simply waiting for compromised packages to be identified and removed before adoption. The technique is free, easy to configure, and among the most effective mitigations against mass-impact supply chain compromises.
Datadog's BewAIre is an LLM-powered system that reviews every pull request for malicious intent. Achieving over 99.3% accuracy and a 0.03% false positive rate after six months of tuning, the system addresses limitations of traditional static analysis by detecting intent-based attacks, such as credential exfiltration and obfuscated payloads, and by using recursive diff chunking to handle large PRs that attackers exploit to hide malicious code. Security teams can apply similar approaches by combining curated malicious datasets, prompt engineering with pattern exclusions, and adversarial red-teaming to build LLM-based detection that complements existing SAST/SCA tooling.
Retrieval-Augmented Generation (RAG) is commonly used by organizations to augment an LLM's knowledge base with additional information, such as internal documentation. RAG poses authorization risks as the LLM performs the search, so any authorization checks at the data source are skipped. This tutorial shows how to use S3 Access Grants with Amazon Bedrock Knowledge Bases to add authorization checks to the RAG flow.
OWASP Social OSINT Agent is an agent designed for OSINT investigations that leverages both text and vision-capable LLMs via any OpenAI-compatible API to gather, analyze, and synthesize user activity across single or multiple social media accounts.