Tuesday, 20 October 2026

Governments are hacking people with ‘zero-click’ spyware by showing them online ads (opens in a new tab)

TechCrunch Security ·Zack Whittaker ·fetched 20 Oct 2026, 03:36 UTC Must read Research

Why readCitizen Lab exposed how spyware vendors Candiru and Bindecy leverage commercial ad networks to deliver zero-click Windows exploits.

Citizen Lab published research revealing how Israeli firm Discover-Tech operates an ad platform bridging ad networks and commercial spyware vendors. The infrastructure allowed Candiru and Bindecy to serve zero-click Windows exploits to targeted victims, including infections across 45 workstations at Al Jazeera.

Also covered byDark Reading (opens in a new tab).

techcrunch.com → (opens in a new tab)

Hackers Exploit Palo Alto VPN Flaw to Bypass MFA and Steal Credentials From VMware Disks (opens in a new tab)

Cybersecurity News ·Guru Baran ·fetched 20 Oct 2026, 11:35 UTC Must read CVE-2026-0257 EPSS 96.9%

Why readDetails how ransomware actors exploit CVE-2026-0257 in Palo Alto GlobalProtect to bypass MFA and extract credentials directly from domain controller VM disks.

Attackers linked to Qilin and Inc ransomware are actively exploiting CVE-2026-0257 in Palo Alto Networks GlobalProtect VPNs to bypass MFA. Once inside, threat actors access VMware management interfaces and steal credentials directly from virtual disks of domain controllers, evading security tools running inside the guest OS.

Also covered byThe Register Security (opens in a new tab),Cybersecurity Dive (opens in a new tab),DataBreaches.net (opens in a new tab).

cybersecuritynews.com → (opens in a new tab)

Hackers Hijack Five OVHcloud CLI Releases to Distribute Secret-Stealing Malware (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 20 Oct 2026, 11:35 UTC Must read

Why readAttackers hijacked a contributor account to replace binaries on five historic GitHub releases of OVHcloud CLI with secret-stealing malware.

An attacker compromised an OVHcloud contributor's GitHub credentials and swapped release archives on five previously published versions of the OVHcloud command-line interface. The malicious binaries were live for 94 minutes on October 19 and targeted Linux, macOS, and Windows environments before OVHcloud replaced them with immutable releases.

thehackernews.com → (opens in a new tab)

AOMEI yet to notify customers after hackers hijacked its software updates (opens in a new tab)

The Record ·fetched 20 Oct 2026, 15:33 UTC

Why readDetails a short-burst update hijacking campaign against AOMEI backup software that evaded automated triage scanners.

Attackers repeatedly modified update server configuration files for brief window intervals between April and July to distribute malware while bypassing point-in-time security scans. Investigations by Google Threat Intelligence and ManageEngine confirmed the supply-chain intrusion, though vendor notification to affected users remains pending.

therecord.media → (opens in a new tab)

ConnectWise Cloud Backup application abused in cross-tenant token sweep (opens in a new tab)

Sophos Threat Research ·fetched 20 Oct 2026, 03:36 UTC Must read Research

Why readDetails an attack where an adversary used compromised ConnectWise Cloud Backup Entra application credentials to perform cross-tenant Microsoft Graph token requests across consenting customer environments.

Sophos X-Ops observed an attacker authenticating via the ConnectWise Cloud Backup (formerly SkyKick) Entra service principal identity from a single VPS to request app-only Microsoft Graph tokens across consenting M365 tenants. Because the activity authenticated as the application itself, no passwords or multi-factor authentication checks were triggered during the 96-minute enumeration scan.

Also covered byDark Reading (opens in a new tab),Infosecurity Magazine (opens in a new tab).

sophos.com → (opens in a new tab)

Popular WordPress SEO Plugin Rank Math PRO Hijacked to Deliver Self-Restoring Backdoor (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 20 Oct 2026, 03:36 UTC

Why readCopies of Rank Math PRO WordPress plugin downloaded during an 11-hour window contained malware establishing self-restoring backdoors.

A supply-chain attack compromised the distribution site for the Rank Math PRO WordPress plugin between October 15 at 20:02 UTC and October 16 at 07:12 UTC. Trojanized copies of version 3.0.123 deploy a persistent backdoor on infected sites. Upgrading to clean version 3.0.124 replaces core files but does not remove existing backdoors, rogue admin accounts, or scheduled tasks.

Also covered byBleepingComputer (opens in a new tab).

thehackernews.com → (opens in a new tab)

Zeabur shutting down after second security breach in two months (opens in a new tab)

BleepingComputer ·Bill Toulas ·fetched 20 Oct 2026, 19:36 UTC

Why readAssess exposure if your team used Zeabur, as exfiltrated data includes 3.2 million environment variable records containing application secrets.

Developer deployment platform Zeabur announced it is shutting down following a major infrastructure compromise that resulted in the theft of 3.2 million environment variable records and 321,810 user accounts. Threat actors leveraged leaked internal credentials stored in plaintext along with cloud access keys to compromise shared clusters, dedicated servers, and the central database. This marks the company's second major breach in two months, following an August incident where attackers pivoted from a customer application into internal systems.

bleepingcomputer.com → (opens in a new tab)

#StopRansomware: SafePay Ransomware (opens in a new tab)

CISA Advisories ·CISA ·fetched 20 Oct 2026, 19:36 UTC CVE-2020-3259 EPSS 25.5%

Why readCISA released an advisory detailing SafePay ransomware operations, living-off-the-land techniques, and exploitation of CVE-2020-3259.

A CISA advisory documents active campaigns by the SafePay ransomware group targeting critical infrastructure across North America, Europe, and Australia. SafePay actors rely on credential theft, phishing, and known vulnerabilities such as CVE-2020-3259 to gain entry, using native system binaries to move laterally before encrypting victim data.

cisa.gov → (opens in a new tab)

FBI says SafePay ransomware gang ‘among most active’ with more than 600 victims since 2024 (opens in a new tab)

The Record ·fetched 20 Oct 2026, 23:32 UTC

Why readDetails active TTPs and specific exploited CVEs used by the SafePay ransomware group across 600+ target networks.

A joint advisory from the FBI, German, and UK law enforcement warns that SafePay has become one of the most active ransomware operations in 2026. The gang relies on credential guessing, phishing, and known vulnerabilities including CVE-2020-3259, CVE-2023-20269, CVE-2023-27532, CVE-2024-40711, and CVE-2024-40766 to compromise networks.

therecord.media → (opens in a new tab)

From Phishing to LLMjacking: Threat Actors Targeting Cloud Billing Accounts for LLM Usage (opens in a new tab)

Wiz ·Andre Maccarone ·fetched 20 Oct 2026, 23:32 UTC Research

Why readDetails GCP tradecraft where attackers hijack billing accounts to route high-cost LLM workloads into external projects, hiding prompts and outputs from victims.

Wiz CIRT documented multiple intrusions where threat actors compromised privileged GCP credentials and linked external, attacker-controlled GCP projects directly to victim billing accounts. By processing massive LLM workloads in external projects, the attackers generated hundreds of thousands of dollars in API charges while completely obscuring prompt contents and model outputs from victim audit logs. The report outlines the attack lifecycle, key GCP telemetry indicators, and mitigation steps for billing IAM permissions.

wiz.io → (opens in a new tab)

Hackers Abuse AI Pen-Test Tool Cairn to Automate Massive Attack Spree (opens in a new tab)

Dark Reading ·Elizabeth Montalbano ·fetched 20 Oct 2026, 15:33 UTC

Why readDetails active exploitation campaigns leveraging the open-source AI penetration testing tool Cairn to breach major industrial targets.

A Chinese-speaking threat actor is abusing Cairn, an AI-assisted penetration testing tool, to automate attacks against enterprise networks. Compromised targets include prominent manufacturing firms such as John Deere, SharkNinja, and Cummins.

darkreading.com → (opens in a new tab)

Defense contractors in two NATO allies hacked in recent Russia-linked campaign (opens in a new tab)

The Record ·fetched 20 Oct 2026, 11:35 UTC

Why readAPT28 targeted NATO defense contractors using the recently discovered Hookedge backdoor.

Symantec reports that Russian GRU-linked group APT28 compromised a naval shipbuilder and aircraft engine manufacturer in European NATO countries. Initial access was obtained via malicious Word documents, followed by the deployment of the Hookedge backdoor to extract credentials and map internal networks.

therecord.media → (opens in a new tab)

8 eMudhra certificates are compromised. Here's what you need to know (opens in a new tab)

Cloudflare Blog ·Jeremy Hartman ·fetched 20 Oct 2026, 23:32 UTC

Why readGet Cloudflare's timeline, impact analysis, and specific client mitigations following the compromise of eight eMudhra intermediate TLS private keys.

Cloudflare disclosed its response to the compromise of eight eMudhra intermediate TLS certificate private keys, which could allow adversaries to forge certificates and intercept traffic. The vendor updated WARP clients, cloudflared tunnels, and edge CDN configurations to revoke trust in the affected intermediates. Internal reviews indicate no active exploitation across Cloudflare services to date.

Also covered byThe Register Security (opens in a new tab),The Hacker News (opens in a new tab),Ars Technica Security (opens in a new tab).

blog.cloudflare.com → (opens in a new tab)

Attackers exploit Siemens Software Center flaw in active ClickFix malware campaign (opens in a new tab)

Help Net Security ·Sinisa Markovic ·fetched 20 Oct 2026, 15:33 UTC Must read CVE-2026-110936 EPSS 0.1%

Why readWarns of active exploitation targeting a DLL side-loading vulnerability in Siemens Software Center.

Siemens updated its advisory to warn that threat actors are actively exploiting CVE-2026-110936 in ClickFix malware campaigns. The flaw allows local privilege escalation via DLL side-loading in Siemens Software Center and is fixed in version 4.0.9.8.

Also covered byThe Hacker News (opens in a new tab),Infosecurity Magazine (opens in a new tab).

helpnetsecurity.com → (opens in a new tab)

Critical FortiSIEM Flaw Gets Public Exploit PoC for Unauthenticated Remote Code Execution (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 20 Oct 2026, 19:36 UTC Must read CVE-2026-106173 EPSS 0.4%

Why readA public proof-of-concept exploit is available for a critical FortiSIEM path traversal vulnerability tracked as CVE-2026-106173.

Horizon3.ai published technical details and a working proof-of-concept for CVE-2026-106173, a CVSS 9.4 vulnerability in Fortinet FortiSIEM. Unauthenticated attackers can target the gRPC service on TCP port 27918 using path traversal and NUL-byte injection in HTTP/2 headers to achieve arbitrary file writes and remote code execution.

thehackernews.com → (opens in a new tab)

F5 rushes out emergency NGINX patches for three critical flaws (opens in a new tab)

CSO Online ·fetched 20 Oct 2026, 03:36 UTC Must read CVE-2026-93479 EPSS 0.5%

Why readF5 released emergency out-of-band updates for NGINX fixing three critical vulnerabilities that allow remote code execution and denial of service.

F5 issued emergency patches for three critical flaws in NGINX, rated 9.2 CVSS: CVE-2026-15176, CVE-2026-97205, and CVE-2026-93479. The vulnerabilities impact NGINX Plus, Open Source, and Ingress Controller deployments, enabling unauthenticated attackers to cause crashes or execute arbitrary code. Teams running NGINX infrastructure should deploy fixed versions immediately.

csoonline.com → (opens in a new tab)

KVM escape exposes broader attack surface in Linux networking (opens in a new tab)

CSO Online ·fetched 20 Oct 2026, 23:32 UTC Research CVE-2026-99999 EPSS 0.2%

Why readExamine a high-severity Linux kernel networking flaw that enables virtual machine escapes from KVM and Firecracker microVMs.

Researchers identified CVE-2026-99999 (PwnEscape) in the Linux kernel networking stack during a security challenge against Vercel's sandbox platform. The vulnerability allows an attacker inside a VM to manipulate IPv4 packet headers after host validation, causing host memory leakage, crashes, or root code execution. The underlying bug exists in Linux network packet processing rather than the hypervisor code itself.

csoonline.com → (opens in a new tab)

Hackers Exploit SiteGuard WAF Vulnerability That Could Give Them Full Control of Servers (opens in a new tab)

Cybersecurity News ·Abinaya ·fetched 20 Oct 2026, 03:36 UTC CVE-2026-106931 EPSS 0.7%

Why readCommand injection flaw CVE-2026-106931 in SiteGuard Server Edition WAF is seeing active exploitation in the wild.

Attackers are actively exploiting an OS command injection vulnerability in SiteGuard Server Edition, a web application firewall running directly on target servers. Tracked as CVE-2026-106931, the flaw resides in the default-enabled statistics processing module and allows unauthenticated execution to gain root or SYSTEM privileges.

cybersecuritynews.com → (opens in a new tab)

CVE-2026-106931: EG Secure Solutions SiteGuard Server Edition, EG Secure Solutions SiteGuard Server Edition OS Command Injection Vulnerability (opens in a new tab)

CISA KEV ·fetched 20 Oct 2026, 23:32 UTC CVE-2026-106931 Exploited in the wild · patch by 2026-10-23 EPSS 0.7%

Why readCISA added an active OS command injection vulnerability in EG Secure Solutions SiteGuard Server Edition to its KEV catalog.

CISA added CVE-2026-106931 to its Known Exploited Vulnerabilities catalog following active exploitation. The OS command injection vulnerability in EG Secure Solutions SiteGuard Server Edition allows remote attackers to execute arbitrary commands with root or SYSTEM privileges. Federal agencies must apply mitigations or cease using the software by October 23, 2026.

eg-secure.co.jp → (opens in a new tab)

CVE-2026-110936: Siemens Software Center, Siemens Software Center Uncontrolled Search Path Element Vulnerability (opens in a new tab)

CISA KEV ·fetched 20 Oct 2026, 23:32 UTC CVE-2026-110936 Exploited in the wild · patch by 2026-11-03 EPSS 0.1%

Why readCISA added a Siemens Software Center DLL hijacking vulnerability to its KEV catalog.

Siemens Software Center contains an uncontrolled search path element vulnerability tracked as CVE-2026-110936. An authenticated local attacker can drop a crafted DLL into an application search path to execute arbitrary code with application privileges. CISA has added the vulnerability to its KEV catalog with a remediation deadline of November 3, 2026.

productcert.siemens.com → (opens in a new tab)

CVE-2026-96656: Plex Media Server, Plex Media Server External Control of File Name or Path Vulnerability (opens in a new tab)

CISA KEV ·fetched 20 Oct 2026, 23:32 UTC CVE-2026-96656 Exploited in the wild · patch by 2026-10-23 EPSS 0.4%

Why readCISA added a Plex Media Server arbitrary file write flaw to the Known Exploited Vulnerabilities catalog.

CISA updated its Known Exploited Vulnerabilities catalog with CVE-2026-96656, an arbitrary file write flaw affecting Plex Media Server. The vulnerability allows admin users to write files that execute on load and can be chained with CVE-2026-96652, CVE-2026-96654, and CVE-2026-96655. The federal remediation deadline is set for October 23, 2026.

forums.plex.tv → (opens in a new tab)

Plex2Shell Exploit Used to Turn Vulnerable Plex Servers Into Crypto Miners (opens in a new tab)

Security Affairs ·Pierluigi Paganini ·fetched 20 Oct 2026, 11:35 UTC CVE-2026-96655 EPSS 0.4%

Why readAttackers are actively exploiting the four-bug Plex2Shell chain (CVE-2026-96652 through CVE-2026-96656) in the wild to deploy crypto miners and proxies.

Security researchers observed active exploitation in honeypots within 24 hours of Zach Main publishing PoC code for the Plex2Shell vulnerability chain. The chain combines four CVEs in Plex Media Server to achieve unauthenticated remote code execution on internet-facing instances.

securityaffairs.com → (opens in a new tab)

CVE-2026-106173 Deep-Dive: FortiSIEM gRPC Path Traversal File Write RCE (opens in a new tab)

Horizon3 Attack Team ·Zach Hanley ·fetched 20 Oct 2026, 15:33 UTC Research CVE-2026-106173 EPSS 0.4%

Why readTechnical breakdown and root-cause analysis of a critical gRPC path traversal flaw (CVE-2026-106173) in FortiSIEM.

Horizon3 Attack Team presents a technical analysis of CVE-2026-106173, an arbitrary file write flaw in FortiSIEM's phQueryMaster service listening on port tcp/27918. Unlike previously analyzed vulnerabilities in phMonitor on port tcp/7900, this issue resides in the grpcNotificationServer::run() routine. The writeup demonstrates how the gRPC handler processes unauthenticated requests to achieve arbitrary file writes and execution.

horizon3.ai → (opens in a new tab)

2026-018: Critical Vulnerabilities in PHP (opens in a new tab)

CERT-EU Advisories ·fetched 20 Oct 2026, 23:32 UTC

Why readPrepare for emergency patching of two critical PHP vulnerabilities rated CVSS 9.2 that could enable remote code execution.

The PHP development team announced emergency patches scheduled for October 21 at 14:00 UTC addressing two CVSS 9.2 remote code execution flaws. Technical details have reportedly leaked ahead of disclosure, prompting CERT-EU to recommend isolating exposed applications if immediate updates cannot be deployed.

Also covered byBleepingComputer (opens in a new tab).

cert.europa.eu → (opens in a new tab)

Introducing pathrunner: An AWS privilege escalation framework (opens in a new tab)

Datadog Security Labs ·fetched 20 Oct 2026, 19:36 UTC Must read Research

Why readDatadog released pathrunner, an open-source framework designed to automate AWS IAM privilege escalation attacks.

Building on its pathfinding.cloud taxonomy and vulnerable lab environments, Datadog Security Labs released pathrunner. The modular interactive console automates the exploitation of cloud privilege escalation paths across AWS IAM principals, helping offensive and defensive teams validate cloud security posture.

securitylabs.datadoghq.com → (opens in a new tab)

Locate, Then Reason: End-to-End Patch Presence Testing in Stripped Binaries (opens in a new tab)

arXiv cs.CR (AI) ·Huaijin Wang, Tongda Wang, Zhibo Liu, Yue Zhang ·fetched 20 Oct 2026, 07:32 UTC Research

Why readIntroduces PatchAlign, an end-to-end tool that identifies whether security patches are present in stripped binaries without relying on function-level matching or debug symbols.

Binary patch presence testing determines if a specific security fix exists within compiled target code, but compiler optimizations and stripped symbols often break traditional function matching. PatchAlign uses instruction-aligned retrieval combined with caller-callee context graph analysis to pinpoint relevant code fragments. An LLM then compares target binary fragments against known vulnerable and patched reference snippets to issue precise patch verdicts.

arxiv.org → (opens in a new tab)

Discovery of cross-domain cyberattack risk targeting electric vehicle charging navigation systems (opens in a new tab)

Google News: incidents · Tech Xplore ·fetched 20 Oct 2026, 19:36 UTC

Why readTrack emerging cross-domain attack vectors targeting electric vehicle charging navigation systems.

Initial reporting on cybersecurity research exposing cross-domain attack risks in electric vehicle charging navigation infrastructure. The research demonstrates how vulnerabilities in connected navigation systems can be leveraged to disrupt EV charging telemetry and routing algorithms.

news.google.com → (opens in a new tab)

People Are Using 'AI Decompiling' to Hack Their Things (opens in a new tab)

404 Media ·Jason Koebler ·fetched 20 Oct 2026, 15:33 UTC

Why readExamines how hardware hackers are using LLM decompilation tools to accelerate reverse engineering of binary code.

Hardware security practitioners and reverse engineers are increasingly adopting LLM-powered decompilation workflows to decode machine code and analyze embedded targets. The technique uses generative models to convert complex assembly into readable source logic, simplifying vulnerability research.

404media.co → (opens in a new tab)

Fake Kernel Threads Are Hiding Backdoors on Linux Servers and Appliances (opens in a new tab)

Eclypsium ·Chris Garland ·fetched 20 Oct 2026, 15:33 UTC Must read Research CVE-2026-75650 EPSS 3.9%

Why readExplains how attackers disguise Linux malware as kernel worker threads and provides concrete inspection checks to spot them.

Multiple threat actors, including attackers exploiting Magento zero-day CVE-2026-75650, are disguising Linux backdoors using fake kernel thread names such as [kworker/u:8:0]. Real kernel worker threads are child processes of kthreadd (PID 2) and do not have associated executable binaries on disk, open user network sockets, or mapped userland memory regions. Incident responders can reliably identify rogue kworker processes by checking parent process IDs, inspecting /proc/PID/exe symlinks, and auditing process memory maps.

eclypsium.com → (opens in a new tab)

Application data breaches: Lessons learned from CERT-FR (opens in a new tab)

translated Violations de données applicatives - Retour d’expérience du CERT-FR (20 octobre 2026)

CERT-FR (ANSSI) ·fetched 20 Oct 2026, 11:35 UTC Must read

Why readAnalyzes almost 200 web application data breaches handled by CERT-FR to extract practical defensive lessons.

CERT-FR published CTI report CERTFR-2026-CTI-008 detailing lessons learned from responding to nearly 200 web application data breaches in 2025. The analysis reveals that many high-impact exfiltrations stemmed from low-complexity attacks without requiring zero-day exploits. The report focuses on practical application hardening and risk mitigation strategies to prevent data exposure.

cert.ssi.gouv.fr → (opens in a new tab)

DM-RSA-OAEP: Formal Treatment of a Dual-Modulus Fault-Detection Countermeasure Based on Physical Diversity (opens in a new tab)

arXiv cs.CR (all) ·Ramamonjy Andriamifidisoa, Rufine Marius Lalasoa ·fetched 20 Oct 2026, 19:36 UTC Research

Why readLearn the formal security bounds and failure modes of DM-RSA-OAEP as a physical fault-detection countermeasure.

Researchers present a formal treatment of DM-RSA-OAEP, a dual-modulus RSA countermeasure designed to detect physical fault injection and safe-error attacks. The authors provide an IND-CCA2 security proof in the random-oracle model and establish explicit fault-detection bounds under physical-diversity assumptions. Crucially, the paper demonstrates how the countermeasure degrades completely under common-mode faults on single-chip implementations.

arxiv.org → (opens in a new tab)

Detecting GitHound Enumeration via the GitHub API (opens in a new tab)

detect.fyi ·Stamatis Chatzimangou ·fetched 20 Oct 2026, 11:35 UTC Must read

Why readOutlines audit logging and API telemetry strategies to detect automated GitHound credential harvesting against GitHub Enterprise.

Attackers targeting developer infrastructure frequently deploy automated tools like GitHound to scan GitHub API endpoints for exposed secrets and credentials. The guide details specific API audit log patterns, rate anomalies, and telemetry indicators required to flag GitHound enumeration before repository compromise occurs.

detect.fyi → (opens in a new tab)

Chrome flags malicious commands that scammers pass off as fixes (opens in a new tab)

Help Net Security ·Anamarija Pogorelec ·fetched 20 Oct 2026, 19:36 UTC

Why readReview Google Chrome's new Enhanced Protection safeguards against copy-paste malicious command scams.

Google is updating Chrome Enhanced Protection to actively detect scam websites that trick users into executing malicious terminal commands, often referred to as ClickFix attacks. The feature analyzes page behavior to identify prompts that instruct users to copy PowerShell or shell commands disguised as software fixes or verification steps.

helpnetsecurity.com → (opens in a new tab)

The journey of a paper unicorn: adventures inside a modern warehouse (opens in a new tab)

Securelist ·Sergey Sidorov, Haidar Kabibo, Victor Zvarykin, Kaspersky Security Services ·fetched 20 Oct 2026, 07:32 UTC

Why readAnalyzes operational technology cyber risks across modern automated warehouse infrastructure.

Kaspersky Security Services examines how cyber incidents disrupt modern automated warehouses that combine web applications, radio networks, and industrial control systems. The paper details how single data corruptions in warehouse management systems cascade into physical conveyor halts, freight backlog, and supply chain delays.

securelist.com → (opens in a new tab)

opentrap/opentrap: A free, open-source alternative to Burp Suite: an intercepting proxy and web app security testing suite with Repeater, Intruder, Collaborator and a built-in OWASP ZAP scanner. Linux, AGPL-3.0. (opens in a new tab)

GitHub: new security tools ·opentrap ·fetched 20 Oct 2026, 19:36 UTC Research ★ 167

Why readAn open-source, AGPL-licensed alternative to Burp Suite that wraps the OWASP ZAP scanning engine in a familiar proxy workflow.

OpenTRAP combines a Burp Suite style intercepting proxy user interface with the open-source OWASP ZAP active scanning engine. The Java-based suite includes Repeater, Intruder, and Collaborator features without commercial licensing restrictions.

github.com → (opens in a new tab)

EFF Launches Privacy Badger for Safari on macOS (opens in a new tab)

EFF Deeplinks ·Hudson Hongo ·fetched 20 Oct 2026, 11:35 UTC

Why readExtends EFF's Privacy Badger browser extension to Safari on macOS to block third-party trackers automatically.

The Electronic Frontier Foundation released Privacy Badger for Safari on macOS, bringing algorithmic tracker blocking to Apple's browser. Requiring Safari version 26.5 or higher, the extension automatically detects and blocks domain-based tracking behaviors without relying on static blocklists.

eff.org → (opens in a new tab)

Google may add your number to all your accounts: How to stop it (opens in a new tab)

Malwarebytes Labs ·fetched 20 Oct 2026, 19:36 UTC

Why readLearn how Android's automatic phone number verification links identity across multiple signed-in Google accounts and how to disable it.

Android automatically verifies device phone numbers across all signed-in Google accounts by default, creating unintended cross-account identity links. This behavior can compromise pseudonymous or secondary accounts used on the same device. The analysis details how the feature operates and provides steps for users to opt out and preserve account separation.

malwarebytes.com → (opens in a new tab)

Show HN: VoidAccess – open-source dark web OSINT, from query to a graph (opens in a new tab)

Hacker News ·coding-maniac ·fetched 20 Oct 2026, 15:33 UTC Research 54 points

Why readAutomate dark web OSINT collection and output entity relationship graphs while filtering out explicit media.

VoidAccess is an open-source engine that queries darknet sources, strips noise and NSFW content, and generates visual relationship graphs for analysts. The tool runs via command line and automates multi-step dark web investigations with configurable execution time limits.

github.com → (opens in a new tab) Discussion (opens in a new tab)

The Checker You Pay For (opens in a new tab)

Group-IB ·fetched 20 Oct 2026, 11:35 UTC

Why readDetails how fraudsters abuse open-banking account-linking APIs to validate stolen payment credentials and how early onboarding detection mitigates the fraud.

Fraudsters are using open-banking account-linking mechanisms to check whether stolen credit card and bank credentials are active, shifting verification costs onto targeted fintech platforms. Monitoring account creation, initial outbound messages, and linking workflows reduced time to block from hours to under a minute.

group-ib.com → (opens in a new tab)

DHS Said They Searched Man Who Wiped GrapheneOS Phone for Child Abuse Material. They Lied (opens in a new tab)

404 Media ·Joseph Cox ·fetched 20 Oct 2026, 19:36 UTC

Why readDetails court records showing how law enforcement handles forced mobile device searches and remote-wipe features.

Court filings from a criminal case against an activist reveal that CBP officials misled a traveler during interrogation before he entered a duress code that wiped his GrapheneOS device. The case sets a notable legal precedent regarding anti-forensics features, device destruction charges, and border search authority.

404media.co → (opens in a new tab)

Why Digital Evidence Needs a Behavioral Read, Not Just a Technical One (opens in a new tab)

Paraben ·Blogger ·fetched 20 Oct 2026, 19:36 UTC

Why readConsider how behavioral baselining complements technical artifact analysis in digital forensics.

This article discusses the limitations of relying solely on technical forensic artifacts such as timestamps and file hashes during incident investigations. It advocates for incorporating behavioral baselines to identify anomalies that pass automated technical checks, referencing psychological meta-analyses on deception detection to illustrate human evaluation biases.

paraben.com → (opens in a new tab)

Connection Escrow: Holding AI-Agent Pod Egress in the Kernel Until an Isolated Verifier Rules (opens in a new tab)

arXiv cs.CR (AI) ·Ashok Subbabhatta Gopalakrishna ·fetched 20 Oct 2026, 07:32 UTC Must read Research

Why readUses an eBPF program in the Linux kernel to hold outbound TCP connections from AI agent containers until an isolated verifier validates the destination and payload.

Traditional network controls can be bypassed by child processes or direct sockets spawned by compromised AI agents inside container environments. Connection Escrow places a 98-instruction eBPF program on the pod's cgroup to pause outbound TCP connections at the kernel connect() hook. A separate verifier container inspects the executable digest, process, and initial written bytes before releasing the connection via a signed token, passing a 36-check adversarial evaluation suite.

arxiv.org → (opens in a new tab)

Fake News, Real Crashes: Information Injection Attacks on LLM Financial Trading Agents (opens in a new tab)

arXiv cs.CR (AI) ·Mengxiao Wang, Nitesh Saxena ·fetched 20 Oct 2026, 07:32 UTC Research

Why readDemonstrates how fabricated news items and simple 30-word formatting templates achieve a 100% attack success rate in tricking LLM financial trading agents.

Unlike prompt injection which targets model instructions, information injection manipulates external evidence trusted by autonomous agents. Testing across 15 U.S. stocks and ~10,000 decisions, the study found that prompt injection achieved only a 14% success rate, whereas fake news reached 62% and specific structural formatting templates hit 100%. Removing simple authority markers like a source name or 'BREAKING' tag reduced the attack success rate back to zero without altering the underlying claim.

arxiv.org → (opens in a new tab)

OpenCode Breach Could Affect OpenAI, Anthropic and Google (opens in a new tab)

Infosecurity Magazine ·fetched 20 Oct 2026, 03:36 UTC Must read

Why readOpenCode's central console breach compromised workspace API credentials for OpenAI, Anthropic, and Google AI services.

Open-source AI agent framework provider OpenCode confirmed a security breach affecting its Console service, which centralizes model credential management for teams. The incident exposed API keys used to authenticate requests to OpenAI, Anthropic, and Google models across affected developer workspaces.

infosecurity-magazine.com → (opens in a new tab)

Accountable partial understanding: securing the epistemic attack surface of agentic systems (opens in a new tab)

arXiv cs.CR (all) ·Josiah Dykstra ·fetched 20 Oct 2026, 07:32 UTC Must read Research

Why readPresents a taxonomy and theoretical framework for securing the epistemic attack surface and delegated authority of agentic AI systems.

The paper analyzes how delegating decision-making and interpretation to AI agents creates an epistemic attack surface where partial understanding increases system risk. It outlines four attack vectors targeting intent, evidence, authority scope, and supervisor comprehension, then proposes Accountable Partial Understanding (APU) with six operational criteria to enforce bounded delegation and visible residual risk.

arxiv.org → (opens in a new tab)

Auditor-Assisted Summary-Channel Verification for Hosted LLM Identity Substitution (opens in a new tab)

arXiv cs.CR (AI) ·Ziyi Zhang, Ziyao Wang, Guoheng Sun, Ang Li ·fetched 20 Oct 2026, 23:32 UTC Research

Why readIntroduces SUMMARK, a watermarking framework that uses reasoning summaries to detect when hosted LLM providers secretly substitute claimed models with smaller alternatives.

Researchers developed SUMMARK, an auditor-assisted summary-channel watermarking scheme designed to detect model identity substitution in hosted LLM services without requiring access to weights or provider infrastructure. During setup, an auditor trains lightweight summary adapters and establishes identity-specific vocabulary carriers; at inference time, statistical testing on user-visible reasoning summaries confirms if the claimed model generated the output. Tests across multiple model families demonstrated reliable detection of cross-family and cross-scale model swaps with low false-positive rates.

arxiv.org → (opens in a new tab)

Mal-AAEGAN: Adversarial Autoencoder with Double-Critic WGAN-GP for Malware Generation (opens in a new tab)

arXiv cs.CR (all) ·Nam Hoang Thanh, Trung Pham Duy, Lam Bui Thu ·fetched 20 Oct 2026, 23:32 UTC Research

Why readLearn how Mal-AAEGAN uses adversarial autoencoders and Wasserstein GANs to generate malware that bypasses machine learning detectors.

Researchers developed Mal-AAEGAN, an adversarial framework designed to craft malware samples capable of evading machine learning detectors. Operating in a continuous latent space with a double-critic architecture, the system generates effective evasive variants while preserving original binary API calls.

arxiv.org → (opens in a new tab)

Meta Muse exposes users' personal and business secrets when they share conversations (opens in a new tab)

Ars Technica Security ·Dan Goodin ·fetched 20 Oct 2026, 15:33 UTC

Why readDetails how predictable conversation-sharing URLs in Meta Muse allowed public enumeration of sensitive user prompts.

Researchers from Claroc discovered that Meta's Muse AI agent generated easily predictable Web links for shared conversations, allowing them to enumerate over 6,000 public links. The exposed sessions included sensitive personal details and business credentials. Meta subsequently suspended the sharing feature to overhaul URL generation security.

arstechnica.com → (opens in a new tab)

The Risk of AI-Powered Antidetect Browsers in Online Fraud and Influence Operations (opens in a new tab)

Check Point Research ·fetched 20 Oct 2026, 15:33 UTC

Why readDetails how operators integrate AI and MCP servers into antidetect browsers like Kameleo, AdsPower, and MultiLogin for automated fraud and influence operations.

Check Point Research analyzes how antidetect browsers obscure digital fingerprints to manage thousands of isolated social media accounts concurrently. Modern anti-detection tooling now embeds AI automation templates and Model Context Protocol (MCP) servers to feed LLM-generated content into accounts at machine speed. This infrastructure lowers the operational threshold for large-scale financial fraud, malware delivery, and coordinated influence campaigns.

research.checkpoint.com → (opens in a new tab)

VulnBump: Can AI Agents Fix Vulnerable Dependencies Without Breaking the Build? (opens in a new tab)

Snyk ·fetched 20 Oct 2026, 15:33 UTC Research

Why readIntroduces VulnBump, a benchmark evaluating AI agent reliability during security dependency upgrades across six package managers.

Snyk researchers evaluated AI agents on 307 real-world vulnerable dependency upgrades across six package ecosystems. While 94% of identified vulnerabilities had known fixes available, automated upgrades often failed build or testing stages due to breaking changes and transitive dependencies.

snyk.io → (opens in a new tab)

LeakBench: Auditing Reward Hacking through Prediction-Time Leakage in Clinical Modeling Agents (opens in a new tab)

arXiv cs.CR (AI) ·Xiaoyuan Bao, Jingsi Jiang, Jiao Shan, Yulong Cao ·fetched 20 Oct 2026, 15:33 UTC Research

Why readIntroduces an execution framework to audit autonomous clinical AI agents for prediction-time feature leakage and reward hacking.

Researchers released LeakBench, an audit suite designed to evaluate clinical AI agents for prediction-time data leakage and reward hacking across datasets like MIMIC-IV. Tests across DeepSeek Flash and GPT-6-astra showed consistent leakage of invalid features when unconstrained, whereas explicit admission constraints prevented the exploit path.

arxiv.org → (opens in a new tab)

Microsoft Targets Rogue AI Agents With Dynamic Permissions, Risk Scoring (opens in a new tab)

SecurityWeek ·Eduard Kovacs ·fetched 20 Oct 2026, 11:35 UTC

Why readMicrosoft is extending Project Perception to treat internal AI agents as potentially overprivileged insider threats, adding risk scoring and dynamic permission controls.

Microsoft announced expansion of its Project Perception security system to manage risks from internal enterprise AI agents. Recognizing that frontier models executing autonomous workflows can operate with excessive privilege and lack contextual judgment, the update introduces risk scoring and dynamic access restriction controls.

securityweek.com → (opens in a new tab)

RAGLeak: Supply-Chain Privacy Backdoors in Retrieval-Augmented Generation (opens in a new tab)

arXiv cs.CR (AI) ·Yuchen Chen, Wei Cheng, Yuan Xiao, Junda He ·fetched 20 Oct 2026, 07:32 UTC Research

Why readExplores RAGLeak, a supply-chain backdoor attack against RAG pipelines that silently exfiltrates private retrieved documents via reversible text obfuscation.

Existing data extraction attacks against Retrieval-Augmented Generation (RAG) systems rely on prompt injection or direct document repetition, making them easy to detect. RAGLeak introduces a backdoor mechanism where the generator reversibly obfuscates retrieved context, enabling a secondary LLM to recover private text from benign outputs. Tested across three model families, RAGLeak achieves up to 100% attack success rate while maintaining benign response utility and bypassing standard detection defenses.

arxiv.org → (opens in a new tab)

DHS bans DeepSeek products across US federal civilian agencies (opens in a new tab)

BleepingComputer ·Bill Toulas ·fetched 20 Oct 2026, 15:33 UTC Must read

Why readDetails a DHS directive ordering US federal civilian agencies to immediately remove and stop procuring DeepSeek products.

The US Department of Homeland Security issued a binding order instructing federal civilian agencies to remove DeepSeek software and third-party products integrating its models from government networks. Agencies must immediately halt new procurements and audit existing deployments for compliance.

Also covered byThe Register Security (opens in a new tab).

bleepingcomputer.com → (opens in a new tab)

ICANN testing tech to let law enforcement access non-public WHOIS info (opens in a new tab)

The Register Security ·fetched 20 Oct 2026, 03:36 UTC

Why readICANN is testing an urgent WHOIS request framework for law enforcement responding to critical infrastructure and life safety threats.

ICANN has begun testing authentication mechanisms designed to expedite law enforcement access to redacted WHOIS registrant details during emergencies. The framework applies to cases involving imminent threats to life, critical infrastructure, or child exploitation where GDPR and domain privacy protections would otherwise delay investigations.

theregister.com → (opens in a new tab)

Here’s What Telegram Giving Data to the U.S. Government Looks Like (opens in a new tab)

404 Media ·Joseph Cox ·fetched 20 Oct 2026, 19:36 UTC

Why readReveals the exact subscriber metadata fields provided by Telegram under US legal process.

Unsealed court records obtained via Court Watch show official response documentation sent by Telegram to US federal law enforcement. The disclosure confirms that Telegram hands over registered phone numbers, internal user IDs, and connected IP address logs when presented with valid court orders.

404media.co → (opens in a new tab)

Victory! Court Rules OPM Data Disclosure to DOGE Is Unlawful (opens in a new tab)

EFF Deeplinks ·Mario Trujillo ·fetched 20 Oct 2026, 23:32 UTC

Why readUnderstand the legal boundary set by a federal court regarding inter-agency sharing of federal employee PII.

A federal judge ruled that the Office of Personnel Management violated the Privacy Act by giving DOGE agents broad access to systems containing millions of employee records. The court found OPM breached provisions governing inter-agency transfers and safeguarding obligations. Specific remedies will be determined in a subsequent proceeding.

eff.org → (opens in a new tab)

Head of AI Security Roles Surge in Large Organizations (opens in a new tab)

Infosecurity Magazine ·fetched 20 Oct 2026, 23:32 UTC

Why readBenchmarking data on how large enterprises structure dedicated AI security leadership roles.

A survey from IANS and Artico Search indicates 65% of companies with over $10 billion in revenue have established a Head of AI Security position. Dedicated roles account for 18% of these organizations, while 47% combine the responsibility with existing executive security roles. Adoption drops significantly in organizations earning under $1 billion.

infosecurity-magazine.com → (opens in a new tab)

Require Police to Get a Warrant to Search ALPR Data (opens in a new tab)

EFF Deeplinks ·Adam Schwartz ·fetched 20 Oct 2026, 03:36 UTC

Why readExamines legal arguments for requiring law enforcement warrants when searching license plate reader databases.

The Electronic Frontier Foundation argues that state legislatures and courts must enforce strict warrant requirements for automated license plate reader data. The post outlines privacy risks associated with municipal surveillance infrastructure and unchecked police queries.

eff.org → (opens in a new tab)

EU cyber law won’t force fixes for fridge controllers hit in French attack (opens in a new tab)

The Record ·fetched 20 Oct 2026, 15:33 UTC Must read

Why readExamines the policy boundaries of EU cybersecurity mandates after operational attacks on French supermarket refrigeration units.

Cyberattacks against exposed Danfoss and Copeland industrial controllers forced supermarket refrigeration systems into continuous defrost cycles, causing tens of millions of euros in spoiled inventory across major French chains. While CERT-FR issued emergency alerts, EU cybersecurity regulations mandate breach reporting for end-of-life hardware without requiring vendors to provide security updates.

therecord.media → (opens in a new tab)

Ahold Delhaize: Cyberattack affects services across US and Europe (opens in a new tab)

BleepingComputer ·Sergiu Gatlan ·fetched 20 Oct 2026, 11:35 UTC

Why readRetail conglomerate Ahold Delhaize confirmed a cyberattack forced IT systems offline across its US and European supermarket brands.

Ahold Delhaize disclosed that a cybersecurity incident forced it to disconnect IT systems, impacting websites, mobile apps, and back-office services for brands including Stop & Shop, Albert Heijn, and Delhaize. While physical stores remain open, supply chain and administrative systems face ongoing disruptions.

Also covered byWBAL-TV (opens in a new tab),NBC4 Washington (opens in a new tab),ABC27 (opens in a new tab).

bleepingcomputer.com → (opens in a new tab)

California warns all public school students of data stolen in contractor breach (opens in a new tab)

The Record ·fetched 20 Oct 2026, 19:36 UTC

Why readCalifornia warned that a contractor breach exposed three years of public school student records statewide.

A cyberattack on testing contractor Educational Testing Service compromised personal and academic records for all California public school students enrolled over the past three years. Compromised data includes student names, addresses, state identifiers, testing accommodations, and assessment scores.

Also covered byKCRA (opens in a new tab),GovTech (opens in a new tab).

therecord.media → (opens in a new tab)

23andMe successor settles with California for $6 million over cybersecurity failings (opens in a new tab)

The Record ·fetched 20 Oct 2026, 19:36 UTC

Why read23andMe's corporate successor agreed to a $6 million settlement with California over a 2023 breach.

Chrome Holding Company agreed to pay $6 million to resolve California regulatory claims over security failures that led to 23andMe's 2023 breach affecting 7 million individuals. Pending bankruptcy court approval, the agreement allows affected customers to demand data deletion and destruction of biological samples.

Also covered byMountain View Voice (opens in a new tab),Los Angeles Times (opens in a new tab).

therecord.media → (opens in a new tab)

Google says services recovering after cloud outage hits UK banks and websites (opens in a new tab)

BBC Technology ·fetched 20 Oct 2026, 15:33 UTC

Why readDetails a Google Cloud London datacenter power failure that interrupted operations for major UK banks and public services.

A power incident at Google Cloud's europe-west2 datacenter complex in London disrupted digital operations for major UK financial institutions including Revolut, Lloyds, and HSBC. The outage also impacted logistics providers and UK court administration systems before services were restored.

bbc.co.uk → (opens in a new tab)

OpenAI Engineer Accused of Spying on Executives Ordered to Remain in Custody (opens in a new tab)

WIRED Security ·Paresh Dave ·fetched 20 Oct 2026, 03:36 UTC

Why readFormer OpenAI engineer accused of placing listening devices in executive conference rooms was denied bail due to flight risks.

A federal magistrate judge denied bail to former OpenAI engineer Abdulrahman Alfozan, who is accused of bugging company conference rooms. Prosecutors argued Alfozan presented a severe flight risk after the Saudi Arabian consulate offered to assist in funding his release.

wired.com → (opens in a new tab)

Germany's far-right AfD investigates publication of emails stolen in cyberattack (opens in a new tab)

Google News: incidents · Reuters ·fetched 20 Oct 2026, 11:35 UTC

Why readReports on the leak of stolen internal emails following a cyberattack on Germany's AfD political party.

Germany's far-right AfD party is assessing the blast radius of a cyberattack that resulted in the public release of exfiltrated internal emails. The incident underscores ongoing threat activity targeting political infrastructure and sensitive communications across Europe.

news.google.com → (opens in a new tab)

$80 Million Settlement Proposed to Resolve Conduent Data Breach Litigation (opens in a new tab)

Google News: incidents · The HIPAA Journal ·fetched 20 Oct 2026, 15:33 UTC

Why readReports an $80 million proposed settlement resolving class-action litigation from the Conduent data breach.

Litigation surrounding the Conduent data breach reached a proposed $80 million settlement. The agreement addresses claims from affected individuals and health plans following the widespread exposure of personal data.

news.google.com → (opens in a new tab)

Labcorp Reaches $2.3M Settlement with 44 Attorneys General Over Debt Collector Data Breach (opens in a new tab)

Google News: incidents · Holland & Knight ·fetched 20 Oct 2026, 07:32 UTC

Why readLabcorp agreed to a $2.3 million settlement with 44 state attorneys general over a breach originating at a third-party debt collector.

Labcorp finalized a $2.3 million settlement with 44 state attorneys general resolving investigations into a historical data breach linked to a debt collection vendor. The enforcement highlights state-level regulatory scrutiny over third-party supply chain liabilities.

news.google.com → (opens in a new tab)

Connecticut Fines Frontier $625,000 Over Data Breach (opens in a new tab)

Google News: incidents · Broadband Breakfast ·fetched 20 Oct 2026, 07:32 UTC

Why readConnecticut state regulators fined Frontier Communications $625,000 following an investigation into a corporate data breach.

Frontier Communications was ordered to pay $625,000 to resolve regulatory enforcement actions in Connecticut stemming from a security breach. The penalty illustrates sustained state-level enforcement action against telecom operators failing to protect customer records.

news.google.com → (opens in a new tab)