Blahalla

The Next AI Security Failure May Start With a Trusted Assistant

www.techrepublic.com · a month ago

There is a class of vulnerability in AI coding assistants where a network sandbox bypass combined with prompt injection can be used to exfiltrate data, and a widely used tool was quietly patched for a SOCKS5 hostname null byte weakness. There are systemic gaps because organizations often rely on model layer controls such as system prompts, output filters, and sandboxes that can be defeated, leaving no independent data layer enforcement to stop unauthorized access. There are recommended data layer controls including authenticating agents and tying requests to the human who authorized them, enforcing attribute based access policies, encrypting returned data, and writing interactions to a tamper evident audit log. There are industry findings that most custom assistants lack adequate protection, many organizations cannot enforce purpose limits or terminate misbehaving agents, and AI enabled adversary activity is rising sharply, underscoring the need to treat assistants as data processing systems with least privilege, encryption, and audit rigor.
UK Cyberspying Chief Calls AI ‘an Unstoppable Force’ and Warns About Russia

www.securityweek.com · a month ago

Artificial intelligence is an unstoppable force being weaponized in ways that fall just short of traditional warfare, and Britain faces a narrowing window to stay ahead as tech companies release AI innovations at rapid pace. The director of GCHQ warned that Russia is scaling up daily hybrid activity targeting critical infrastructure, democratic processes, supply chains and public trust, and cited new intelligence suggesting almost half a million Russian soldiers have been killed since February 2022. GCHQ is planning to harden machine speed cyber defenses by integrating agentic AI, and officials urged governments, companies and citizens to treat cybersecurity with far greater urgency while reinforcing international partnerships, particularly the U.K.-U.S. intelligence relationship.
Latin American Cybercriminals Hoover Up Government Data

www.darkreading.com · a month ago

There are reports of a purported leak exposing 5.8 million records of Uruguayan citizens after a group calling itself La Pampa Leaks claimed to have compromised the government-sponsored identity service managed by telecommunications provider Antel, which responded that passwords, signature PINs, private keys tied to digital certificates, and credentials were not compromised. There are broader trends of Latin American cybercriminal groups monetizing government and citizen data, including a claim by the Chronus Group that it stole data from 25 Mexican government agencies and reporting that Colombia's health ministry faced more than 23 million attempted attacks in March. There are analyses from security firms noting that public administration was the most-breached sector in the region, accounting for about 21 percent of breaches, and that regional actors are shifting to pure extortion and high-volume data exfiltration instead of deploying encryption. There are recommendations to harden identity controls, reduce exposed services and open ports, and prioritize patching of vulnerabilities to build resilience against these targeted attacks.
AI-Assisted Exploit Development Outpaces Scanner Detection

www.darkreading.com · a month ago

Analysis of 69,159 CVEs shows attackers using AI reduced the time to develop a working exploit for a disclosed vulnerability from about 125 days in January 2025 to roughly half a day by April 2026. That acceleration created visibility gaps for defenders, with 83.2 percent of critical vulnerabilities producing a visibility gap, 55.7 percent of critical CVEs never receiving detection coverage from major scanners, and 62 percent of vulnerabilities having exploits circulating before detection signatures were available. Median detection lags among leading scanners were measured at 0.1 days for Tenable, 2.9 days for Qualys, and 5.1 days for Rapid7, prompting recommendations to adopt continuous software inventory checks, software bill of materials matching, and threat intelligence feeds so teams can identify affected assets within minutes of disclosure.
Grandoreiro Malware and BTMOB RAT Campaigns Target Windows and Android Users

thehackernews.com · a month ago

Two banking trojan campaigns are targeting Windows and Android users across Latin America and Europe, with Grandoreiro focusing on companies in Spain, Portugal, and Mexico and BTMOB aimed at mobile users in Brazil. Grandoreiro operations use DLL side-loading to launch Delphi-built DLLs that incorporate sgcWebSockets and rely on WebRTC, STUN, or ICE to hide in web-conferencing traffic, and attackers deliver payloads via phishing lures such as Mediafire-hosted ZIPs containing obfuscated Visual Basic scripts that prompt fake Adobe Reader updates to steal banking credentials. BTMOB is an Android remote access trojan sold as malware-as-a-service with an APK builder that spreads through phishing sites and fake app listings, abuses Android accessibility services to automate credential theft and remote control, and as of May 2026 is at version 4.5.5 with commercial pricing and leaked source code increasing its availability.
New Phishing Technique - Vaultjacking: One Captured PIN, the Entire Google Password Manager Vault

phishu.net · a month ago

A new phishing technique called Vaultjacking uses a single AiTM landing page that spoofs a Google passkey or password manager PIN to capture that PIN and then access and exfiltrate all of a victim's third party credentials saved in Chrome, including passkeys. A single successful phish on one site can therefore lead to total compromise of a user’s Chrome saved credentials, and the technique is demonstrated in the Phishu framework.
Malicious npm Package Stole Files From Claude AI User Directory via GitHub

thehackernews.com · a month ago

A malicious npm package named mouse5212-super-formatter was found to exfiltrate files from the /mnt/user-data directory used by Anthropic's Claude AI in an operation codenamed Malware-Slop. The package masquerades as an archive deployment sync utility but authenticates to GitHub during postinstall using an environment access token or a hard coded fallback, creates a repository if needed and recursively uploads local files to a threat actor controlled GitHub account while writing fake network logs. Stolen files were stored in randomly named folders, the package leaked the operator's private token and account details, and it remained available on npm with an estimated 676 downloads before the linked GitHub account, created on May 26, 2026, became unavailable. Security researchers warned that AI assistance is lowering the bar for creating malware and may increase the volume of sloppy supply chain threats.
7-Eleven Breach: Hackers Claim 600,000 Records Stolen

www.techrepublic.com · a month ago

A system intrusion at 7-Eleven on April 8 exposed internal systems used to store franchise application documents, and breach notification letters dated May 1 said attackers accessed personal information submitted during the application process including names, addresses, phone numbers and dates of birth. Analysis of leaked files linked to the incident estimated roughly 185,300 affected individuals and reported Social Security numbers in a subset of records, while the extortion group ShinyHunters claims to have stolen more than 600,000 records from a connected Salesforce environment and later published a 9.4 gigabyte archive, although 7-Eleven has not officially attributed the breach to that group. 7-Eleven hired a forensic investigation firm, is offering up to 24 months of identity theft protection and CyberScan monitoring through IDX, and cybersecurity experts say ShinyHunters increasingly targets Salesforce related environments through phishing, third party integrations and cloud misconfigurations while the FBI warns organizations not to pay extortion demands.
FastAPI-based AI tools exposed to authentication bypass by flaw in Starlette framework

www.csoonline.com · a month ago

A single malformed character in an HTTP Host header can allow an unauthenticated attacker to bypass access controls in applications built on the Starlette Python framework that powers FastAPI, an issue tracked as CVE-2026-48710. The flaw stems from Starlette rebuilding the request address by joining the Host header to the path while validating the combined URL and its parts with different rules, so a Host value containing a slash, question mark, or hash can shift where the path begins and cause middleware or endpoints to see a different path than the server routes. Researchers demonstrated that the bug can turn a forbidden response into a successful one without credentials, enabling authentication bypass and potentially chaining into server side request forgery and in some cases remote code execution against downstream projects, and it affects model serving tools, proxies, gateways and other FastAPI ecosystem components. Maintainers released a patch and advised upgrading to Starlette version 1.0.1 or later, with severity rated between CVSS 6.5 by the maintainer and 7.0 by the researchers, and a public test was published to help detect vulnerable deployments.
MalShark: MCP-Powered Malware Traffic Analysis — Benchmarked Against Real Malware

mohitdabas.in · a month ago

MalShark is an MCP server that wraps tshark to provide AI-accessible malware traffic analysis, allowing a user to drop a pcap, request analysis in plain English, and receive a structured forensic report. Its toolset runs multiple tshark passes in parallel and includes pcapsummary, extractiocs, c2beaconing, finddownloads, extractcredentials, httpsessions, detectdnstunneling, and live capture to detect DNS and TLS SNI indicators, non‑TLS traffic on port 443, and SYN‑only offline C2s. Every detection rule is validated against real malware samples in a blind benchmark process; example results include a May 8, 2026 macOS Shub stealer sample with six of six network IOCs detected and a May 22, 2026 SmartApeSG to NetSupport RAT sample with 87 percent detection coverage, and the project documents fixes such as SYN‑only detection and SNI‑based false positive suppression driven by those samples.
Vulnerability in Popular Conference Software Granted Attackers a 100% Talk Acceptance Rate

www.securityweek.com · a month ago

Researchers at Novee discovered a high severity account takeover vulnerability in the open source call for papers and scheduling platform Pretalx. The flaw tracked as CVE-2026-41241 is a stored cross site scripting issue that allowed registered speakers to embed code that executed in an organizer's browser when submissions were searched, enabling silent account compromise. Many conferences share the same Pretalx codebase so an attacker could submit booby trapped proposals across events and, in practice, achieve a one hundred percent talk acceptance rate, and the issue was patched in Pretalx version 2026.1.0.
FBI’s 2025 Internet Crime Report

www.schneier.com · a month ago

There were 1,008,597 complaints to the Internet Crime Complaint Center in 2025 reporting $20.9 billion in losses, a roughly 26 percent increase from 2024. Cryptocurrency-related scams produced the largest share of losses, with about 181,565 complaints and roughly $11.36 billion reported, while investment scams overall accounted for about $8.6 billion. For the first time the report separated AI-related complaints, documenting about 22,364 AI-nexus complaints and $893 million in losses, and it also identified business email compromise and account takeover as high-cost threats. The report highlighted emerging tactics such as deepfakes, QR code lures, and rising bitcoin ATM fraud that criminals use to move funds quickly and irreversibly.
Can you enforce strong Active Directory password rules without frustrating users?

www.bleepingcomputer.com · a month ago

There are practical ways to enforce strong Active Directory password policies without frustrating users, including prioritizing passphrases by raising minimum length for example to 15 characters, blocking weak and breached passwords, and avoiding frequent mandatory expirations unless a compromise is suspected. Organizations should pair these controls with approved password managers, secure self-service password resets verified with multi factor authentication, clear expiry and lockout notifications, and real time, specific feedback at password creation to reduce helpdesk tickets and risky workarounds. Continuous breached password checking and custom banned word lists stop reused or compromised credentials at creation, while length based aging can reward longer passwords with extended expiry, improving resilience without increasing support burdens.
A week after Dutch FIOD seized 800+ servers, the hosting network's ASN (AS209847) is still scanning at its normal daily rate

ellio.tech · a month ago

Dutch law enforcement FIOD seized more than 800 servers and arrested two operators on May 18, yet scanning from the bulletproof hosting network's ASN ranges has largely continued. About a third of recently active ranges including legacy Stark blocks 94.131.105.0/24 and 92.118.232.0/24 were withdrawn from global routing while surviving ranges under AS209847 WorkTitans THE.Hosting remain announced and continue scanning at the network's normal daily rate. Sibling ASNs AS213999 and Moscow based AS33993 are still routed but currently idle. Recent activity skews toward database and industrial control system discovery including MongoDB Redis PostgreSQL Oracle LDAP and protocols DNP3 and EtherNet IP alongside known exploit probes targeting CVE-2017-17215 and WinRM.
Spencer Pratt Wants to Be Mayor. His Fans Want Him to Be Batman.

www.nytimes.com · a month ago

There is an AI-generated video that casts Spencer Pratt as a Batman-like vigilante in a viral campaign ad portraying Los Angeles as a dystopian landscape and depicting Mayor Karen Bass and other leaders as corrupt figures. The video, created by outside supporters and reposted by Pratt, has racked up millions of views and prompted widespread debate over the use of artificial intelligence in political messaging, with some officials and commentators criticizing the violent imagery while others praise its persuasive creativity. Observers and campaign experts warn that AI-generated ads could rapidly reshape political advertising even as it remains uncertain whether viral AI spectacles will translate into actual votes.
5 Steps to Managing Shadow AI Tools Without Slowing Down Employees

thehackernews.com · a month ago

There are five practical steps to manage shadow AI without slowing employees: build a full inventory of AI tools including OAuth apps, browser extensions, and embedded vendor AI; write a usable AI governance policy specifying approved tools, data classification, and data training opt-out requirements; create a fast approval lane using structured intake and evaluation criteria; and provide browser-native monitoring and just-in-time coaching to make secure choices easy. Across most organizations, employees run three to five AI tools daily that often bypass network controls by accessing data through browser sessions or OAuth, and Gartner reports 69 percent of organizations suspect or have confirmed prohibited AI tool use while only 37 percent have an AI governance policy. When security teams publish an open approved tool list, shorten review turnaround, and give real-time visibility, shadow AI usage typically declines as employees prefer approved alternatives.
Glassworm botnet disrupted after resilient C2 infrastructure takedown

www.bleepingcomputer.com · a month ago

The Glassworm botnet that targeted developers with malicious extensions and software supply chain attacks has been disrupted after researchers deprived it of a resilient command and control infrastructure that used Solana blockchain memo fields, the BitTorrent distributed hash table, a public calendar service, and traditional servers. On May 26, 2026, CrowdStrike, Google, and The Shadowserver Foundation coordinated a takedown that cut off the botnet’s four C2 channels simultaneously, preventing infected machines from receiving new instructions. Following the disruption, compromised hosts are now beaconing to an address operated by CrowdStrike, and organizations are advised to look for that network indicator and use published YARA rules to confirm and remediate infections.
Link11 is fully committed to Europe and is opening a Customer Excellence Hub in Lisbon

gbhackers.com · a month ago

On May 27, 2026, Link11 announced it is opening a Technical Customer Excellence Hub in Lisbon and relocating its technical customer service operations to the European Union. The European cloud based IT security provider, which specializes in network and web application security, said the move is intended to strengthen its commitment to digital sovereignty and centralize technical customer support within Europe. The new hub will serve as a regional center for customer excellence and support Link11's continued expansion across European markets.
SecurityWeek to Host AI Risk Summit August 11-12 at the Ritz-Carlton, Half Moon Bay

www.securityweek.com · a month ago

SecurityWeek will host the 2026 AI Risk Summit on August 11 and 12, 2026, at the Ritz Carlton in Half Moon Bay, California. Now in its third year, the summit will bring together CISOs, security leaders, AI researchers, developers, policymakers, and enterprise risk professionals for keynote presentations, expert panel discussions, fireside chats, technical briefings, and networking focused on securing enterprise AI, adversarial AI threats, deepfakes and misinformation, AI governance and compliance, risk management frameworks, and protecting sensitive data in AI workflows. The call for presentations is open through May 31, 2026, and registration includes a discounted rate of one thousand seven hundred ninety five dollars available until May 31, 2026.
Teachers Union Urges Schools to Curb A.I. Chatbots and Screen Time

www.nytimes.com · a month ago

The American Federation of Teachers recommended that students in second grade or younger have no screens at all in school. The union also urged that A.I. chatbots be barred from use by students in elementary school.
HN Security - AI Reporter - Let's automate reporting in Burp Suite!

hnsecurity.it · a month ago

AI Reporter is a Burp Suite extension that automatically generates structured vulnerability findings from HTTP request and response pairs using Burp AI, a local Ollama instance, or any OpenAI compatible service. The tester right clicks a request, selects Report with AI, provides a vulnerability title severity and confidence, and the extension creates a Burp issue and can export findings as Markdown ready for reports. Features include editable prompts and templates, dynamic tags to inject request data, a lightweight chat tab, support for Community Edition via local models, and the option to avoid sending data to third parties by using a local model, while cloud based AI usage consumes credits and Burp Suite Pro currently includes 10,000 free AI credits.
BadHost Vulnerability Exposes Sensitive AI Agent Server Endpoints to Attackers

gbhackers.com · a month ago

There is a critical vulnerability called BadHost, tracked as CVE-2026-48710, in the Starlette web framework that lets attackers craft malicious Host headers to control request.url and bypass path-based authentication middleware. Thousands of FastAPI and Starlette applications used across AI infrastructure, including vLLM, LiteLLM, Model Context Protocol servers, OpenAI-compatible APIs and various AI agent frameworks, are affected and exploitation does not require authentication. Successful exploitation can expose protected model endpoints, prompt data and API keys, and enable unauthorized use of compute resources or lateral movement within AI environments. Mitigations include upgrading to Starlette version 1.0.1 or later, validating and normalizing Host headers at reverse proxies and application layers, avoiding path-based auth in middleware in favor of route-tied mechanisms or using the ASGI scope path, and deploying RFC-compliant reverse proxies.
Cybersecurity Evolution: How We Went From Perimeter Defense to AI-Native Security

www.darkreading.com · a month ago

The cybersecurity industry has shifted from a narrow focus on antivirus and firewalls in 2006 to a strategic, board-level priority driven by cloud, mobile, the Internet of Things, and now artificial intelligence. Infrastructure changes such as widespread cloud migration and the proliferation of connected devices expanded the attack surface and forced defenses to move from perimeter appliances to software and service-based platforms. Industry spending has moved heavily toward security software and consolidated platforms like SASE, zero trust, and XDR, with analysts projecting software to account for roughly half of security spending by 2026. As AI drives the next wave of innovation, defenders are adopting diverse telemetry, stronger identity and data protection, and platform consolidation while still relying on fundamentals such as patching and proper access management.
RevEng.AI Raises $15 Million to Hunt for Flaws and Backdoors in Software Binaries

www.securityweek.com · a month ago

RevEng.AI raised $15 million in a Series A investment led by the NATO Innovation Fund with participation from Sands Capital, In-Q-Tel, IQ Capital, and Episode One, bringing the startup's total funding to $19.5 million. The London-based company uses an AI model called BinNet to analyze compiled software at the binary level without access to source code, enabling automated detection of vulnerabilities, hidden backdoors, dangerous behaviors, and abnormal release changes. BinNet, reportedly trained with allied government cyber units and commercial security teams, can analyze software in seconds and integrates with existing security and software delivery workflows to verify the integrity of closed-source, internally developed, and third-party software.
FBI warns of in-person data theft attacks from extortion gang

www.bleepingcomputer.com · a month ago

The FBI warns that the Silent Ransom Group, also known as Luna Moth, Chatty Spider, and UNC3753, is targeting U.S. law firms with in-person data theft attacks. Actors pose as IT support via phone calls or phishing to obtain remote desktop access and, if that fails, will travel to a victim site to plug in USB drives or external hard drives to steal data. Stolen data is used to extort victims through ransom emails that threaten to sell or post it on leak sites and by calling employees or clients to pressure victims into paying; indicators include unauthorized external drives and unidentified individuals claiming to be IT support. The group has been active since at least 2022 and has targeted legal and financial organizations in the United States since early 2023.
CISA Warns LiteSpeed cPanel Plugin Vulnerability Is Being Exploited in Attacks

gbhackers.com · a month ago

On May 26, 2026, CISA added a critical LiteSpeed cPanel Plugin vulnerability to its Known Exploited Vulnerabilities catalog and confirmed active exploitation in the wild. The flaw, tracked as CVE-2026-48172, stems from improper privilege management and allows any authenticated cPanel user to execute arbitrary scripts with root level permissions, enabling full server takeover and lateral movement in multi tenant hosting. CISA mandated remediation by May 29, 2026 under Binding Operational Directive 22 01 and organizations are urged to apply vendor patches immediately or disable or restrict the plugin, audit accounts, monitor logs, and enforce strict access controls to reduce exposure.
GlassWorm Malware Takedown Disrupts Developer Supply Chain Attack Infrastructure

thehackernews.com · a month ago

CrowdStrike, working with Google and the Shadowserver Foundation, simultaneously disrupted all command and control channels used by GlassWorm, a persistent software supply chain campaign that has targeted software developers since at least early 2025. The campaign delivered trojanized Visual Studio Code extensions and compromised npm and Python packages to install a data theft framework that harvests developer credentials and crypto wallets, deploys a WebSocket based JavaScript RAT, installs malicious Chrome extensions, and converts infected hosts into SOCKS proxies, hidden VNC servers, and remote execution nodes. GlassWorm relied on four resilient C2 layers including Solana blockchain memo fields, the BitTorrent DHT, Google Calendar event titles, and commercial VPS hosts, and the coordinated takedown neutralized those channels so infected machines can no longer receive new instructions, with CrowdStrike attributing the operation to likely Russia based actors.
3 SOC Steps that Shut Down Incident Risks Early

thehackernews.com · a month ago

Organizations are shifting SOC strategy from building stronger perimeter defenses to reducing operational uncertainty by shrinking the time between a change and a clear understanding of its meaning, because attackers often enter disguised as routine activity and accumulate risk before being labeled an incident. Three core steps to shut down incident risk early are continuously updating monitoring and threat intelligence feeds to spot emerging IOCs, enriching alerts with complete triage context so analysts receive investigation ready data, and producing automated response ready reports from sandbox analysis to eliminate investigation bottlenecks. Implementing these steps lowers attacker dwell time, reduces false positives, speeds triage and remediation, and cuts the probability of operational disruption, ransomware escalation, compliance failures, supply chain propagation, and expensive incident recovery. The page also references CVE-2026-45585, CVE-2026-31635 and CVE-2026-42945.
Romanian Hacker Sentenced to Prison in US for Selling Access to State Network

www.securityweek.com · a month ago

A Romanian national named Catalin Dragomir was sentenced in the United States to four years and eight months in prison after admitting to selling access to an Oregon state government office network and pleading guilty to obtaining information from a protected computer and aggravated identity theft. He was arrested in Romania in November 2024, extradited to the United States in January 2025, and pleaded guilty in February 2026, with the judge crediting two months he spent in Romanian custody at sentencing. Prosecutors say he hacked the Oregon state network in June 2021, sold access for three thousand dollars in Bitcoin, admitted selling information from at least ten other organizations, and caused losses exceeding two hundred fifty thousand dollars.
BTMOB Malware Allows Cybercriminals to Remotely Hijack Android Phones

gbhackers.com · a month ago

A newly observed Android malware strain named BTMOB, first identified in early 2025, has evolved into a full featured remote access trojan that grants attackers near complete control over infected devices. Researchers say the malware combines powerful remote access capabilities with ease of deployment, enabling cybercriminals to remotely hijack phones rather than only stealing banking credentials. Security teams warn of rapid evolution across variants and recommend monitoring Android devices for indicators of compromise and applying mobile security best practices.
Lastwall Raises $11.5 Million for Quantum-Resilient Identity Platform

www.securityweek.com · a month ago

Lastwall has raised 11.5 million dollars in a Series A extension led by BDC Capital’s StrongNorth Fund, with participation from NBIF, Frostbite Capital, Blue Bear Capital, 18West, and Blue Wing Ventures, to accelerate its North American expansion. The company provides secure identity management for defense and government organizations and offers the IDCommand Suite, a unified platform that combines zero trust architecture and public key infrastructure and supports passwordless authentication in three configurations: Enterprise, a cloud native identity provider; Tactical, a field deployable on premises solution for denied degraded intermittent and low bandwidth environments; and BOLT, a biometric add on that delivers AAL 3 assurance. The IDCommand Suite evaluates about 200 contextual signals including keystroke cadence, mouse movement, and hardware profiling to assign a risk score to each login and trigger additional authentication when needed. Lastwall also offers Quantum Shield, a quantum safe TLS terminator and load balancer available through the AWS Marketplace that applies NIST approved post quantum cryptography while maintaining backward compatibility and updating as PQC standards evolve, and the company has earned FedRAMP Moderate authorization for U.S. government use.
The Credential Crisis: How Stolen Credentials Defeat Modern Security

www.securityweek.com · a month ago

There is a growing credential crisis as AI driven phishing, deepfakes, session hijacking, infostealers, and social engineering accelerate credential theft and enable attackers to operate with legitimate user access. Credentials now span human and machine identities including passwords, passkeys, biometrics, tokens, API keys, certificates, and session cookies, and stolen sessions or tokens can preserve access even after password changes. Security teams are advised to combine breach and dark web monitoring, infostealer log intelligence, identity security posture management, and phishing resistant MFA such as WebAuthn and passkeys to detect compromise and reduce risk, but detection remains difficult and MFA bypass rates are rising.
Threat Intel: Lithuania Investigates B2B Credential Misuse Exposing 600,000 National Registry Records

www.technadu.com · a month ago

There was a large scale data exfiltration from Lithuania's State Enterprise Centre of Registers in which more than 600,000 records were copied from the national Real Estate and Legal Entities Registers. Forensic tracking indicates the attackers used valid high privilege inter agency B2B login credentials, reportedly those of the Department of Migration under the Ministry of the Interior, to issue large backend queries and bypass perimeter defenses. Exfiltrated data include full legal names, dates of birth, national identification numbers, registered physical addresses, corporate entity structures, and cadastral and property registry extracts, while telephone numbers, email addresses, bank account numbers, and raw cadastral measurement files were not included. The breach was detected in early April 2026, prompted revocation of compromised accounts, mandatory credential rotation and query throttling, and led to the resignation of the Centre director on May 25, 2026, with official attribution still open and analysts warning the dataset could be used for profiling, spear phishing, or coercion.
Windows Kernel Vulnerability Lets Attackers Modify Kernel Memory Counters

gbhackers.com · a month ago

CVE-2026-40369 is an untrusted pointer dereference in the Windows kernel component ntoskrnl.exe that can be triggered by calling NtQuerySystemInformation with information class 253 and allows arbitrary kernel-address increments. Any unprivileged process, including browser renderer sandboxes, can reach the bug with a single NtQuerySystemInformation call and reliably escalate privileges to SYSTEM on affected builds. Affected releases include Windows 11 versions 24H2, 25H2 and 26H1 and Windows Server 2025, and Microsoft has published fixes for specific build numbers so applying vendor updates is required to mitigate the issue. The vulnerability is tracked as CVE-2026-40369 with a CVSS v3.1 base score of 7.8 and is classified as CWE-822 untrusted pointer dereference.
‘SymJack’ Attack Turns AI Coding Agents Into Supply Chain Attack Delivery Systems

www.securityweek.com · a month ago

SymJack is an attack that uses malicious repositories and disguised symlinks to trick AI coding agents into silently installing attacker controlled MCP servers that can steal secrets, compromise continuous integration pipelines, and deploy malicious code. The chain requires attacker control of the coding agent repository, a ready made malicious MCP server, and a developer using an AI coding tool, and works by renaming a symlink to appear innocuous so a copy command can insert the payload into the agent configuration where a startup command runs it unsandboxed. Adversa’s proof of concept worked across five major coding agents including Claude Code, Gemini CLI and Antigravity CLI, Cursor Agent CLI, Grok Build CLI, and GitHub Copilot CLI, and a single malicious pull request against CI runners can exfiltrate all necessary secrets before any human review. Mitigations include resolving symlinks and showing real destination paths before asking for approval and refusing unexpected copy operations to prevent registration and spawning of a malicious MCP.
GlassWorm Botnet Disrupted

www.securityweek.com · a month ago

The GlassWorm botnet that targeted the open source software ecosystem for over six months has been disrupted after CrowdStrike, Google and the Shadowserver Foundation simultaneously took down the malware's four command and control channels. GlassWorm used the Solana blockchain, Google Calendar, the BitTorrent peer to peer network and traditional VPS servers as layered command and control infrastructure, encoding addresses in blockchain memo fields, storing Base64 encoded command and control paths in calendar event titles, and keeping configuration data on BitTorrent to resist takedowns. Initially spread via trojanized Visual Studio extensions on the OpenVSX marketplace and later observed on GitHub, the malware steals developer credentials and cryptocurrency extension funds, deploys SOCKS proxies and hidden VNC servers, and poses a supply chain risk to downstream consumers of affected software. Evidence points to operators of likely Russian origin who avoided infecting machines in CIS countries, and organizations are advised to check for connections to IP address 164.92.88.210 to identify potential infections.
GitHub Enterprise Server 3.20.3 Addresses Critical Security Flaws

gbhackers.com · a month ago

GitHub released Enterprise Server version 3.20.3 on May 26, 2026, to address multiple critical and high severity vulnerabilities that could allow attackers to access internal services, escalate privileges, or extract sensitive data. The most severe is a pre authentication server side request forgery tracked as CVE-2026-9312 in an upload endpoint; other addressed vulnerabilities include CVE-2026-43284 and CVE-2026-43500, Dirty Frag Linux kernel issues allowing local privilege escalation, and CVE-2026-8606, a timing side channel combined with SSRF affecting GitHub Packages. GitHub revoked the existing release signing key and requires administrators to rotate GPG public keys before applying the patch, providing a script to assist with the transition. The release also removes the vulnerable package URL endpoint, increases the OpenTelemetry collector memory to 4096 MiB, fixes several bugs, and warns administrators to reapply custom firewall rules and check for locked root accounts after upgrades while treating the update as high priority.
Gitea Vulnerability Exposes Private Container Images without Authentication

thehackernews.com · a month ago

A vulnerability in Gitea, an open source self hosted platform for version control, allows unauthenticated remote attackers to pull private container images from Gitea deployments without requiring an account password or other credentials. The flaw is tracked as CVE-2026-27771 and affects all versions of Gitea prior to 1.26.2. Administrators are urged to upgrade to version 1.26.2 to prevent unauthorized access to private container images.
CISA gives feds 4 days to patch actively exploited cPanel plugin flaw

www.bleepingcomputer.com · a month ago

CISA ordered U.S. federal agencies to patch a critical, actively exploited privilege escalation flaw in the LiteSpeed cPanel user-end plugin within four days, setting a deadline of midnight on May 29, 2026 under Binding Operational Directive 22-01. The flaw, tracked as CVE-2026-48172, stems from incorrect privilege assignment in the lsws.redisAble function that mishandles Redis enable and disable features and allows remote attackers with no privileges to execute arbitrary scripts as root. LiteSpeed released urgent updates and warned that plugin versions from 2.3 through 2.4.4 are at risk, and organizations were urged to apply vendor mitigations immediately, examine logs for unauthorized activity, and prioritize patching.
LA Metro Cyberattack Linked to Iranian State-Sponsored Hackers

www.securityweek.com · a month ago

The Los Angeles County Metropolitan Transportation Authority discovered a breach in mid March that disrupted internal operations and required hundreds of servers to be inspected, while rail and bus services remained unaffected. The attack was claimed by a pro Iran hacktivist group called Ababil of Minab, which allegedly wiped hundreds of terabytes of data, exfiltrated more than one terabyte of files, and published screenshots and videos showing access to a core virtualization management platform, a Microsoft IIS web server, and an operational technology system used to monitor trains. Forensic analysis by security firms linked the operation to infrastructure previously used by Iran linked actors including Black Shadow and reported similar exfiltration and destructive activity against organizations in the United States, Israel, Saudi Arabia, and Turkey.
Dutch police arrests suspect linked to Ajax football club hack

www.bleepingcomputer.com · a month ago

Dutch National Police arrested a 35-year-old man in Buren on May 26, 2026, suspected of repeatedly intruding into AFC Ajax computer systems earlier in 2026. The intrusions exploited vulnerabilities that granted access to data for a few hundred individuals and allowed modification of stadium bans and transfer of purchased tickets for fewer than twenty people. Investigators and reports demonstrated the flaw could reassign VIP season tickets in seconds, manipulate 538 supporter stadium bans, affect 42,000 season tickets, and expose details on more than 300,000 accounts. Ajax has patched the exploited vulnerabilities and notified the Dutch Data Protection Authority and police, and the suspect remains in custody while the investigation continues.
The NSA, ‘Mythos’ and the quiet emergence of AI cyber doctrine

www.csoonline.com · a month ago

There is a nascent AI driven cyber doctrine emerging as frontier models enable autonomous reconnaissance, vulnerability discovery and multi stage exploit chaining at machine speed and scale, giving attackers compressed time, cost and expertise advantages while defenders remain indexed to human decision cycles. Anthropic's Mythos preview reportedly surfaced thousands of high severity findings and autonomously chained exploits, including a 17 year old FreeBSD NFS remote code execution tracked as CVE-2026-4747, and spurred a coordinated defensive coalition of major technology and infrastructure organizations with substantial usage credits and donations. Security leaders are urged to treat AI agents as security principals with identity and access controls, invest in adaptive AI driven detection and response rather than static signatures, and reframe risk assumptions to expect continuous low level compromise and optimize for detection, containment and bounded loss.
RCE in Strix Agent(Sandbox): A practical guide to prompt injections with impact

baldur.dk · a month ago

There is a demonstrated remote code execution technique against the Strix AI pentester that leverages prompt injection to coerce the agent into executing attacker controlled shell commands. An attacker can inject crafted content such as a fake JavaScript blob or comment, lure Strix to fetch an attacker hosted script, then exploit a time of check to time of use swap so the agent first validates a benign payload and later retrieves and executes a malicious reverse shell, achieving code execution as the pentester user inside the scan container. Mitigations discussed include sandboxing, network restrictions, and strict guardrails, but the information warns that prompt injection is a fundamental architectural risk for LLM agent systems and that containerization reduces but does not eliminate potential impact.
Windows 11 KB5089573 update released with performance improvements

www.bleepingcomputer.com · a month ago

There is a preview cumulative update KB5089573 for Windows 11 versions 25H2 and 24H2 that delivers 30 changes focused on performance and reliability improvements. The update is an optional non-security monthly preview release intended for testing before Patch Tuesday and raises affected systems to builds 26200.8524 and 26100.8524. The update accelerates app launch and core shell experiences such as Start menu, Search, and Action Center, improves Windows Hello sign-in behavior to favor face or fingerprint when available, and fixes reliability issues in File Explorer, sign-in and lock screens, Settings theme changes, touch gestures, and Modern Standby resume. Additional fixes improve Task Manager CPU reporting for virtual machines after hibernation, increase sensor and HID power resilience to prevent battery drain, and add high-confidence device targeting for the phased rollout of updated Secure Boot certificates ahead of their late June expiry.
FBI: Hackers Sending Operatives in Person to Insert USB Drives and Steal Data

www.securityweek.com · a month ago

The Silent Ransom Group has been impersonating IT support to gain remote access through callback phishing and social engineering, and when remote access fails they are sending operatives in person to insert USB or external drives into target computers to image devices or create backups and steal data. After gaining access they escalate privileges and exfiltrate data using legitimate tools such as WinSCP and variants of Rclone or by copying to internal file shares and cloud drives including Google Drive and Microsoft OneDrive, and then extort victims without deploying file encrypting ransomware. Recommended defenses include verifying credentials for anyone requesting access, restricting remote access and external drive installation, implementing phishing resistant multi factor authentication, blocking commonly exploited ports, maintaining backups, and training employees to recognize phishing and social engineering.
AI Chatbot Recommendations Redirect Users to Cryptojacking Malware Sites

thehackernews.com · a month ago

Microsoft warned of an active cryptojacking campaign that leverages AI chatbot interactions to surface attacker controlled download sites, with more than 150 malicious domains identified. The operation impersonates legitimate system utilities to target systems with high performance GPUs, delivering ZIP archives that sideload a rogue DLL to install a remote access client and then use a hollowed trusted binary to deploy miners including gminer, lolMiner, and SRBMiner MULTI while establishing persistence and configuring Defender exclusions. Threat actors also abused compromised edge appliances and follow on techniques to pivot internally, including exploitation tied to CVE-2025-33073, prompting advice to validate vendor behavior, monitor for unauthorized remote access deployments, and harden defenses against AI assisted social engineering.
New Zero-Click WhatsApp Account Takeover Attack Targets iOS 16 Users

gbhackers.com · a month ago

There is a newly uncovered zero-click attack against iPhones running iOS 16 that hijacks WhatsApp accounts without any user interaction or visible unknown linked-device prompts. The exploit chain leverages CVE-2025-43300, an out-of-bounds write in Apple ImageIO, and CVE-2025-55177, an authorization flaw in WhatsApp’s linked-device synchronization, to extract session cryptographic material and silently attach a second client that sends fraudulent payment requests to recent contacts. Mitigations include updating iOS and WhatsApp to patched versions, reinstalling or moving the account to a fresh device to evict rogue sessions, enabling WhatsApp chat lock, and verifying money requests by calling contacts rather than replying in the same chat.
Hackers Exploit Shared CDN Edge IPs to Evade Protective DNS Filtering

gbhackers.com · a month ago

Attackers are exploiting shared CDN edge IPs to bypass DNS based security controls by resolving trusted domains and then initiating TLS connections to the same IP with a different hostname in the Server Name Indication or HTTP Host header, a technique known as Underminr. Underminr differs from legacy domain fronting and supports multiple modes including simple, split, ECH and direct to IP, enabling covert command and control, data exfiltration and policy evasion while blending into legitimate CDN traffic. Defenders are advised to correlate DNS telemetry with network and application layer signals, monitor for mismatches between DNS resolution and TLS session parameters, restrict direct IP egress and deploy advanced traffic analysis, and ADAMnetworks has published research, an impact report and an online assessment tool to help organizations identify exposure.
DSPM buyer’s guide: Top 10 data security posture management tools

www.csoonline.com · a month ago

Data security posture management tools help security teams examine their entire data environment to find shadow data across cloud and on premises systems, classify and map sensitive data, and reduce the risk of data loss or exposure. The DSPM market is rapidly consolidating, with major vendors acquiring niche providers to expand discovery, classification, and protection capabilities, and research firms position DSPM as bridging the gap between discovery and automated remediation. Core capabilities to evaluate include agentless and agent collectors, metadata based scanning, prebuilt and AI driven classifiers, data lineage mapping, continuous scanning and auditing, and broad integrations with SIEM, SOAR, CNAPP, DLP, IAM, and other security and data tools for remediation. Pricing is substantial and variable, typically starting around thirty thousand dollars for small deployments and often exceeding one hundred thousand dollars annually, with some vendors charging per terabyte or offering named starting prices for specific packages.
CISA Urges Immediate Patching of Exploited LiteSpeed cPanel Plugin Zero-Day

www.securityweek.com · a month ago

There is a critical privilege escalation vulnerability in the LiteSpeed user-end plugin for cPanel tracked as CVE-2026-48172 with a CVSS score of 9.8 that has been exploited in the wild to allow execution of arbitrary scripts with root privileges. CISA added CVE-2026-48172 to its Known Exploited Vulnerabilities catalog and urged federal agencies to patch or remove vulnerable user-end plugin versions between v2.3 and v2.4.4 by May 29, 2026, while cPanel pushed an update that removed the user-end plugin. Mitigations recommended include upgrading to LiteSpeed WHM Plugin version 5.3.1.0 or later, which bundles user-end plugin version 2.4.7, or completely removing the plugin if patching is not possible.