Blahalla

54 EDR Killers Use BYOVD to Exploit 35 Signed Vulnerable Drivers and Disable Security

thehackernews.com · 7 months ago

Fifty four EDR killer tools were found to leverage the bring your own vulnerable driver technique by abusing thirty four signed vulnerable drivers to gain kernel mode privileges and disable endpoint protections. These tools are commonly used in ransomware intrusions to neutralize security software immediately before deploying file encryptors, and operators include closed ransomware groups, attackers who fork proof of concept code, and criminals selling the utilities on underground marketplaces. Recommended defenses include blocking commonly misused drivers, implementing layered detection and response, and proactively monitoring to flag, contain, and remediate attempts to disable security controls.
Ransomware group exploited Cisco firewall vulnerability as a zero day, weeks before a patch appeared

www.csoonline.com · 7 months ago

Interlock began exploiting a critical Cisco Secure Firewall Management Center deserialization flaw CVE-2026-20131 on January 26, 2026, roughly 38 days before Cisco released a patch on March 4, 2026. CVE-2026-20131 carries a maximum CVSS score of 10, and Cisco also patched a separate FMC authentication bypass vulnerability CVE-2026-20079 in the same update. AWS honeypots captured the attack chain, including a malicious ELF binary and a single poorly secured staging server, enabling attribution to the Interlock ransomware family based on embedded ransom notes and TOR negotiation infrastructure. Organizations using affected FMC versions are likely at risk and are advised to apply the March 4, 2026 updates appropriate to their software version and to employ defense in depth controls.
A.I. Bots Can Act as Personal Digital Assistants, but There Are Serious Risks

www.nytimes.com · 7 months ago

New A.I. bots can do more than just chat. They can edit files, send emails and book trips for users, and those capabilities can also enable harmful or unintended actions when agents are misused or make mistakes.
Post-Quantum Web Could be Safer, Faster

www.darkreading.com · 7 months ago

NULL
1stProtect Emerges From Stealth With $20 Million in Funding

www.securityweek.com · 7 months ago

false
we found a memory exhaustion CVE in a library downloaded 29 million times a month. AWS, DataHub, and Lightning AI are in the blast radius.

www.periphery.security · 7 months ago

A memory exhaustion vulnerability in DeepDiff allows a 42-byte pickle payload to trigger multi-gigabyte allocations during deserialisation, crashing processes and enabling persistent crash loops via queues or uploaded artifacts. This issue is tracked as CVE-2026-33155 and follows a prior fix for CVE-2025-58367 that hardened the restricted unpickler but did not validate constructor arguments, creating a false sense of security for teams that believed they were protected. Immediate remediation is to update DeepDiff to version 8.6.2 or later, pin downstream packages, and ensure no untrusted data can reach pickleload or the Delta constructor.
Critical ScreenConnect Vulnerability Exposes Machine Keys

www.securityweek.com · 7 months ago

There is a critical vulnerability identified as CVE-2026-3564 with a CVSS score of 9.0 that could allow attackers to access cryptographic material used for session authentication. Previously, unique ScreenConnect machine keys were stored in server configuration files and could be exfiltrated, and ScreenConnect version 26.1 adds encrypted storage and management of machine key material to prevent unauthorized access and reduce the risk of server compromise. Users are advised to update to version 26.1 immediately, review and restrict access to configuration files and backups, and monitor logs for unusual activity; ConnectWise says it has no evidence to validate claims that the flaw was exploited by Chinese state-sponsored actors.
Privacy Platform Cloaked Raises $375M to Expand Enterprise Reach

www.securityweek.com · 7 months ago

Cloaked raised 375 million dollars in a Series B and growth financing to accelerate its product roadmap and scale for consumers and businesses, bringing total funding to more than 400 million dollars. The platform generates unlimited unique virtual identities such as phone numbers, email addresses, and passwords, runs an automated data removal tool that continuously scans and deletes personal information from data brokers and people search sites, and offers an AI powered Call Guard that screens incoming calls to block spam, scams, and fraud. The company plans to use the capital to introduce AI driven personal agents that act on users' behalf to monitor, manage, and enforce privacy preferences and security posture, and to expand Cloaked Enterprise, a Digital Workforce Protection program for identifying and mitigating cybersecurity risks from employees' personal digital exposure. Cloaked said it has disguised more than 10 million identities, removed over a billion records from data broker sites, and screened more than 40 million calls.
Bitrefill blames North Korean Lazarus group for cyberattack

www.bleepingcomputer.com · 7 months ago

Bitrefill says the cyberattack at the start of March, with technical issues first reported on March 1, 2026, was likely carried out by North Korean Lazarus group operators associated with the Bluenoroff cluster. Attackers compromised an employee laptop, stole legacy credentials, accessed a snapshot containing production secrets, escalated to parts of the database and some cryptocurrency wallets, and exploited supplier purchasing and gift card supply lines. About 18,500 purchase records containing customer email addresses, IP addresses, and cryptocurrency payment addresses were exposed and about 1,000 purchases also included customer names, and Bitrefill warns attackers may have obtained decryption keys. Bitrefill says losses were minimal and will be covered from its capital, that attackers appear to have been after crypto and gift card inventory rather than customer data, and that it is expanding security reviews, pen testing, access controls, logging, monitoring, and automated shutdown mechanisms while most services have returned to normal and customers should treat incoming communications with extra caution.
Sandboxing AI coding agents with kernel-level enforcement: built-in profiles for Claude Code, Codex, and OpenCode

nono.sh · 7 months ago

A new open source sandbox called nono enforces kernel level filesystem and command policies for AI coding agents using Landlock on Linux and Seatbelt on macOS so the agent process cannot bypass protections at the application layer. It includes ready to use profiles for Claude Code, Codex, and OpenCode and applies a default policy that blocks common credential stores such as SSH credentials, AWS credentials, and Kubernetes config, along with browser cookies, shell history, and destructive commands like rm, dd, sudo, and shutdown. A learn mode traces an agent's actual file and command accesses and generates a tailored policy profile listing the exact paths to permit.
Beijing wants its own quantum-resistant encryption standards rather than adopt NIST’s

www.csoonline.com · 7 months ago

China plans to develop national post-quantum cryptography standards within the next three years, prioritizing finance and energy, and favors structureless lattice algorithms over the algebraic lattice designs adopted by most other countries. Leading cryptographer Wang Xiaoyun and other experts warn that algebraic or structured lattices may contain exploitable patterns, and China launched a global call for post quantum algorithm proposals in February 2025. The United States and partners published NIST post quantum standards in 2024 and selected a code based backup algorithm in March 2025, with migration timelines set between 2030 and 2035. Experts say the resulting five year gap creates a difficult compliance choice for organizations operating in China and advise immediate hybrid deployments that layer NIST approved post quantum algorithms alongside existing cryptography to reduce harvest now decrypt later risk and preserve the ability to swap algorithms as standards evolve.
FBI seizes Handala data leak site after Stryker cyberattack

www.bleepingcomputer.com · 7 months ago

Federal authorities seized two websites used by the Handala hacktivist group after a destructive cyberattack on medical technology giant Stryker that wiped approximately 80,000 devices. Handala is Iranian linked and reportedly tied to Iran's Ministry of Intelligence and Security, and investigators say the group compromised a Windows domain administrator account, created a new global administrator account, and issued a Microsoft Intune wipe command that factory reset company devices and some employees' personal devices. Handala acknowledged the seizures and said it will rebuild infrastructure while Microsoft and the United States Cybersecurity and Infrastructure Security Agency released guidance on hardening Windows domains and securing Intune to prevent similar attacks.
Exploiting a PHP Object Injection in Profile Builder Pro in the era of AI

blog.sicuranext.com · 7 months ago

There is an unauthenticated PHP Object Injection in Profile Builder Pro affecting versions 3.14.5 and below that accepts attacker-controlled serialized data in the args POST parameter for the wppbrequestuserspins AJAX action and was fixed in version 3.14.6. Researchers chained available gadgets to exploit the flaw, using a novel Monolog POP chain in The Events Calendar plugin and a Guzzle file-write primitive in another plugin to deploy a web shell and achieve remote code execution. AI assistance, specifically Claude Code with Opus 4.6, was used to triage exposed hooks and help identify usable gadget chains, accelerating the vulnerability analysis. Scanning of roughly 11,700 reachable sites found about 1,590 vulnerable instances and the report notes a related Monolog gadget case tracked as CVE-2024-8016.
Iran Readied Cyberattack Capabilities for Response Prior to Epic Fury

www.securityweek.com · 7 months ago

There was a six month buildup of Iran linked cyber infrastructure, including US based shell companies and multi tier hosting arrangements designed to weather kinetic strikes and sustain global hacking operations. Multiple Iran linked APTs showed increased infrastructure activity in the six months before the February 28, 2026 US and Israeli strikes, and around sixty Iran linked hacktivist groups established coordinated operations within 24 hours to target US, Israeli, and Gulf entities. Examples of the staged infrastructure include bulletproof hosts and shell firms such as RouterHosting LLC, Cloudblast and UltaHost, ICANN issued a breach notice against UltaHost Inc on February 5, 2025, and MuddyWater had multiple CIDRs flagged in mid September 2025 consistent with pre operational staging.
The Long Farewell to Mark Zuckerberg’s Metaverse

www.nytimes.com · 7 months ago

false
The Long Farewell to Mark Zuckerberg’s Metaverse

www.nytimes.com · 7 months ago

false
Marquis Data Breach Affects 672,000 Individuals

www.securityweek.com · 7 months ago

There are roughly 672,000 individuals affected by a Marquis data breach after the Texas based marketing and compliance provider discovered in August 2025 that attackers had accessed its systems. Personal information stolen included names, addresses, Social Security numbers, dates of birth, taxpayer identification numbers, and financial information including payment card numbers that Marquis stored on behalf of dozens of the roughly 700 banks and credit unions it serves. Earlier estimates based on state filings and disclosures by affected financial institutions ranged from at least 780,000 to as many as 1.6 million potentially impacted people, and some reported counts may overlap when customers hold accounts at multiple institutions. The company reported the attackers exploited a SonicWall firewall vulnerability, a timeline that matched increased exploitation by the Akira ransomware group, and there are unconfirmed reports that Marquis paid a ransom.
[Virtual Event] Shields Up: Key Technologies Reshaping Cybersecurity Defenses

www.darkreading.com · 7 months ago

false
When tax season becomes cyberattack season: Phishing and malware campaigns using tax-related lures

www.microsoft.com · 7 months ago

There is an observable uptick in tax themed phishing and malware campaigns as the United States tax filing deadline approaches, with attackers using W‑2 and Form 1099 lures, refund notices, CPA impersonations, and other tax related messages to trick individuals and accounting professionals into opening malicious attachments, scanning QR codes, or following multi step link chains. Threat actors are using phishing as a service platforms such as Energy365 and SneakyLog to harvest credentials and bypass multifactor authentication, and some campaigns deliver remote monitoring and management tools including ScreenConnect to maintain persistence or enable hands on keyboard activity. Observed campaigns in January and February 2026 used personalized Excel and OneNote attachments hosted on legitimate services like OneDrive, QR codes that embed recipient specific URLs, and sets of tax themed domains registered to impersonate firms and agencies. Defensive guidance emphasizes user education, configuring email security controls, and hunting for indicators of compromise related to these tax themed lures.
Russian hackers exploit Zimbra flaw in Ukrainian govt attacks

www.bleepingcomputer.com · 7 months ago

APT28, a threat group linked to Russia's GRU, is exploiting a Zimbra Collaboration Suite stored cross site scripting vulnerability tracked as CVE-2025-66376 that can lead to remote code execution and full compromise of Zimbra servers and user mailboxes. The campaign, named Operation GhostMail, used phishing emails with no malicious attachments or links but with obfuscated JavaScript embedded in the HTML body that runs in vulnerable Zimbra webmail sessions to harvest credentials, session tokens, backup two factor codes, browser saved passwords, and up to 90 days of mailbox contents for exfiltration over DNS and HTTPS. CISA added CVE-2025-66376 to its catalog of known exploited vulnerabilities and ordered federal agencies to secure affected servers within two weeks, while researchers reported the attacks targeted Ukrainian government entities including the State Hydrology Agency.
Telnet vulnerability opens door to remote code execution as root

www.csoonline.com · 7 months ago

A critical Telnet vulnerability with a CVSS score of 9.8 allows unauthenticated remote code execution as root against systems running GNU inetutils telnetd and is tracked as CVE-2026-32746. The flaw stems from an out of bounds write caused by a buffer overflow in the telnetd LINEMODE Set Local Characters handler and can be triggered during protocol negotiation before authentication, enabling arbitrary code execution after a connection to TCP port 23 is established. Affected deployments include legacy infrastructure, networking equipment, embedded and IoT devices, and Linux distributions that ship inetutils or enable telnetd; vendors were notified on March 11, 2026 and a patch was planned for release by April 1, 2026. Immediate mitigations include disabling telnetd, migrating to SSH, running telnetd without root privileges where possible, and blocking or restricting access to port 23 at the network perimeter.
ThreatsDay Bulletin: FortiGate RaaS, Citrix Exploits, MCP Abuse, LiveChat Phish & More

thehackernews.com · 7 months ago

An emerging Ransomware-as-a-Service operation called The Gentlemen is exploiting a critical FortiOS authentication bypass tracked as CVE-2024-55591 to control roughly 14,700 exploited FortiGate devices and nearly 969 validated brute-forced VPN credentials while using bring your own vulnerable driver techniques to evade defenses. Security researchers reported a pre authentication remote code execution chain in BMC FootPrints involving CVE-2025-71257 CVE-2025-71258 CVE-2025-71259 and CVE-2025-71260, and noted a surge of Citrix NetScaler exploit attempts targeting CVE-2025-5777 and CVE-2023-4966. Meanwhile malware loaders such as Hijack Loader are deploying new C2 frameworks and updated stealers, and phishing campaigns via Microsoft Teams and LiveChat are increasingly used to gain remote access or harvest credentials and payment data.
Security Firm Aura Discloses Data Breach Impacting 900,000 Records

www.securityweek.com · 7 months ago

Online safety platform Aura experienced a data breach after an employee fell for a targeted phone phishing attack that gave attackers access to the employee's account for about an hour. Attackers accessed roughly 900,000 records stored in a marketing tool acquired in 2021, mostly names and email addresses, and the compromised data includes names, email addresses, addresses, and phone numbers for about 20,000 current and 15,000 former customers. No Social Security numbers, passwords, or financial information were exposed, and sensitive customer information is stored encrypted and protected by restricted access. Aura terminated the account access, activated its incident response plan, engaged external cybersecurity and legal experts, notified law enforcement, and began notifying impacted customers while assessing risk.
CISO Whisperer Names 11 Vendors Leading the Shift from Tools to Outcomes at RSA Conference 2026

gbhackers.com · 7 months ago

There is a move toward continuous intelligence driven operations at RSA Conference 2026, with CISOs prioritizing outcome based security, AI driven automation, and integrated platforms as attack surfaces expand. Eleven vendors were highlighted for exemplifying this shift, including Daylight Security with outcomes as a service and agentic AI, Reclaim Security with an AI Security Engineer that remediates vulnerabilities, and CyCognito for attacker centric external asset discovery. Established vendors including Cisco, Splunk, Palo Alto Networks, Fortinet, Nokia, Zscaler, CrowdStrike, and Check Point are emphasizing platform consolidation, identity first models, agentic SOC workflows, and hands on demonstrations to help organizations separate signal from noise.
7 Ways to Prevent Privilege Escalation via Password Resets

www.bleepingcomputer.com · 7 months ago

There is a high risk of privilege escalation when password reset workflows are weaker than login controls, because attackers who compromise low-privilege accounts, intercept reset tokens, or socially engineer helpdesk staff can reset credentials for higher-value accounts. There are seven recommended mitigations: require multi-factor authentication with phishing-resistant methods for high-value accounts, limit resets to trusted managed devices and apply device posture checks, enforce strong password policies and block breached passwords, educate users and helpdesk teams on identity verification, audit and monitor reset activity, implement least privilege and time-bound access, and avoid knowledge-based authentication in favor of possession-based checks. Organizations should scope who can reset others' passwords, log and alert on unusual reset patterns such as repeated or out-of-hours requests, and remove overly broad reset permissions that enable lateral movement. The guidance stresses strengthening the full account recovery lifecycle to reduce escalation risk while preserving user experience.
A Copy-Paste Bug That Broke PSpice® AES-256 Encryption

jtsylve.blog · 7 months ago

There is a copy-paste bug in PSpice mode four key derivation that causes the short derived key to be passed to the AES engine instead of the intended extended key, so 28 of the 32 AES-256 key bytes are known or zero. This reduces the effective keyspace to two to the thirty-second power, about 4.3 billion candidates, enabling brute-force recovery in seconds on modern hardware with AES-NI or a GPU. A fixed header prefix provides a known-plaintext crib to validate keys, and recovering the four-byte portion allows reconstruction of the full user key and decryption of all files protected with that key. The weakness likely dates to the April 2014 introduction of AES-256 support and fixing it now would break compatibility with many years of encrypted models.
OpenSIPS SQL Injection to Authentication Bypass (CVE-2026-25554)

aisle.com · 7 months ago

There is a high severity SQL injection vulnerability in OpenSIPS authjwt that enables a full authentication bypass and is tracked as CVE-2026-25554 with a severity score of 8.3. The flaw occurs because the JWT tag claim is decoded without verification and interpolated into a raw SQL query without escaping, allowing an attacker to inject a UNION SELECT that returns attacker-controlled secrets so the server will validate forged JWTs and impersonate arbitrary SIP users. All OpenSIPS versions 3.1 and later using authjwt with database mode enabled are affected, exposing risks including unauthorized registration, call placement, toll fraud, and wiretapping, and an upstream fix implemented on February 2, 2026 escapes the tag value to prevent injection.
Hacker Conversations: Ben Harris, From Unintentional Young Hacker to Intentional Adult CEO

www.securityweek.com · 7 months ago

Ben Harris is a British security practitioner who moved from youthful school network tinkering to a professional pentester and security executive, holding roles at Portcullis, MWR Infosecurity, and F-Secure before founding and becoming CEO of WatchTowr in August 2021. His hacking was driven by curiosity about how computers communicate rather than malice, and he says he has a moral boundary against causing damage or selling discovered weaknesses to criminals or foreign governments. WatchTowr provides a platform that builds a real time attack view of customers environments and continuously identifies and validates exploitable vulnerabilities so they can be addressed before being leveraged.
New Apple Hack: Up to 270M iPhones Vulnerable to ‘DarkSword’ Exploit

www.techrepublic.com · 7 months ago

Researchers uncovered DarkSword, a powerful iPhone exploit kit hosted on compromised websites that can silently compromise unpatched devices running iOS 18.4 through 18.6.2. The fileless attack hijacks iOS system processes to steal iCloud Keychain passwords, messages from iMessage, WhatsApp and Telegram, browser history, photos, calendar entries, notes, health data, email contents and data from cryptocurrency wallet apps, then deletes logs and temporary files so a reboot clears the active implant while the stolen data remains compromised. The tool has been observed in campaigns across Ukraine, Saudi Arabia, Turkey and Malaysia, researchers estimate roughly 221 million to 270 million iPhones remain vulnerable, and immediate protections include updating to iOS 26.3.1 or iOS 18.7.6, enabling Lockdown Mode for high risk users and avoiding untrusted sites.
BYOUD - Bring Your Own Unwind Data - By KlezVirus

klezvirus.github.io · 7 months ago

BYOUD is a framework that spoofs Windows call stacks by tampering with unwind metadata rather than altering return addresses, enabling believable stack reconstruction while remaining compatible with Intel Control Flow Enforcement Technology and Hardware Stack Protection. The technique locates and edits UNWINDINFO and runtime function table entries in process memory so crafted frame sizes and proxy functions cause stack walkers and debuggers to reconstruct a different, legitimate looking call chain while the shadow stack and return flow remain intact. There are three main variants plus a JIT specific approach that demonstrate scanning the stack to calculate concealment distances and abusing proxy calls such as NdrServerCall2 to present fully unwindable fake subtrees without touching the shadow stack.
Russian APT Exploits Zimbra Vulnerability Against Ukraine

www.securityweek.com · 7 months ago

A Russian state sponsored threat actor has exploited a high severity stored cross site scripting vulnerability in Zimbra Collaboration’s Classic UI tracked as CVE-2025-66376 with a CVSS score of 7.2, which was fixed in November 2025 in Zimbra versions 10.1.13 and 10.0.18. Insufficient sanitization of CSS content in HTML email allowed attackers to abuse CSS import directives to reference external resources or inject inline scripts that execute when a message is opened in a browser, enabling theft of credentials, session tokens, backup two factor authentication codes, saved passwords, and mailbox contents going back 90 days and facilitating remote code execution. CISA added CVE-2025-66376 to its Known Exploited Vulnerabilities catalog and urged federal agencies to patch within two weeks, while Seqrite Labs linked exploitation in attacks against Ukrainian targets to APT28 in a campaign named Operation GhostMail.
I built a zero-knowledge CLI password manager from scratch. AES-256-GCM, Argon2id, 22 secret types, MCP support....

aaravmaloo.github.io · 7 months ago

APM is a zero-knowledge command-line password manager written in Go that provides encrypted-at-rest vaults, multi-cloud synchronization, AI-agent integration via an MCP server, a Windows autofill daemon, a manifest-based plugin system, and a Team Edition with role-based access controls. Its encryption stack uses Argon2id for key derivation with memory-hard defaults and support up to 512 megabytes, AES-256-GCM for authenticated encryption, and HMAC-SHA256 integrity signatures, and the master password is never stored. APM supports more than twenty five structured secret types including passwords, TOTP, SSH keys, API keys, certificates, and documents, offers a YAML policy engine and shell-scoped sessions with inactivity timeouts, and exposes a security contact for alerts.
AI Replica of Val Kilmer to Appear in ‘As Deep as the Grave’

www.nytimes.com · 7 months ago

false
CISA Calls on Organizations to Strengthen Microsoft Intune Security After Stryker Incident

gbhackers.com · 7 months ago

There is an urgent CISA alert dated March 18, 2026, calling on organizations to harden Microsoft Intune and other endpoint management configurations after a March 11, 2026 cyberattack against Stryker that abused Intune to issue mass device wipe commands and disrupted the companys Microsoft environment. CISA recommends applying least privilege to administrative roles, enforcing multi factor authentication and privileged access hygiene, requiring multi administrator approval for high risk actions, restricting and monitoring service principals and app permissions, rotating and revoking stale credentials and tokens, and increasing audit logging and threat detection across endpoint management systems. Organizations are urged to implement Microsoft and CISA hardening guidance immediately to reduce the risk of similar operationally destructive attacks.
Max severity Ubiquiti UniFi flaw may allow account takeover

www.bleepingcomputer.com · 7 months ago

Ubiquiti patched two vulnerabilities in the UniFi Network Application, including a maximum-severity path traversal flaw tracked as CVE-2026-22557 that affects UniFi Network versions 10.1.85 and earlier and is fixed in versions 10.1.89 or later. Successful exploitation can allow low-privilege or network-accessible attackers to access files on underlying systems and potentially hijack user accounts in low-complexity attacks that do not require user interaction. Ubiquiti also fixed an authenticated NoSQL injection vulnerability that could let an authenticated low-privilege actor escalate privileges.
SpyCloud’s 2026 Identity Exposure Report Reveals Explosion of Non-Human Identity Theft

www.csoonline.com · 7 months ago

SpyCloud's 2026 Identity Exposure Report shows a 23 percent expansion in its recaptured identity datalake to 65.7 billion distinct identity records and a sharp rise in non human identity exposure, including 18.1 million exposed API keys and tokens and 6.2 million credentials or authentication cookies tied to AI tools. There are 28.6 million phished identity records from 2025 with nearly half belonging to corporate users, and modern phishing datasets increasingly include session cookies, authentication tokens, and MFA workflow data that enable session takeover. Session theft and MFA bypass remain widespread, with 8.6 billion stolen cookies and session artifacts recaptured and a 51 percent overlap between combolists and infostealer logs, and authorities executed a coordinated disruption of the Tycoon 2FA phishing infrastructure on March 4, 2026. Credential exposure remains severe with 5.3 billion credential pairs recaptured, 80 percent of exposed corporate credentials containing plaintext passwords, 1.1 million password manager master passwords circulating, and over 642.4 million exposed credentials recovered from 13.2 million infostealer infections.
A timeline of MCP security breaches: Tool poisoning, RCE via mcp-remote, sandbox escapes, and 7,000+ exposed servers

brightbean.xyz · 7 months ago

There are nearly 7,000 exposed Model Context Protocol servers on the open internet and the protocol commonly ships without built in authentication, least privilege controls, or audit trails, creating a blind spot for zero trust architectures. There have been documented incidents including WhatsApp data exfiltration, GitHub repository leaks, a command injection disclosed as CVE-2025-6514 in the mcp-remote package, and two critical Anthropic filesystem flaws CVE-2025-53109 and CVE-2025-53110 that allowed sandbox escape and symlink containment bypass. Recommended mitigations include treating MCP connections as privileged access, sanitizing natural language context before models process it, enforcing least privilege with short lived tokens, sandboxing MCP runtimes, monitoring for tool mutation, and requiring human approval for sensitive operations.
New Perseus Android Banking Malware Monitors Notes Apps to Extract Sensitive Data

thehackernews.com · 7 months ago

A new Android malware family called Perseus is being actively distributed via dropper apps on phishing sites to conduct device takeover and financial fraud. Built on the Cerberus and Phoenix codebases, Perseus leverages Accessibility based remote sessions to run overlay attacks, capture keystrokes, and monitor note taking apps for high value personal and financial information. Operators can issue commands to scan notes, start near real time visual streams, interact programmatically with user interface elements, force installation from unknown sources, mute audio, and display black screen overlays while the malware performs checks for debuggers and emulators and reports a suspicion score to its command and control server.
CVE-2026-22729: JSONPath Injection in Spring AI’s PgVectorStore

blog.securelayer7.net · 7 months ago

A JSONPath injection vulnerability in Spring AI's PgVectorStore can allow attackers to craft malicious filter expressions that bypass metadata filtering and access controls, tracked as CVE-2026-22729. The flaw stems from unsafe handling of JSONPath filter expressions when converting filter expressions into PostgreSQL JSON path queries, enabling unauthorized retrieval or deletion of vector store records via specially crafted queries.
Deep dive into how OS-level age verification systems can be bypassed

thecybersecguru.com · 7 months ago

Recent laws in California and Brazil now require strict age verification and privacy defaults at the operating system and platform level, prompting companies to block access rather than implement costly compliance measures. Open source projects and smaller vendors are being region blocked, and communities are responding with network obfuscation such as privacy focused VPNs, peer to peer distribution of operating system images, and protest modifications like Ageless Linux that disable or stub age verification APIs. Users are also turning to radical self hosting and local AI workflows using tools like Docker, SearXNG, LM Studio, and Ollama to avoid centralized age and identity checks, widening a divide between regulated platforms and technically literate users.
Claude Vulnerabilities Allow Data Exfiltration and Malicious Redirect Attacks

gbhackers.com · 7 months ago

Security researchers uncovered a critical attack chain within Anthropic's Claude.ai platform named Claudy Day that enables silent extraction of sensitive user data through prompt manipulation and malicious redirects. The chain combines three flaws: invisible prompt injection using hidden HTML in prefilled chat prompts, data exfiltration by forcing the model to search chat history and upload files using an attacker controlled API key, and open redirect vulnerabilities on the main domain that can be abused via advertising to deliver malicious links. The exploit operates within a default Claude session without external tools and can escalate if users connect Model Context Protocol servers, third party APIs, or corporate files, allowing the AI to access internal resources. Organizations should audit AI integrations, restrict unnecessary MCP servers and API access, apply prompt injection fixes, and train users to avoid clicking shared links and prefilled prompts.
Fake Tools and CDNs Power New “Vibe-Coded” Malware Campaign

gbhackers.com · 7 months ago

Vibe coding has moved from buzzword to battleground as attackers use AI-assisted development to produce and distribute more than 440 malicious ZIP archives in January 2026 that masquerade as AI tools, game mods, drivers, VPNs, and utilities hosted on content delivery and file-sharing platforms. Each ZIP bundles a clean-looking executable that sideloads a malicious WinUpdateHelper.dll which executes verbose PowerShell payloads in memory, creates a persistent service named Microsoft Console Host, generates time-based command and control domains that rotate roughly every 58 days, and downloads miner binaries that impersonate legitimate Windows processes while disabling defenses and adding Windows Defender exclusions. Operators mine privacy coins and Ravencoin via mining pools, have processed visible Bitcoin inflows of about eleven thousand five hundred dollars, and sometimes swap miners for stealers and remote access tools; users and organizations are advised to install software only from official vendor sites, monitor for unusual miner traffic and time-based xyz domains, and watch for changes to Defender exclusions and suspicious services masquerading as native components.
CISA urges US orgs to secure Microsoft Intune systems after Stryker breach

www.bleepingcomputer.com · 7 months ago

There are warnings from CISA urging U.S. organizations to harden Microsoft Intune after a March 11, 2026 cyberattack on Stryker in which attackers allegedly created a new Global Administrator from a compromised admin account, stole about 50 terabytes of data, and used Intune's built in wipe command to erase nearly 80,000 devices. There are Microsoft guidance recommendations to apply least privilege using role based access control, enforce multi factor authentication and Microsoft Entra conditional access and risk signals, practice privileged access hygiene, and require multi admin approval for sensitive actions such as device wipes and RBAC changes. There are reports that the attack was claimed by Handala, a hacktivist group linked to Iran's Ministry of Intelligence and Security that has a history of stealing and leaking data.
How Ceros Gives Security Teams Visibility and Control in Claude Code

thehackernews.com · 7 months ago

false
Raven Emerges From Stealth With $20 Million in Funding

www.securityweek.com · 7 months ago

Raven, a Palo Alto California based cybersecurity startup founded in 2023 by Israeli cyber intelligence veterans, emerged from stealth with twenty million dollars in combined seed and post seed funding led by Norwest and Elron Ventures with additional investors including SentinelOne and others. The company has built a runtime platform that observes applications to detect anomalous behavior, uncover and monitor AI agents in production, and block unsafe actions and malicious code in real time even when no known vulnerabilities exist. Raven will use the funding to accelerate product development, expand its research and engineering teams, and scale go to market efforts.
New ‘Perseus’ Android malware checks user notes for secrets

www.bleepingcomputer.com · 7 months ago

There is a new Android malware called Perseus that checks user curated notes to steal sensitive information such as passwords recovery phrases and financial data. Perseus is distributed via sideloaded APKs disguised as IPTV apps and can bypass Android 13 and later sideloading restrictions to gain a foothold on devices. By abusing Android Accessibility Services operators can take full remote control to capture continuous screenshots stream device displays perform overlay attacks simulate user input and keylog to steal credentials. The malware performs extensive anti analysis and evasion checks including device fingerprints and SIM and hardware data sends a suspicion score to a command and control panel before data theft and users are advised to avoid sideloading install apps from the official store and enable Play Protect.
Pyronut Package Backdoors Telegram Bots With RCE

gbhackers.com · 7 months ago

A trojanized Python package named Pyronut backdoors Telegram bots and userbots, giving attackers remote code execution in both the Telegram session and the underlying host system. Pyronut impersonated the pyrogram framework on PyPI and shipped three malicious releases numbered 2.0.184, 2.0.185, and 2.0.186 that were detected and quarantined on March 18, 2026. At runtime it modifies the Client.start method to import a hidden module that registers handlers for the commands /e and /shell, enabling remote evaluation of Python code via the meval library and arbitrary shell execution by invoking /bin/bash -c with results exfiltrated over Telegram. Detection and response guidance includes searching dependency manifests for pyronut versions, flagging unexpected use of meval, inspecting for Python spawned shell processes and unusual file or network activity, revoking affected Telegram sessions and tokens, rebuilding environments from known good baselines, and rotating exposed credentials.
Critical Microsoft SharePoint flaw now exploited in attacks

www.bleepingcomputer.com · 7 months ago

A critical Microsoft SharePoint vulnerability patched in January is now being exploited in attacks, affecting SharePoint Enterprise Server 2016, SharePoint Server 2019, and SharePoint Server Subscription Edition. Successful exploitation allows unauthenticated attackers to achieve remote code execution by exploiting a deserialization of untrusted data weakness; the issue is tracked as CVE-2026-20963 and was addressed in Microsofts January 2026 Patch Tuesday. The Cybersecurity and Infrastructure Security Agency added the flaw to its catalog of actively exploited vulnerabilities, ordered federal civilian executive branch agencies to secure servers by March 21, 2026, and strongly urged all network defenders to apply mitigations and patch against related issues including CVE-2025-40551.
Analyzing the Current State of AI Use in Malware

unit42.paloaltonetworks.com · 7 months ago

There are three primary ways AI is being applied to malware: to write malware, to provide remote decision making via an AI-backed command and control, and to perform local agentic decision making, though only the first two have been observed in the wild. Researchers analyzed a trio of.NET infostealers that call OpenAI GPT-3.5-Turbo via HTTP to run four LLM-driven functions that mainly log suggested evasion or obfuscation names and in one case return a millisecond delay used for sleeping, and a Golang dropper that queries an LLM to assess the environment and gate execution. The findings show AI can lower the skill barrier for attackers but current integrations are often superficial or poorly implemented, and deploying local models for meaningful autonomous attack flows remains challenging for adversaries.
The multi-billion dollar mistake: Why cloud misconfigurations are your biggest security threat

www.csoonline.com · 7 months ago

Misconfigured cloud settings are the primary cause of breaches, with a 2024 Cloud Security Alliance analysis finding configuration errors enabled nearly every incident. The financial scale is steep, with IBM reporting average breach costs of 4.44 million globally and 10.22 million in the United States, and several high profile incidents exposed hundreds of millions of records or affected hundreds of companies. Root causes include human error, sprawling complex environments, shadow IT, unclear ownership, and a lack of monitoring, with industry data showing most errors are human driven, many storage locations go unmonitored, and remediations often take around three months. Recommended actions include enforcing multi factor authentication everywhere, locking down public storage and overly permissive network rules, enabling comprehensive logging and monitoring, using cloud security posture management and infrastructure as code checks, adopting least privilege, and investing in cross training and leadership support.