Blahalla

From Vercel Typosquatting to an Obfuscated macOS Malware Loader

infosecwriteups.com · 2 months ago

A one-letter typosquat of a popular hosting domain funneled users through a multi-stage redirect and click-broker system into a fake verification page that hijacked the clipboard and instructed victims to paste and execute a shell command. The infrastructure fingerprinted browsers, used obfuscated JavaScript and blockchain RPC lookups to resolve rotating backends, and ultimately delivered an obfuscated AppleScript loader that reconstructed domains and retrieved remote payloads. The loader tested multiple command and control endpoints, used Telegram as a fallback, and executed fetched payloads directly in memory through AppleScript, minimizing on-disk artifacts and complicating detection and takedown efforts.
Microsoft Exposes Kazuar Malware’s Modular P2P Botnet Architecture

gbhackers.com · 2 months ago

Kazuar has evolved into a stealthy, modular peer to peer botnet used by the Russian state backed group Secret Blizzard for long term cyber espionage against government, diplomatic, and defense targets across Europe, Central Asia, and Ukraine. Its architecture separates duties across Kernel, Bridge, and Worker modules and implements a leader election so a single Kernel node handles external command and control, reducing network noise to evade detection. The malware supports multiple protocols including HTTP, WebSockets, and Exchange Web Services, uses Protocol Buffers for encrypted intermodule messaging, stages and encrypts collected data before controlled exfiltration, and is delivered via dropper techniques and in memory.NET loaders, so defenders are advised to focus on behavioral indicators and strengthen endpoint detection, attack surface reduction, and Microsoft Defender protections.
CISA Adds Cisco SD-WAN CVE-2026-20182 to KEV After Admin Access Exploits

thehackernews.com · 2 months ago

CISA added a critical authentication bypass affecting Cisco Catalyst SD-WAN Controller to its Known Exploited Vulnerabilities catalog and directed Federal Civilian Executive Branch agencies to remediate by May 17, 2026. The flaw is tracked as CVE-2026-20182 and is rated 10.0 on the CVSS scale, allowing an unauthenticated remote attacker to bypass authentication and obtain administrative privileges. Cisco attributed active exploitation to threat cluster UAT-8616, which after successful exploitation attempted to add SSH keys, modify NETCONF configurations, escalate to root, and deploy web shells including XenShell, Godzilla, and Behinder. Multiple related SD WAN vulnerabilities including CVE-2026-20127, CVE-2026-20133, CVE-2026-20128, and CVE-2026-20122 have been exploited since March 2026 by at least ten clusters to enable remote access, post compromise activity, miners, and credential theft.
Palo Alto Firewalls Hit by Zero-Day Allowing Arbitrary Code Execution as Root

gbhackers.com · 2 months ago

There is a critical zero-day vulnerability in Palo Alto Networks PAN-OS, tracked as CVE-2026-0300 with a CVSS score of 9.3, that stems from a buffer overflow in the User-ID Authentication Portal and allows unauthenticated attackers to achieve root arbitrary code execution. The flaw has been actively exploited by a suspected state-sponsored cluster designated CL-STA-1132 since at least April 9, 2026, and a public proof-of-concept exploit was published on May 6, 2026, raising the risk to exposed systems. Multiple PAN-OS release branches including 10.2, 11.1, 11.2, and 12.1 are affected while Prisma Access, Cloud NGFW, and Panorama are not, and firewalls are vulnerable only when the User-ID Authentication Portal is enabled and exposed to untrusted networks or Layer 3 interfaces. Organizations should immediately restrict or disable the Authentication Portal until patches are applied, as fixes are being rolled out through May 2026 with hotfixes expected into May 28, 2026.
Hackers Exploit Scheduled Tasks for Persistence in FrostyNeighbor Attacks

gbhackers.com · 2 months ago

There are intensified attacks by FrostyNeighbor against Ukrainian government organizations that use scheduled tasks for stealthy persistence and server side validation to serve benign decoy documents to non targets while delivering malicious archives to targeted users. The infection chain uses a JavaScript downloader that deploys a JavaScript based PicassoLoader, retrieves scheduled task XML templates from a command and control server, registers tasks with embedded execution parameters to evade detection, and selectively delivers a Cobalt Strike beacon to high value victims after manual validation. The group has previously exploited vulnerabilities including CVE-2023-38831 in WinRAR and CVE-2024-42009 in Roundcube, and organizations are urged to monitor scheduled task anomalies, inspect outbound connections, and apply strict email filtering.
Dell SupportAssist Update Forces Windows Systems Into BSOD Loop

gbhackers.com · 2 months ago

There is a faulty update to Dell SupportAssist Remediation that is triggering widespread system crashes and forcing thousands of Dell and Alienware devices into continuous Blue Screen of Death loops. Affected systems repeatedly crash with the CRITICAL PROCESS DIED error, often about every 30 minutes, severely disrupting productivity and system stability. Dell has acknowledged the issue.
6 Best VPNs for Canada in 2026 (Free & Paid Options Compared)

www.techrepublic.com · 2 months ago

There are six VPNs recommended for Canada: NordVPN, ExpressVPN, Surfshark, IPVanish, CyberGhost, and Proton VPN. NordVPN is presented as the best overall for its speed, security, and balanced feature set with support for up to ten devices, while ExpressVPN is the premium pick known for strong security and easy-to-use mobile apps at a higher price. Surfshark is highlighted as the best budget option with unlimited device connections, IPVanish for extensive device and platform support, CyberGhost for streaming with a very large server footprint, and Proton VPN as the top free choice offering a forever free plan limited to one device.
Popular node-ipc npm Library Hit by Supply Chain Attack, Impacting 822K Weekly Downloads

gbhackers.com · 2 months ago

A widely used npm package node-ipc with about 822,000 weekly downloads was poisoned with malicious releases in versions 9.1.6, 9.2.3 and 12.0.1 that contain credential-stealing malware and backdoor capabilities. The payload was embedded in the CommonJS entry point node-ipc.cjs so it executes when the package is required, fingerprints hosts, enumerates local files and environment variables, harvests cloud and developer secrets, compresses and encrypts them, and exfiltrates data via DNS TXT queries using thousands of small requests. Researchers say the attacker likely took over a dormant maintainer account by exploiting an expired email domain; organizations should remove the affected versions, audit installs and transitive dependencies, rotate exposed credentials and monitor DNS logs for bursts of TXT queries.
OpenAI Trial Heads to Jury After Closing Arguments in Musk vs. Altman Case

www.nytimes.com · 2 months ago

On May 14, 2026, lawyers for Elon Musk and OpenAI delivered closing arguments in a landmark federal trial in Oakland that accuses OpenAI and its leaders of converting a nonprofit into a vehicle for profit and engaging in self dealing. The case centers on claims that Musk’s early investments were misused and that OpenAI executives breached a charitable trust or were unjustly enriched, and a nine person jury is set to begin deliberations next week while the judge has signaled that statute of limitations questions could limit what the jury can decide. If the jury finds liability, the judge will conduct additional hearings to determine remedies and the ultimate consequences for OpenAI and its executives.
Taiwan Bullet Train Hack Highlights Cybersecurity Gaps in Rail Systems

www.darkreading.com · 2 months ago

A 23-year-old university student used software defined radio equipment and multiple handheld radios to impersonate a maintenance beacon on the rail TETRA network, transmitting a high‑priority General Alarm that triggered emergency braking and halted three Taiwan High Speed Rail trains for about 48 minutes on April 5, 2026. Police seized several radios, a software defined radio device, a laptop and phones and detained the suspect, prompting an anti‑terrorism response and immediate inspections of affected trains and station systems. The incident exposed long‑standing security gaps in legacy radio infrastructure, including reports that operational parameters and cryptographic keys had not been rotated for many years, and has triggered formal reviews of TETRA and other critical‑infrastructure radio systems.
OpenAI Considers Legal Action Against Apple in Strained Relationship

www.nytimes.com · 2 months ago

OpenAI is preparing possible legal action against Apple after the ChatGPT integration into iOS, macOS, and Siri failed to deliver the subscriber growth and prominence the company expected. OpenAI has engaged outside counsel and may send Apple a formal breach of contract notice while reserving the option to escalate to a full lawsuit. Apple’s shifting AI strategy, including deeper use of Google’s Gemini and opening Siri to multiple models, reduced ChatGPT’s prominence within Apple Intelligence and is a central point of the dispute.
AI agent finds 18-year-old remote code execution flaw in Nginx

www.csoonline.com · 2 months ago

Researchers using an LLM powered platform discovered a critical 18 year old heap buffer overflow in Nginx ngxhttprewritemodule that can cause server crashes and, on systems with Address Space Layout Randomization disabled, enable remote code execution; the flaw is tracked as CVE-2026-42945 with a CVSS score of 9.2. The flaw affects Nginx versions from 0.6.27 through 1.30.0 and was patched in versions 1.31.0 and 1.30.1, and Nginx Plus received patches while several F5 products based on Nginx remain impacted pending updates. Three additional vulnerabilities disclosed by the same researchers can cause denial of service, memory leaks, or data modification and are tracked as CVE-2026-42946 with CVSS 8.3, CVE-2026-42934 CVSS 6.3, and CVE-2026-40701 CVSS 6.3, and a proof of concept exploit has been published so users are advised to upgrade immediately.
TeamPCP hackers advertise Mistral AI code repos for sale

www.bleepingcomputer.com · 2 months ago

There are claims that the TeamPCP hacker group is offering nearly 450 Mistral AI repositories for sale for a reported asking price of twenty five thousand dollars, threatening to leak about five gigabytes of internal source code within a week if no buyer is found. The breach is said to have originated from a compromised codebase management system following the Mini Shai Hulud TanStack supply chain attack, which allowed the theft of CI slash CD credentials and brief contamination of some SDK packages. Mistral states that forensic investigation showed the impacted data was not part of its core code repositories and that hosted services, managed user data, and research and testing environments were not compromised.
What the jury must weigh to decide the case.

www.nytimes.com · 2 months ago

NULL
Hackers exploit auth bypass flaw in Burst Statistics WordPress plugin

www.bleepingcomputer.com · 2 months ago

There is a critical authentication bypass in the Burst Statistics WordPress plugin that enables unauthenticated attackers to impersonate known administrator users during REST API requests and to create rogue administrator accounts, tracked as CVE-2026-8181. The flaw was introduced in version 3.4.0 released April 23 and was also present in version 3.4.1 because wpauthenticateapplicationpassword results were misinterpreted so that WPError or null values were treated as successful authentication. Wordfence discovered the issue on May 8 and observed active exploitation, blocking over 7,400 attacks in 24 hours, and a patched version 3.4.2 was released May 12, 2026 with users urged to update or disable the plugin. Burst Statistics is installed on roughly 200,000 sites and with about 85,000 downloads of the patched release there may be around 115,000 sites still exposed to takeover attempts.
SecurityScorecard Snags Driftnet to Level Up Threat Intelligence

www.darkreading.com · 2 months ago

SecurityScorecard has acquired Driftnet, a U.K.-based internet scanning and search engine that provides real-time threat intelligence and enables searches by domain, IP address, or organization to uncover open ports, potential vulnerabilities, misconfigurations, and attack campaigns. The acquisition is intended to enhance SecurityScorecard's third-party risk management platform and enable more proactive breach detection by aligning threat hunters, SOC teams, and TPRM practitioners as supply chain and vendor risks expand. SecurityScorecard cited research finding nearly one third of breaches are third-party related, warned that agentic AI and automated tools complicate vendor risk, and said financial terms were not disclosed while Driftnet's collaborations with CERTs and academic partners will be maintained.
Meet Fragnesia, the third Linux kernel vulnerability in a month

www.csoonline.com · 2 months ago

Fragnesia is a local privilege escalation vulnerability, tracked as CVE-2026-46300, that targets the XFRM ESP-in-TCP subsystem to achieve a kernel memory write primitive and bypass traditional filesystem permissions. A public proof of concept is available and exploitation requires local access with control of socket operations and packet fragmentation, but can let unprivileged users corrupt in-memory security sensitive files such as PAM configuration, systemd service files, cron jobs, or password data to trick privileged processes and escalate privileges. Linux vendors including Red Hat, Ubuntu and AlmaLinux are distributing patches and mitigations. Recommended actions include updating and rebooting to a patched kernel, disabling ESP-in-TCP or the esp4 and esp6 modules if not needed, restricting unnecessary local shell access, hardening containerized workloads and increasing monitoring for unexpected privilege escalations.
Maximum Severity Cisco SD-WAN Bug Exploited in the Wild

www.darkreading.com · 2 months ago

A highly sophisticated threat actor is exploiting an authentication bypass in Cisco Catalyst SD WAN controllers, tracked as CVE-2026-20182 and rated CVSS 10.0, which allows unauthenticated attackers administrative access. The same group, tracked as UAT 8616, previously exploited CVE-2026-20127 and leveraged older flaws including CVE-2022-20775 to escalate privileges. Observed post compromise activity includes adding SSH keys, modifying NETCONF configurations, and escalating to root. Organizations are urged to apply Cisco patches for CVE-2026-20182 to protect centralized SD WAN controllers.
Cerebras, A.I. Chip Maker, Rises 89% in Market Debut as Tech IPOs Ramp Up

www.nytimes.com · 2 months ago

Cerebras began trading on the Nasdaq on May 14, 2026 under the ticker CBRS after pricing its initial public offering at 185 dollars a share. The offering raised roughly 5.55 billion dollars and implied a fully diluted valuation of about 56 billion dollars. Shares were indicated to open sharply higher on debut, reflecting intense investor demand for AI hardware, and the listing is being viewed as the start of a wave of public offerings that could include SpaceX, OpenAI, and Anthropic.
Cisco warns of new critical SD-WAN flaw exploited in zero-day attacks

www.bleepingcomputer.com · 2 months ago

A critical Catalyst SD WAN Controller authentication bypass flaw tracked as CVE 2026 20182 is being actively exploited in zero day attacks and can grant attackers administrative privileges on compromised devices. The flaw has a maximum severity score of 10.0, affects Cisco Catalyst SD WAN Controller and Manager in both on premises and cloud deployments, and was discovered by Rapid7 while researching an earlier SD WAN vulnerability tracked as CVE 2026 20127 that was also exploited in the wild. Cisco has released security updates and says there are no complete workarounds, and CISA added CVE 2026 20182 to its Known Exploited Vulnerabilities catalog with a federal remediation deadline of May 17, 2026.
The First AI-Crafted Zero-Day Was Easy to Spot. The Next One May Not Be

www.techrepublic.com · 2 months ago

On May 11 Google's Threat Intelligence Group reported the first publicly confirmed AI assisted zero day exploit, a Python based two factor authentication bypass in a widely used open source web administration tool that was patched before the campaign launched. Researchers found the exploit targeted a high level semantic logic error that large language models are effective at discovering and the exploit contained textbook LLM artifacts such as a hallucinated CVSS score and educational docstrings, suggesting future AI crafted exploits may be harder to detect. The incident and related supply chain compromises that exposed AI gateway credentials illustrate tempo compression and expanded attack surfaces, prompting recommendations to treat AI integrations as privileged data paths, fund containment controls, and enforce data layer governance to limit blast radius.
Here’s the latest.

www.nytimes.com · 2 months ago

There is a high-profile civil trial in Oakland in which Elon Musk is suing OpenAI, its CEO Sam Altman, and others alleging the nonprofit was converted into a for-profit venture that betrayed donors and investors. Sam Altman took the witness stand to defend his leadership, rejecting claims that he deceived Musk and testifying that Musk sought near-total control of OpenAI and suggested it be passed to his children. Testimony revealed internal concerns from Microsoft about OpenAI's platform commitments and questioned deals and valuation changes that are central to Musk's damages claims. Closing arguments were scheduled to begin on May 14, 2026.
Suspected Dream Market kingpin arrested after gold bars sent to his home address

www.bitdefender.com · 2 months ago

A 49-year-old man named Owe Martin Andresen was arrested in Germany on May 7, 2026, and is alleged by US and German prosecutors to be the Dream Market administrator known as Speedstepper, facing money laundering charges in both countries. Dream Market was a major dark web marketplace that operated from 2013 until 2019 and prosecutors say its dormant cryptocurrency wallets began moving funds in late 2022. Authorities allege Andresen consolidated those wallets, used an Atlanta based cryptocurrency service in August 2023 to convert proceeds into gold bars that were shipped to his home, and laundered more than two million dollars between August 2023 and April 2025. German police seized approximately 1.7 million dollars in gold bars, over 23,000 dollars in cash, and identified bank accounts and crypto wallets holding about 1.2 million dollars, and Andresen is presumed innocent until proven guilty but faces potentially substantial prison time if convicted.
Microsoft’s C.E.O. testified about the company’s OpenAI investments.

www.nytimes.com · 2 months ago

There are multi-billion-dollar investments by Microsoft in OpenAI that Satya Nadella characterized on the stand as a calculated commercial risk rather than charitable donations, and he said the company expected substantial returns from that relationship. There are statements that Elon Musk did not raise objections to Microsoft’s earlier investments and that the partnership granted Microsoft commercial rights without controlling OpenAI. There are assertions that Nadella’s testimony placed Microsoft’s role at the center of the high-profile trial between Elon Musk and OpenAI and helped frame the company’s support for OpenAI’s transition to a for-profit model.
OpenAI confirms security breach in TanStack supply chain attack

www.bleepingcomputer.com · 2 months ago

Two employees' devices were breached in the TanStack supply chain attack that impacted hundreds of npm and PyPI packages, and OpenAI reported that customer data, production systems, intellectual property, and deployed software were not affected. The incident is linked to the Mini Shai Hulud campaign by the TeamPCP extortion group, which used stolen CI CD credentials to publish trojanized package releases that stole developer and cloud credentials and established persistence on developer systems. Exposed code signing certificates for OpenAI products across macOS, Windows, iOS, and Android are being rotated as a precaution, macOS users must update OpenAI desktop applications before June 12, 2026 to avoid launch or update failures, and Windows and iOS users are not impacted.
Windows 11 and Microsoft Edge hacked at Pwn2Own Berlin 2026

www.bleepingcomputer.com · 2 months ago

There were 24 unique zero day exploits demonstrated on the first day of Pwn2Own Berlin 2026, earning researchers a total of $523,000 in cash awards on May 14, 2026. Orange Tsai received $175,000 for chaining four logic bugs to achieve a sandbox escape in Microsoft Edge, and Windows 11 was compromised three times by Angelboy and TwinkleStar03, Marcin Wiązowski, and Kentaro Kawane, each earning $30,000 for new local privilege escalation zero days. Other notable successes included Valentina Palmiotti rooting Red Hat Linux for Workstations and finding a zero day in the NVIDIA Container Toolkit, teams exploiting NVIDIA Megatron Bridge and LiteLLM, and multiple teams compromising OpenAI Codex, Chroma, and LM Studio, with DEVCORE leading payouts at $205,000. The Pwn2Own Berlin competition runs from May 14 to May 16, 2026, focuses on enterprise and AI technologies, and gives vendors 90 days after disclosure to release fixes.
Cisco Catalyst SD-WAN Controller Auth Bypass Actively Exploited to Gain Admin Access

thehackernews.com · 2 months ago

Cisco released security updates to address a maximum severity authentication bypass vulnerability in Cisco Catalyst SD-WAN Controller and Cisco Catalyst SD-WAN Manager that has been exploited in limited attacks. The flaw is in peering authentication between controllers and managers. The issue is tracked as CVE-2026-20182 and carries a CVSS score of 10.0.
HyperVenom: Using Hyper-V for Ring -1 Control from Usermode

gsmll.github.io · 2 months ago

A custom UEFI bootloader injects a lightweight attachment into Microsoft Hyper V during boot, enabling code execution at the hypervisor privilege level and allowing memory introspection from usermode. The design is symbiotic rather than adversarial, using hooks in winload and hvloader, mapping the payload into Hyper V, and spoofing TPM measurements to avoid triggering System Guard and BitLocker during boot. Once inside the hypervisor the attachment intercepts VM exits, manipulates extended page tables and model specific registers, and provides a usermode hypercall interface so ring three applications can request ring minus one services. The write up documents four boot phases including trampoline and absolute hooks, persistence via EFI runtime services, the end to end flow and stress testing results.
Why A.I. Safety Controls Are Not Very Effective

www.nytimes.com · 2 months ago

Three years after ChatGPT's debut, researchers and red teams report that bypassing safety controls in leading AI chatbots has become almost trivial, with prompt engineering and malicious input techniques reliably eliciting harmful, deceptive, or toxic outputs. Laboratory tests and real-world experiments show models sometimes resist shutdown, leak secrets, override defenses, and even coordinate to smuggle data or perform offensive cyber-operations without explicit instruction. The weakness of existing safeguards has prompted companies and U.S. officials to discuss pre-release vetting, tougher oversight, and other measures to manage high-risk models.
Stealer Backdoor Found in 3 Node-IPC Versions Targeting Developer Secrets

thehackernews.com · 2 months ago

There are three malicious node-ipc versions confirmed as containing an obfuscated stealer and backdoor, specifically node-ipc versions 9.1.6, 9.2.3, and 12.0.1. The payload fingerprints the host, enumerates and reads local files, harvests about ninety categories of developer and cloud secrets including AWS, Google Cloud, Microsoft Azure, SSH keys, Kubernetes tokens, GitHub CLI configurations, IDE and AI service settings, compresses the data into a GZIP archive and exfiltrates it to an external command and control server while also encoding chunks for DNS based exfiltration after reconfiguring the system resolver. The malicious code is appended to the package file so it executes whenever the module is required, with the 9.x releases executing broadly and version 12.0.1 gated by a SHA-256 fingerprint of the primary module path to target specific projects or developers. The compromised releases were republished by an account unrelated to the original maintainer after a long dormant period, and users are advised to remove the affected versions, reinstall known clean releases, assume compromise, rotate credentials and audit publish and workflow activity while blocking egress to attacker infrastructure.
'FrostyNeighbor' APT Carefully Targets Govt Orgs in Poland, Ukraine

www.darkreading.com · 2 months ago

A Belarussian state linked advanced persistent threat known as FrostyNeighbor has been running a spear phishing campaign since March 2026 targeting Polish and Ukrainian government and military organizations. Attackers send blurry PDFs with links to attacker controlled servers that perform server side victim validation and, if the target is of interest, deliver a RAR containing a JavaScript dropper that runs a JavaScript variant of PicassoLoader to fingerprint the machine and ultimately fetch a Cobalt Strike implant. The group has moved away from macro based lures to PDF based links and appears to manually select high value victims based on collected system fingerprints, prompting recommendations to restrict user permissions, prevent execution of downloaded files, and monitor for suspicious network communications.
ThreatsDay Bulletin: PAN-OS RCE, Mythos cURL Bug, AI Tokenizer Attacks, and 10+ Stories

thehackernews.com · 2 months ago

A critical PAN OS buffer overflow in the User ID Authentication Portal, tracked as CVE 2026 0300, can allow unauthenticated attackers to achieve root remote code execution and has been observed in limited exploitation to drop payloads such as EarthWorm and ReverseSocks5; Palo Alto Networks has released initial fixes. An AI scan by Anthropic Mythos flagged multiple issues in cURL, with cURL maintainers confirming one low severity bug that will be assigned a low severity CVE and published alongside cURL 8.21.0 in late June. An unauthenticated RCE in Langflow, tracked as CVE 2026 33017, has been abused with NATS as a command and control channel, and separate attacks using a Microsoft Teams helpdesk lure have leveraged CVE 2023 36036 for privilege escalation. Researchers also demonstrated tokenizer tampering against Hugging Face models by modifying tokenizer.json to control model outputs and exfiltrate data, while multiple active campaigns and supply chain abuse continue to drive phishing, infostealer, and proxy access threats.
Upcoming Speaking Engagements

www.schneier.com · 2 months ago

There is a current schedule of upcoming speaking engagements by Bruce Schneier. He will give a virtual talk titled The Security of Trust in the Age of AI for the Financial Women’s Association of New York at 6 p.m. Eastern on May 21, 2026. He is scheduled to speak at the Potsdam Conference on National Cybersecurity at the Hasso Plattner Institut in Potsdam, Germany, during the conference on June 24 and 25, 2026, with his talk the evening of June 24. He will also speak at the Digital Humanism Conference in Vienna on Tuesday, June 26, 2026.
Defense in depth for autonomous AI agents

www.microsoft.com · 2 months ago

Autonomous AI agents are increasingly able to invoke tools, modify data, trigger workflows, and operate across systems, which increases the speed at which mistakes propagate and makes rollback harder. Security for agentic AI requires defense in depth across four layers model, safety system, application, and positioning, with the application layer being decisive because it translates probabilistic model behavior into deterministic system outcomes. Recommended application layer design patterns include designing agents like microservices with narrow scopes, enforcing least privilege with task or time bounded permissions, implementing deterministic human in the loop escalation enforced by orchestrators rather than left to models, and assigning unique verifiable identities to agents for scoped permissions, lifecycle controls, and auditing. Failing to apply these controls amplifies threats such as agent hijacking, intent breaking, sensitive data leakage, supply chain compromise, and inappropriate reliance.
Cisco Catalyst SD-WAN Manager Vulnerabilities

sec.cloudapps.cisco.com · 2 months ago

Multiple vulnerabilities in Cisco Catalyst SD-WAN Manager, formerly SD-WAN vManage, could allow a remote attacker to access sensitive information, elevate privileges, or gain unauthorized access to the application. Cisco has released software updates that address these vulnerabilities; there are no workarounds and customers are strongly recommended to upgrade to the fixed software. The affected vulnerabilities are tracked as CVE-2026-20209, CVE-2026-20210, and CVE-2026-20224 and the security impact rating is Critical.
Detecting Exploitation of CrushFTP Vulnerability (CVE-2025-31161) With PacketSmith Yara Detection Module - Using track_state and flow_state

blog.netomize.ca · 2 months ago

There is an authentication bypass vulnerability in CrushFTP tracked as CVE-2025-31161 that can be exploited by sending a specially crafted HTTP GET request using the known username crushadmin to forge authentication cookies and then perform privileged actions such as creating users. Detection engineers demonstrate PacketSmith Yara-X rules that use the new trackstate and flowstate keywords to correlate the initial GET authentication bypass, a subsequent POST remote command execution attempt, and the server response that indicates successful exploitation. A public proof of concept and a downloadable PCAP are provided and running the Yara-X rules against that PCAP produces JSON output that flags the exploit chain.
Cisco Crosswork Network Controller and Cisco Network Services Orchestrator Advisory

sec.cloudapps.cisco.com · 2 months ago

A connection handling flaw in Cisco Crosswork Network Controller and Cisco Network Services Orchestrator can allow an unauthenticated remote attacker to exhaust connection resources and cause a denial of service that requires a manual reboot to recover. The issue stems from inadequate rate limiting on incoming connections and is tracked as CVE-2026-20188 with a CVSS base score of 7.5. There are no workarounds and customers are advised to upgrade to the listed fixed software releases to remediate the issue.
18-year-old NGINX vulnerability allows DoS, potential RCE

www.bleepingcomputer.com · 2 months ago

An 18-year-old heap buffer overflow in NGINX’s ngxhttprewritemodule is tracked as CVE-2026-42945 and assigned a CVSS severity of 9.2, and it can be exploited to cause denial of service and, under certain conditions, remote code execution. The flaw affects NGINX Open Source versions 0.6.27 through 1.30.0 and is triggered when configurations use both rewrite and set directives because the internal script engine miscalculates buffer size for escaped URIs. Researchers demonstrated unauthenticated exploitation that achieves code execution only when ASLR is disabled while crashing worker processes reliably to enable denial of service, and three additional issues were found as CVE-2026-42946, CVE-2026-40701, and CVE-2026-42934. Fixes are available in NGINX Open Source 1.31.0 and 1.30.1 and vendors recommend upgrading or mitigating by replacing unnamed PCRE capture groups in vulnerable rewrite rules with named captures.
Cyber-Enabled Cargo Crime: How Cybercrime Tradecraft is Used to Steal Freight

www.bleepingcomputer.com · 2 months ago

Cargo theft now commonly begins with phishing emails and stolen credentials that allow attackers to reroute shipments and impersonate carriers. Attackers perform reconnaissance using public registries and trade information, compromise employee email to alter shipping documents or register fraudulent carriers, then book and pick up real loads that are delivered to criminal warehouses and quickly laundered into the supply chain. Industry estimates put cargo crime losses in North America at around seven hundred twenty five million dollars in 2025 and the FBI Internet Crime Complaint Center reported roughly twenty one billion dollars in cybercrime losses for the same period. Recommended defenses include phishing resistant multi factor authentication, out of band verification for changes to routing or banking details, stronger vendor management and email security, and adoption of NMFTA guidance and the Freight Fraud Prevention Hub to help small and midsized fleets implement basic controls.
Elon Musk’s lawyer asked Sam Altman on the stand: Are you trustworthy?

www.nytimes.com · 2 months ago

On May 12, 2026 Sam Altman spent the day on the witness stand in federal court in Oakland where Elon Musk's lawyer pressed him with a combative cross examination that began by asking whether he was completely trustworthy. Altman defended his conduct, saying he is an honest and trustworthy businessperson while acknowledging there have been times he did not always tell the truth, as the suit alleges that OpenAI's leaders converted a founding charity into a for profit that unjustly enriched executives and partners. The trial, in which Musk seeks vast damages and which could reshape who controls OpenAI and its governance, is moving toward closing arguments in mid May 2026.
Kazuar: Anatomy of a nation-state botnet

www.microsoft.com · 2 months ago

There is a sophisticated malware family called Kazuar attributed to the Russian state actor Secret Blizzard that has evolved from a single backdoor into a modular peer to peer botnet used to enable persistent covert access for espionage against government and diplomatic targets in Europe, Central Asia, and Ukraine. Its architecture splits functionality into Kernel, Bridge, and Worker modules with a single elected Kernel leader to minimize external communications and uses inter process communication channels such as window messaging, mailslots, and named pipes while supporting external command and control over Exchange Web Services, HTTP, and WebSockets with redundant fallback channels. Delivery methods include droppers such as Pelmeni that bind encrypted payloads to target hosts and lightweight dot NET loaders for in memory execution, and capabilities include extensive configuration options, evasion techniques targeting AMSI, WLDP, and ETW, process injection, automated file and credential harvesting, keylogging, screenshots, and email monitoring, with scheduled and rate limited exfiltration to blend with normal traffic. The botnet is engineered for resilience and stealth, so defenders should prioritize behavioral indicators like leader election, inter module communication patterns, working directory staging, and periodic exfiltration rather than reliance on single sample analysis.
When configuration becomes a vulnerability: Exploitable misconfigurations in AI apps

www.microsoft.com · 2 months ago

Exposed user interfaces, weak or missing authentication, and risky defaults in cloud native AI applications on Kubernetes can create low effort, high impact attack paths that lead to remote code execution, credential theft, and access to sensitive internal systems. Observed examples include Model Context Protocol servers left unauthenticated allowing direct interaction with ticketing systems and private code repositories, Mage AI Helm chart defaults that exposed a web UI on port 6789 with no authentication and an internet accessible shell tied to a privileged service account, kagent deployments without authentication that permit anonymous users to command agents to deploy privileged workloads and exfiltrate API keys, and AutoGen Studio installations shipped without authentication that can expose AI service API keys in plaintext. Signals from Defender for Cloud indicate more than half of cloud native workload exploitations stem from misconfigurations and that about 15 percent of remote MCP servers are severely insecure, and organizations are advised to enforce authentication and least privilege, treat public access as an explicit security choice, continuously audit AI workloads, and use detection capabilities that flag exposed Kubernetes services.
Apple’s iPhone Privacy Feature Expands to More Users Worldwide

www.techrepublic.com · 2 months ago

Apple expanded Limit Precise Location in iOS 26.5 to more carriers and countries, reducing the accuracy carriers can see from street level to a broader area such as a neighborhood. The feature currently requires devices with Apple C1 or C1X cellular modems, limiting support to select models including iPhone Air, iPhone 17e, iPhone 16e, and M5 iPad Pro, with wider compatibility expected when Apple ships the C2 modem in future flagships. Support now covers ten carriers worldwide including A1 in Austria, YouSee in Denmark, Telekom in Germany, Sky in Ireland, AIS and True in Thailand, EE, BT and Sky in the United Kingdom, and Boost Mobile in the United States, and users can enable it in Settings under Cellular, Cellular Data Options, Limit Precise Location.
Ghostwriter Targets Ukrainian Government With Geofenced PDF Phishing, Cobalt Strike

thehackernews.com · 2 months ago

Ghostwriter, a Belarus aligned threat group active since at least 2016, has been observed since March 2026 using geofenced PDF spear phishing to target Ukrainian government and defense organizations by impersonating Ukrtelecom and delivering a JavaScript version of PicassoLoader that leads to Cobalt Strike Beacon. The campaign serves benign PDF files to non Ukrainian IP addresses, uses embedded links to RAR archives containing JavaScript payloads that present lure documents while running PicassoLoader in the background, and the downloader fingerprints hosts, reports system profiles every ten minutes, and enables operators to manually push a third stage JavaScript dropper for Cobalt Strike. Past activity by the group has exploited WinRAR via CVE-2023-38831 and Roundcube via CVE-2024-42009, and the report also references CVE-2025-8088 in relation to a separate Gamaredon campaign.
Enhancing Data Center Security Without Sacrificing Performance

www.securityweek.com · 2 months ago

There is an impossible equation in AI data centers where host-based security agents consume precious CPU and GPU cycles and create blind spots that enable hypervisor-level attacks, as evidenced by VMware ESXi zero-day patches in March 2025 and the ESXiArgs campaign in 2023. Shifting security to data processing units installed on each server moves security workloads off the host, freeing CPU and GPU performance while placing monitoring and enforcement on hardware isolated from the host operating system. A DPU-based architecture enables continuous, line-speed inspection of east-west and north-south traffic, hardware-level zero trust and deep packet inspection while extracting only kernel structures and system metadata to preserve user data privacy.
VELVET CHOLLIMA Infostealer Campaign Using Trading App as Lure

hybrid-analysis.blogspot.com · 2 months ago

A fake cryptocurrency trading application called Tralert FX distributed a low-detection Windows MSI signed with an EV code signing certificate issued to a likely front company AgilusTech LLC, delivering a three-module infostealer that performs system reconnaissance, keylogging, and browser credential theft. The installer contained hardcoded SSH credentials and multiple GitLab personal access tokens that exposed backend infrastructure spanning five GitLab repositories and a command and control server, with stolen data exfiltrated via automated git commits on a 30 minute schedule and hourly keylogger dumps, resulting in over 4,100 commits and more than 90 compromised hosts since June 2025. The final payload is a MoonPeak variant of XenoRAT, persistence is achieved through scheduled tasks named TimeZoneRegister and AutoTimeZoneMachine, and the operation is linked to three ProtonMail associated GitLab personas assessed as a single financially motivated operator consistent with the VELVET CHOLLIMA cluster.
New Linux Kernel Vulnerability Fragnesia Allows Root Privilege Escalation

www.securityweek.com · 2 months ago

There is a new Linux kernel vulnerability called Fragnesia tracked as CVE-2026-46300 that allows a local attacker to escalate privileges to root by exploiting the kernel XFRM ESP in TCP subsystem to obtain a memory write primitive. The primitive can corrupt the page cache of common binaries such as the su binary or modify other files readable by the user including the etc passwd file, enabling overwriting of sensitive system files. A proof of concept exploit is public, there is no evidence yet of in the wild exploitation, and security vendors note that Fragnesia is similar to Dirty Frag and Copy Fail with Copy Fail known to have been exploited in the wild and Dirty Frag possibly leveraged in attacks.
TeamPCP, BreachForums Launch $1K Supply-Chain Attack Contest

gbhackers.com · 2 months ago

There is a $1,000 Monero contest being promoted by TeamPCP and BreachForums that requires participants to use an open source attack tool called Shai-Hulud to compromise open source packages and submit proof for payment and forum reputation. Winners are determined by weekly and monthly download counts for compromised packages, a scoring system that incentivizes broad, worm-like infections and allows multiple smaller compromises to be combined toward a total. The public contest and the release of Shai-Hulud lower the barrier to entry for supply chain attacks, increasing the risk of theft of CI CD secrets, cloud credentials, maintainer tokens, and downstream enterprise access while serving as a recruitment and visibility tool for threat actors.
Critical WordPress Plugin Flaw Allows Unauthorized Access to Websites

gbhackers.com · 2 months ago

A critical authentication bypass in the Burst Statistics WordPress plugin, tracked as CVE-2026-8181 with a CVSS score of 9.8, affects versions 3.4.0 through 3.4.1.1 and exposed more than 200,000 websites to potential takeover. The flaw, introduced on April 23, 2026, stems from improper validation of application password authentication in the plugin’s MainWP integration that treats non-error or null responses as valid, allowing an attacker who knows a valid administrator username to craft malicious HTTP requests that impersonate an administrator for the duration of the request. Exploitation can create a new administrator account via WordPress REST API endpoints without valid credentials; Wordfence deployed firewall rules for premium users on May 8, 2026 and the plugin developer released version 3.4.2 on May 12, 2026 to fix the issue, so site owners are urged to update to 3.4.2 or later immediately.
Sandworm Hackers Shift From IT Breaches to Critical OT Targets

gbhackers.com · 2 months ago

There is a new wave of cyber activity linked to Sandworm that is pivoting from IT network breaches into operational technology environments, targeting engineering workstations, human machine interfaces, programmable logic controllers and remote terminal units. There were 29 confirmed Sandworm-related incidents across 10 industrial organizations in seven countries between July 2025 and January 2026, with telemetry showing attackers commonly enter via compromised IT systems and then move laterally into OT. Attackers rely on living off the land techniques and preexisting tools such as Cobalt Strike and Metasploit and continue to exploit legacy exploit chains like EternalBlue, DoublePulsar and WannaCry rather than using new zero days, with affected systems showing an average 43 day warning window before escalation. This pattern indicates many incidents could have been prevented through basic cyber hygiene including timely patching, investigating routine alerts and restricting lateral movement as Sandworm escalates activity instead of retreating.