Monday, 19 October 2026

Threat Brief: Active Exploitation of PAN-OS CVE-2026-0257 (Updated October 19) (opens in a new tab)

Unit 42 ·Andy Piazza and Unit 42 ·fetched 19 Oct 2026, 23:35 UTC Must read CVE-2026-0257 EPSS 96.9%

Why readRansomware groups including Qilin and Inc are actively exploiting an authentication bypass in PAN-OS GlobalProtect portals and gateways (CVE-2026-0257).

Unit 42 updated its threat advisory regarding active exploitation of CVE-2026-0257, an authentication bypass affecting Palo Alto Networks PAN-OS GlobalProtect components. Ransomware operators including Spikey Scorpius (Qilin) and Tarnished Scorpius (Inc) are attempting to leverage the flaw to establish unauthorized VPN connections. Organizations using affected PAN-OS versions should immediately check for gateway-connected events and review hunting indicators provided by Unit 42.

unit42.paloaltonetworks.com → (opens in a new tab)

CVE-2025-8088: Silent Lynx attacks governments via WinRAR (opens in a new tab)

Google News: enforcement · SecNews.gr ·fetched 19 Oct 2026, 07:35 UTC Must read CVE-2025-8088 EPSS 94.5%

Why readTrack active government-targeting campaigns exploiting WinRAR via CVE-2025-8088 with an EPSS score of 0.945.

Threat actors associated with Silent Lynx are targeting government entities using exploits for CVE-2025-8088 in WinRAR. High EPSS ratings indicate widespread exploitation activity, requiring immediate patching of WinRAR installations.

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

ConnectWise suspends Cloud Backup service after supply chain attack (opens in a new tab)

The Record ·fetched 19 Oct 2026, 23:35 UTC Must read

Why readRevoke OAuth permissions for ConnectWise Cloud Backup after attackers leveraged compromised credentials to access customer Microsoft 365 environments.

ConnectWise took its Cloud Backup service offline following unauthorized access that targeted customer Microsoft 365 environments. Security researchers at Huntress and Microsoft determined attackers compromised service credentials and abused high-level permissions, including Directory.ReadWrite.All, to read and modify cloud data. ConnectWise suspended affected services while investigations and remediation efforts continue.

Also covered byTruesec (opens in a new tab).

therecord.media → (opens in a new tab)

Rapid7 Analysis: NetScaler intrusions chaining CVE-2026-88771 and CVE-2026-88779 (opens in a new tab)

Rapid7 ·Rapid7 ·fetched 19 Oct 2026, 19:33 UTC Must read CVE-2026-88771 EPSS 0.6%

Why readUnpacks forensic details of Citrix NetScaler intrusions involving root persistence, Go recon tools, and FreeBSD proxy agents.

Rapid7 details post-exploitation activity on Citrix NetScaler ADC appliances where attackers gained root execution and exfiltrated configuration data. The threat actors established persistence via the crash-handling workflow, deployed PHP web shells, and leveraged custom Go reconnaissance tools and FreeBSD proxies.

rapid7.com → (opens in a new tab)

Feds Suspect Ransom Negotiator Ran ShinyHunters Extortion Site (opens in a new tab)

Krebs on Security ·BrianKrebs ·fetched 19 Oct 2026, 15:35 UTC

Why readUS federal investigators suspect an executive at a ransomware negotiation firm helped operate ShinyHunters' extortion site.

Federal investigators pursuing the ShinyHunters extortion group have linked executives at two ransomware negotiation firms to close ties with the hackers. One US-based negotiator reportedly helped run the site used to extort victims, following ShinyHunters' recent breach of the FBI's fbijobs.gov recruitment portal and extortion demands exceeding $70 million this year.

krebsonsecurity.com → (opens in a new tab)

Attackers Tampered with AOMEI Backupper's Update Servers for Minutes to Deliver a Backdoor (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 19 Oct 2026, 03:35 UTC

Why readLearn how attackers tampered with AOMEI Backupper update servers in short windows to push backdoors via unverified configuration files.

Threat actors compromised update infrastructure for AOMEI Backupper, modifying down_html.ini and Notifycfg.ini during 5- to 10-minute windows between May and July 2026. The program downloaded and executed backdoors over HTTPS without verifying digital signatures or checksum integrity. While AOMEI has remediated its servers, the vendor's fix does not automatically remove backdoors installed on compromised hosts.

Also covered byMalwarebytes Labs (opens in a new tab).

thehackernews.com → (opens in a new tab)

19th October – Threat Intelligence Report (opens in a new tab)

Check Point Research ·fetched 19 Oct 2026, 15:35 UTC

Why readWeekly report detailing major breach disclosures at CenterPoint Energy, Burger King Japan, and ConnectWise Cloud Backup.

Check Point summarizes active incidents, including a CenterPoint Energy data breach exposing PII of over 1 million Texas residents and a Burger King Japan breach affecting 11.18 million app users. Additionally, compromised application credentials in ConnectWise Cloud Backup permitted unauthorized access to at least 173 customer environments.

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

Supply Chain Zero Days: Closing the Gap Between Publication and Detection (opens in a new tab)

Wiz ·Tal Moriah ·fetched 19 Oct 2026, 15:35 UTC

Why readWiz research data shows supply chain attacks rose to 25% of major incidents in H1 2026, highlighting zero-day window risks before CVE publication.

Wiz research shows supply chain compromises accounted for 25% of significant security incidents in H1 2026, up from 10% in late 2025. The analysis details how attackers exploit automated dependency updates to breach pipelines before zero-day package compromises receive public CVE identifiers.

wiz.io → (opens in a new tab)

Two more Serbian opposition figures targeted with spyware, researchers say (opens in a new tab)

The Record ·fetched 19 Oct 2026, 19:33 UTC

Why readReports Pegasus spyware infections targeting Serbian opposition figures confirmed by Citizen Lab ahead of elections.

Citizen Lab confirmed that Pegasus spyware was used to compromise the phone of Vladan Đokić, rector of the University of Belgrade and supporter of student anti-government protests in Serbia. A second device belonging to prosecutor Rade Bajić is currently under forensic analysis. The spyware targeting comes immediately prior to Serbia's national elections.

therecord.media → (opens in a new tab)

Siemens Software Center Vulnerability Exploited in ClickFix Malware Attacks (opens in a new tab)

SecurityWeek ·Eduard Kovacs ·fetched 19 Oct 2026, 19:33 UTC

Why readWarns of ClickFix social engineering campaigns abusing authentic, signed Siemens Software Center binaries for DLL side-loading.

Threat actors are distributing malicious DLLs alongside legitimate, signed copies of Siemens Software Center to bypass detection during ClickFix lure attacks. By pairing trusted software with side-loaded payloads, the attack establishes persistence on Windows endpoints without triggering binary signature alerts. Organizations using Siemens industrial software should monitor for unexpected DLL loadings alongside signed Siemens executables.

securityweek.com → (opens in a new tab)

Hackers steal customer data from major Russian hosting provider FirstVDS (opens in a new tab)

The Record ·fetched 19 Oct 2026, 15:35 UTC

Why readRussian host FirstVDS suffered a data breach after attackers exploited a vulnerability in VMmanager 6 virtualization software.

Russian hosting provider FirstVDS confirmed attackers breached its infrastructure by exploiting a vulnerability in ISPsystem's VMmanager 6 software. The threat actors deployed malware to exfiltrate virtual server customer data, including API credentials, configuration files, and cryptocurrency wallets.

therecord.media → (opens in a new tab)

OnlyFans Copyright Takedowns Expose Compromised Websites (opens in a new tab)

Dark Reading ·Kristina Beek ·fetched 19 Oct 2026, 15:35 UTC

Why readAbuse of copyright takedown requests reveals compromised university and government web infrastructure used in SEO poisoning schemes.

Investigation by UpGuard shows threat actors exploiting compromised government and academic websites for adult content SEO manipulation. Attackers leverage OnlyFans copyright takedown filings to manipulate search engine indexation and hide compromised pages.

darkreading.com → (opens in a new tab)

CVE-2026-102764: Critical BeyondTrust Remote Support Authentication Bypass Exploited in the Wild (opens in a new tab)

Rapid7 ·Rapid7 ·fetched 19 Oct 2026, 15:35 UTC Must read CVE-2026-102764 EPSS 1.2%

Why readUnauthenticated authentication bypass in BeyondTrust Remote Support (CVE-2026-102764) is being actively exploited in the wild.

BeyondTrust disclosed a critical unauthenticated authentication bypass in Remote Support versions 26.2.3 and earlier, allowing remote file operations. CISA added CVE-2026-102764 to its KEV list following confirmed active exploitation against hosted and self-hosted instances. Affected organizations should immediately apply patch BT26-05-RS.

rapid7.com → (opens in a new tab)

Hackers Exploit Plex2Shell Flaws to Install Crypto Miners and Linux Proxy Backdoor (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 19 Oct 2026, 23:35 UTC Must read

Why readWarns of active in-the-wild exploitation chaining four Plex Media Server vulnerabilities one day after public PoC release.

Threat actors are actively exploiting a four-vulnerability chain named Plex2Shell to deploy cryptominers and Linux proxy backdoors on unpatched Plex Media Servers. The exploit chain leverages an unauthenticated SSRF in CVE-2026-96652 combined with auth bypass flaw CVE-2026-96654 to hijack owner tokens. Active exploitation began roughly 24 hours after a working proof of concept was published.

Also covered byArs Technica Security (opens in a new tab),Hacker News (opens in a new tab).

thehackernews.com → (opens in a new tab)

Linux KVM guest-host escape rated critical, may represent a new way to attack penguin-powered networking (opens in a new tab)

The Register Security ·fetched 19 Oct 2026, 03:35 UTC Must read CVE-2026-99999 EPSS 0.2%

Why readDetails CVE-2026-99999, a critical Linux KVM guest-host escape flaw triggerable via malformed IPv4 packets.

Security researcher Paulos Yibelo disclosed CVE-2026-99999 (PwnEscape), a CVSS 9.3 guest-host escape vulnerability in the Linux KVM hypervisor. The flaw enables an attacker inside a guest VM to execute code on the host using malformed IPv4 packets, creating severe risks for hypervisor isolation in cloud infrastructure.

Also covered byRisky Business News (opens in a new tab).

theregister.com → (opens in a new tab)

One parser, two bugs, many routers: Exploiting Broadcom WiFi across eCos, Linux (opens in a new tab)

Hacker News ·zoobab ·fetched 19 Oct 2026, 03:35 UTC Must read Research 57 points

Why readDetails a zero-click remote code execution flaw in Broadcom's Wi-Fi SDK affecting eCos and Linux access points.

A long-standing flaw in Broadcom's proprietary 802.11 implementation allows unauthenticated remote attackers to achieve arbitrary code execution over Wi-Fi without knowing the network key. Broadcom provided an incomplete fix in 2018 and recommended migrating to hostapd, leaving legacy eCos and Linux chipsets (including BCM3380, BCM3383, BCM3390, and BCM68xx) permanently vulnerable. The author provides a YARA rule and fingerprinting script to help defenders identify affected devices.

Also covered bySecurity Affairs (opens in a new tab).

quentinkaiser.be → (opens in a new tab) Discussion (opens in a new tab)

Japan’s SiteGuard WAF flaw CVE-2026-106931 is already being exploited (opens in a new tab)

Security Affairs ·Pierluigi Paganini ·fetched 19 Oct 2026, 19:33 UTC CVE-2026-106931 EPSS 0.7%

Why readUrgent advisory on actively exploited OS command injection vulnerability CVE-2026-106931 in SiteGuard Server Edition WAF.

EG Secure Solutions released an alert confirming in-the-wild exploitation of CVE-2026-106931, an OS command injection flaw in the SiteGuard web application firewall. The bug resides in the default statistics feature and allows local low-privilege attackers to execute commands with root or SYSTEM privileges. Affected installations include Apache, Nginx, and IIS deployments prior to version 8.00 Update 5, requiring immediate updating.

securityaffairs.com → (opens in a new tab)

CVE-2026-112005 (CVSS 9.2): WhoDB 0.104.0 before 0.135.0 contains a path traversal vulnerability that allows remote attackers to read and write arbitrary host files by running Du (opens in a new tab)

NVD ·fetched 19 Oct 2026, 23:35 UTC CVE-2026-112005 CVSS 9.2 EPSS 0.6%

Why readUpgrade WhoDB to 0.135.0 to fix a path traversal flaw allowing unauthenticated file read and write via DuckDB functions.

WhoDB versions 0.104.0 through 0.134.9 contain a path traversal vulnerability in the DuckDB query editor interface. Attackers can connect without credentials and execute DuckDB functions like read_text, COPY TO, or ATTACH to read session keys and database credentials or overwrite system files.

nvd.nist.gov → (opens in a new tab)

CVE-2026-85746 (CVSS 9.9): MariaDB server is a community developed fork of MySQL server. From 10.6.1 until 10.6.28, 10.11.19, 11.4.13, 11.8.9, 12.3.3, and 13.0.2, ST_AsText can (opens in a new tab)

NVD ·fetched 19 Oct 2026, 11:36 UTC CVE-2026-85746 CVSS 9.9 EPSS 0.6%

Why readMariaDB ST_AsText function contains a memory corruption flaw during nested GeometryCollection processing that enables remote code execution.

MariaDB versions 10.6.1 through 10.6.27, 10.11.19, 11.4.13, 11.8.9, 12.3.3, and 13.0.2 contain a heap buffer overflow in ST_AsText during recursive WKT serialization. A low-privileged database user can submit a crafted GIS object to corrupt memory and potentially achieve remote code execution. Upgrades have been released across affected stable branches.

nvd.nist.gov → (opens in a new tab)

CVE-2026-111882 (CVSS 8.6): phpMyFAQ 3.2.0-alpha before 4.1.9 and 4.2.0-alpha before 4.2.0-beta contains an authentication bypass vulnerability that allows IdP-authenticated user (opens in a new tab)

NVD ·fetched 19 Oct 2026, 23:35 UTC CVE-2026-111882 CVSS 8.6 EPSS 0.4%

Why readUpgrade phpMyFAQ to 4.1.9 to fix a two-factor authentication bypass affecting Keycloak and Entra ID SSO callbacks.

phpMyFAQ versions prior to 4.1.9 contain an authentication bypass vulnerability during OIDC SSO processing. Users authenticating via Keycloak or Entra ID callbacks can bypass required TOTP two-factor authentication steps, granting full session access even on 2FA-protected administrator accounts.

nvd.nist.gov → (opens in a new tab)

CVE-2026-111927 (CVSS 8.7): Coder before 2.29.20, 2.33.0 before 2.33.12, 2.34.0 before 2.34.8, and 2.35.0 before 2.35.4 contains an improper hostname validation flaw in GetWorksp (opens in a new tab)

NVD ·fetched 19 Oct 2026, 19:33 UTC CVE-2026-111927 CVSS 8.7 EPSS 0.2%

Why readA regex prefix matching bug in Coder workspace proxies enables theft of application connection keys via domain matching.

Coder versions prior to 2.29.20, 2.33.12, 2.34.8, and 2.35.4 perform prefix-based SQL pattern matching during workspace proxy host validation. Attackers registering domains that mirror valid proxy prefix strings can deceive the proxy trust check. Luring an authenticated user to a malicious auth-redirect endpoint allows attackers to harvest session tokens and access workspace applications.

nvd.nist.gov → (opens in a new tab)

CVE-2026-112006 (CVSS 8.3): OpenTelemetry opentelemetry-exporter-otlp-proto-http 1.5.0 before 1.45.0 contains an improper certificate validation vulnerability that disables TLS v (opens in a new tab)

NVD ·fetched 19 Oct 2026, 23:35 UTC CVE-2026-112006 CVSS 8.3 EPSS 0.2%

Why readUpdate OpenTelemetry OTLP HTTP exporter packages to 1.45.0 to prevent silent TLS certificate verification bypasses caused by empty environment variables.

The OpenTelemetry OTLP HTTP exporter library completely disables TLS certificate validation when the OTEL_EXPORTER_OTLP_CERTIFICATE variable is set to an empty string. Network attackers in a man-in-the-middle position can intercept telemetric data, trace spans, and sensitive authentication headers passed via export configurations.

nvd.nist.gov → (opens in a new tab)

CVE-2026-95859 (CVSS 8.7): libp2p-rust is the official Rust language implementation of the libp2p networking stack. Prior to 0.49.5, Gossipsub GRAFT control-message handling unc (opens in a new tab)

NVD ·fetched 19 Oct 2026, 07:35 UTC CVE-2026-95859 CVSS 8.7 EPSS 0.3%

Why readUnbounded memory allocation in libp2p-rust's Gossipsub protocol allows unauthenticated peers to crash nodes running versions prior to 0.49.5.

Versions of libp2p-rust prior to 0.49.5 fail to verify topic subscriptions when processing Gossipsub GRAFT control messages. Remote peers can flood nodes with novel GRAFT topic hashes, causing cumulative unbounded memory growth that leads to process failure.

nvd.nist.gov → (opens in a new tab)

CVE-2026-102552 (CVSS 8.2): Improper verification of the source of a communication channel in the DNS resolver in Amazon FreeRTOS-Plus-TCP before V4.4.2 might allow remote third- (opens in a new tab)

NVD ·fetched 19 Oct 2026, 07:35 UTC CVE-2026-102552 CVSS 8.2 EPSS 0.2%

Why readUpgrade FreeRTOS-Plus-TCP to V4.4.2 and enable source IP verification to defend against DNS cache poisoning.

The DNS resolver in Amazon FreeRTOS-Plus-TCP before version V4.4.2 fails to properly verify the source of incoming DNS communication channels. Remote attackers can forge DNS responses to inject arbitrary IP mappings into a device's DNS cache; mitigation requires updating to V4.4.2 and enabling ipconfigDNS_CHECK_REPLY_SOURCE_IP.

nvd.nist.gov → (opens in a new tab)

Behind Closed Doors: How Physical Penetration Testing Secures Sensitive Conversations (opens in a new tab)

Bishop Fox ·fetched 19 Oct 2026, 03:35 UTC

Why readExplores physical penetration testing methodology and social engineering vectors targeting executive conference rooms and sensitive spaces.

Physical security assessments frequently reveal that access controls and room monitoring fail during confidential meetings. The post details how red teams test access controls around sensitive areas and outlines practical steps to improve physical audits and visitor management.

bishopfox.com → (opens in a new tab)

Bad Diagnosis: Analyzing Consent and Data Collection Practices of Mobile Medical Apps (opens in a new tab)

arXiv cs.CR (all) ·Ethan Myers, Lorenzo De Carli, Shantanu Sharma, Ritwik Banerjee ·fetched 19 Oct 2026, 03:35 UTC Research

Why readQuantifies permission misuse and consent failures across 10,000+ Android mHealth apps.

Researchers analyzed over 10,000 mHealth applications on Google Play and found 69.15% requested permissions unrelated to stated functionality, including location (52.9%) and camera (50.5%). Furthermore, nearly 93% of the analyzed applications failed to present users with required regulatory consent forms.

arxiv.org → (opens in a new tab)

Beyond IoCs: Start Hunting Their Patterns (opens in a new tab)

detect.fyi ·Sergio Albea ·fetched 19 Oct 2026, 11:36 UTC

Why readPractical methodology for transforming static ClickFix IoCs into behavioral hunting patterns and Indicators of Attack across SIEM and XDR platforms.

The author presents a framework for threat hunters to move beyond static IoCs by analyzing execution context and behavioral patterns. Using the ClickFix attack chain as an example, the guide demonstrates how to extract repeatable behavioral rules that detect campaign variations across Defender XDR, Sentinel, and MISP.

detect.fyi → (opens in a new tab)

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

EFF Deeplinks ·Alexei Miagkov ·fetched 19 Oct 2026, 23:35 UTC

Why readEFF released Privacy Badger for Safari on macOS to block invisible third-party web trackers.

The Electronic Frontier Foundation has ported its open-source Privacy Badger browser extension to Safari on macOS version 26.5 and higher. The tool automatically detects and blocks non-consensual web tracking and data collection without relying on static blocklists.

eff.org → (opens in a new tab)

Bearer CLI: Free, source-available code security scanning tool (SAST) (opens in a new tab)

Help Net Security ·Anamarija Pogorelec ·fetched 19 Oct 2026, 07:35 UTC

Why readFree SAST scanner prioritizes source code vulnerabilities by tracing data flows to sensitive assets.

Bearer CLI is a source-available static application security testing tool designed to scan codebases without executing them. It maps sensitive data flows to prioritize high-risk code paths and reduce alert fatigue for development teams.

helpnetsecurity.com → (opens in a new tab)

The Q3 Vulnerability Pulse: Discovery ≠ Exploitation (opens in a new tab)

Cisco Talos ·Thorsten Rosendahl ·fetched 19 Oct 2026, 15:35 UTC CVE-2015-1000013 EPSS 2.0%

Why readCisco Talos analyzes why CVE growth (up 105% YoY) vastly outpaces real-world exploitation tracked by CISA KEV (up 37%).

Cisco Talos reviews Q3 vulnerability volume, noting that while total CVE assignments surpassed 70,000 and entered six-digit identifier territory, CISA KEV additions grew at a much lower rate. The post highlights the divergence between vulnerability discovery rate and actual exploited risk.

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

Flatpak – An Insecurity Nightmare (opens in a new tab)

Hacker News ·jannesan ·fetched 19 Oct 2026, 11:36 UTC 52 points

Why readCritiques Flatpak security posture and desktop Linux sandboxing limitations.

An architectural review outlines security trade-offs in Flatpak container packaging on desktop Linux systems. It details how runtime defaults and portal permissions can weaken process isolation compared to standard container platforms. The post also highlights operational risks related to static dependency bundling.

oro.gay → (opens in a new tab) Discussion (opens in a new tab)

Task-aware Kernel Stein certificates for Rényi differential privacy (opens in a new tab)

arXiv cs.CR (all) ·Leoni Carla Wirth, Tara Trautwein, Gesine Reinert ·fetched 19 Oct 2026, 07:35 UTC Research

Why readDemonstrates kernel Stein discrepancy bounds for certifying Rényi differential privacy guarantees in non-Gaussian mechanisms.

Researchers propose task-aware kernel Stein certificates to measure Rényi differential privacy in non-Gaussian settings. By aligning the kernel with specific privacy directions, the method offers exact identities for Gaussian and Laplace distributions alongside bounds for exponential mechanisms.

arxiv.org → (opens in a new tab)

Graph reconnaissance and ServiceNow user enumeration in a modern identity intrusion (opens in a new tab)

Sophos Threat Research ·fetched 19 Oct 2026, 23:35 UTC Must read Research

Why readDetails how attackers pivot from M365 account compromise via vishing to Microsoft Graph recon and ServiceNow user enumeration.

Sophos X-Ops analyzed an identity intrusion where attackers used vishing and self-service password reset abuse to compromise an M365 account and mint a Primary Refresh Token session. The actor then performed Microsoft Graph reconnaissance and enumerated ServiceNow user directories to move laterally across SaaS systems. The write-up outlines the specific post-compromise telemetry and audit events left across identity providers and SaaS platforms.

sophos.com → (opens in a new tab)

Investigating WMI Permanent Event Subscription Persistence (opens in a new tab)

Forensic Focus ·Forensic Focus ·fetched 19 Oct 2026, 15:35 UTC

Why readDetailed DFIR walkthrough for detecting, investigating, and remediating WMI ActiveScriptEventConsumer persistence mechanisms.

Forensic Focus covers the complete life cycle of WMI permanent event subscription persistence, detailing execution paths and memory artifacts. The guide includes Sigma detection rules and actionable steps for cleaning up malicious consumers.

forensicfocus.com → (opens in a new tab)

ABF-Bench: A Verifiable Benchmark for Agentic Blockchain Forensics (opens in a new tab)

arXiv cs.CR (AI) ·Xiaohang Yu, Hejia Geng, William Knottenbelt ·fetched 19 Oct 2026, 07:35 UTC Research

Why readIntroduces ABF-Bench, a verifiable evaluation framework for automated agents performing blockchain forensic investigations.

Researchers created ABF-Bench, an interactive benchmark that evaluates AI agents on complex cryptocurrency fund-tracing tasks built from real-world laundering cases. The framework requires agents to cite ledger records for every hop across mixers, DeFi protocols, and cross-chain bridges, providing a reproducible testbed for forensic automation.

arxiv.org → (opens in a new tab)

Atola TaskForce Gets A Massive Update Including A New Compare Module (opens in a new tab)

Forensic Focus ·Atola Technology ·fetched 19 Oct 2026, 11:36 UTC

Why readAtola TaskForce firmware 2026.10 adds a sector-by-sector drive Compare module and Seagate used-drive scam detection.

Atola Technology released firmware version 2026.10 for TaskForce and TaskForce 2 forensic imagers. The update introduces a Compare module that performs sector-level verification between source drives and images, alongside automated detection of drive modification scams on Seagate hardware and new Web API endpoints.

forensicfocus.com → (opens in a new tab)

CVE-2026-96620 (CVSS 9.2): Claude Agent SDK is a Python SDK for Claude Code. Prior to 0.2.121, the resume option is passed to the Claude CLI as a separate token without validati (opens in a new tab)

NVD ·fetched 19 Oct 2026, 19:33 UTC Must read CVE-2026-96620 CVSS 9.2 EPSS 0.6%

Why readUnvalidated session resume parameters in the Claude Agent Python SDK allow arbitrary command execution via MCP config flag injection.

Claude Agent SDK versions prior to 0.2.121 pass session resume identifiers directly to the Claude CLI without input sanitization. Passing a dash-prefixed string enables CLI flag injection, such as inserting --mcp-config to spawn a malicious stdio Model Context Protocol server. This results in arbitrary local command execution outside SDK permission controls whenever an application resumes sessions using untrusted input.

nvd.nist.gov → (opens in a new tab)

US bans DeepSeek AI from government use as tech tension with China continues (opens in a new tab)

BBC Technology ·fetched 19 Oct 2026, 19:33 UTC Must read

Why readThe US Department of Homeland Security issued a ban on DeepSeek AI technology across federal civilian agencies.

The Department of Homeland Security ordered federal civilian agencies to remove DeepSeek AI software and derivative models from government systems. The directive covers civilian agencies including Treasury, State, and Justice, while military components under the Department of Defense are excluded.

bbc.co.uk → (opens in a new tab)

Ignored, Not Refused: What Low Prompt-Injection Success Rates Hide in Small Tool-Using Agents (opens in a new tab)

arXiv cs.CR (AI) ·Farhan Zarif, Kazi Shahid Ahmed Galib, Synthia Akter Promi, Shabnam Sharmin Tarin ·fetched 19 Oct 2026, 23:35 UTC Research

Why readDemonstrates that low prompt-injection attack success rates in AI agents stem from ignored instructions rather than active security refusals.

An audit of 33,000 trajectories across eleven open-weight AI models revealed that 76 to 80 percent of failed prompt-injection attacks were silently ignored rather than explicitly blocked. Modifying system prompts on identical model weights changed attack success rates from 0.2 percent to 43 percent. The authors also found that 21.7 percent of safe runs failed to complete the user task, highlighting flaws in current evaluation benchmarks.

arxiv.org → (opens in a new tab)

Loud, Fast, and Everywhere: Anatomy of a Real AI-Driven Cloud Attack (opens in a new tab)

Wiz ·Kurt Giacchino ·fetched 19 Oct 2026, 19:33 UTC

Why readWiz CIRT details real-world intrusions where autonomous AI agents executed entire attack lifecycles across cloud and CI/CD environments.

Recent incident response investigations reveal a shift toward fully autonomous AI agents conducting cloud intrusions at machine speed. These systems perform initial access, credential harvesting, platform pivoting, and real-time script debugging without human intervention. Because the resulting activity spans identity providers, control planes, and workload logs, defenders must correlate fragmented telemetry across multiple cloud layers to detect agentic attacks.

wiz.io → (opens in a new tab)

MobileInject: Evaluating Reasoning-to-Action Security of MLLM-Based Mobile Agents under Cross-Modal Consistency Injection (opens in a new tab)

arXiv cs.CR (AI) ·Songping Wang, Peiyang Li, Yi Huang, Fengting Li ·fetched 19 Oct 2026, 19:33 UTC Must read Research

Why readIntroduces MobileInject, a benchmark for evaluating multimodal LLM mobile agents against cross-modal indirect injection attacks.

Researchers created MobileInject, an interactive benchmark with 108 executable scenarios across 11 mobile environments to evaluate MLLM agents under injection risks. The paper introduces Cross-Modal Consistency Injection, showing how cross-application content can steer autonomous agent behavior.

arxiv.org → (opens in a new tab)

Monitor-Validity Failures: LLM Safety Evidence Paths at Evolving Framework Boundaries (opens in a new tab)

arXiv cs.CR (AI) ·Junchi Liao ·fetched 19 Oct 2026, 15:35 UTC Research

Why readResearch demonstrates 18 evidence-path failure modes where LLM safety monitors recognize policy violations but fail to pass them to enforcement points.

Authors analyze interface-reliability breakdowns in LLM application frameworks where runtime safety monitor detections are stranded before enforcement decisions occur. Using a testing protocol called canary contracts across 11 agent safety projects, the study identified 18 confirmed evidence-path failures spanning multiple implementation ecosystems.

arxiv.org → (opens in a new tab)

Putting models to the secure coding test: Security rules (opens in a new tab)

Datadog Security Labs ·fetched 19 Oct 2026, 15:35 UTC Research

Why readTests whether supplying explicit security rules to AI coding agents measurably improves generated code safety.

Datadog Security Labs evaluated AI coding agents to measure the practical security impact of injecting pre-set system security rules into prompt contexts. The research compares vulnerability rates between baseline agent code generation and runs augmented with explicit secure coding guidelines.

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

HardEval: An Execution-Verified Benchmark for AI-Driven Security Hardening and Compliance Remediation (opens in a new tab)

arXiv cs.CR (all) ·Manohar Vellala, Soumya Banerjee, Sandip Roy, Ross Gore ·fetched 19 Oct 2026, 11:36 UTC Research

Why readEvaluates language model performance on OS security hardening using live execution of RHEL 8 STIG controls.

HardEval is a benchmark designed to test AI models on operating system hardening and compliance remediation tasks against Red Hat Enterprise Linux 8 STIG controls. Rather than evaluating code snippets, it runs models against live systems and verifies persistent configuration changes in SSH, audit rules, kernel parameters, and authentication using hidden OVAL checks.

arxiv.org → (opens in a new tab)

CVE-2026-49410 (CVSS 8.1): OpenHuman is an open-source agent harness available as a desktop application and Rust library. Versions before 0.56.0 allow cross-site request forgery (opens in a new tab)

NVD ·fetched 19 Oct 2026, 19:33 UTC CVE-2026-49410 CVSS 8.1 EPSS 0.2%

Why readUpdate OpenHuman agent harness to 0.56.0 to fix a local loopback CSRF vulnerability that allows attackers to hijack agent sessions and access memory.

OpenHuman versions before 0.56.0 are vulnerable to cross-site request forgery via the unauthenticated GET /auth/telegram loopback endpoint on port 7788. A malicious web page can trigger a login token consumption without origin checks, attaching the user's local agent session to the attacker's account and exposing agent memory and interactions.

nvd.nist.gov → (opens in a new tab)

TUPLE-Bench: Evaluating and Mitigating Privacy-Utility Trade-offs in Typed Decision Models (opens in a new tab)

arXiv cs.CR (all) ·Yiqu Yang, Qiaoyuan Zheng, Yang Tian, Yu Fan ·fetched 19 Oct 2026, 15:35 UTC Research

Why readResearch introduces TUPLE-Bench, demonstrating that typed decision AI models leak underlying operational causes 91% to 99% of the time.

Researchers evaluated nine typed decision AI models to measure privacy leakage when instructions explicitly request withholding specific causes. Even with cause candidates removed, model probability and confidence outputs enabled up to 75% accuracy in reconstructing restricted cause data.

arxiv.org → (opens in a new tab)

CVE-2026-112051 (CVSS 8.3): Xinference through 4.0.1 contains a path traversal vulnerability on Windows that allows unauthenticated attackers to write and delete arbitrary .json (opens in a new tab)

NVD ·fetched 19 Oct 2026, 23:35 UTC CVE-2026-112051 CVSS 8.3 EPSS 0.4%

Why readApply path validation controls or restrict Windows deployments of Xinference due to an unauthenticated JSON file write and delete vulnerability.

Xinference through version 4.0.1 on Windows contains a path traversal vulnerability via backslash sequences in the model_name parameter. Unauthenticated attackers can submit custom model registration requests to overwrite or delete arbitrary JSON configuration files accessible to the application process.

nvd.nist.gov → (opens in a new tab)

From Billion-Scale to Million-Scale Language Models: Operational Capabilities for Embedded Post-Exploitation (opens in a new tab)

arXiv cs.CR (all) ·Gabriel Hortea, Simon Pietro Romano, Juan Tapiador, Mario Varlese ·fetched 19 Oct 2026, 19:33 UTC Research

Why readEvaluates seven small language models from 20M to 1.5B parameters executing directory enumeration and backdoor generation tasks in resource-constrained environments.

Researchers benchmarked ultra-compact language models to assess their potential for local, offline post-exploitation in autonomous malware. The study found that offensive code generation capabilities do not scale linearly with parameter count; instead, specific task formulation and execution-feedback loops had a larger impact on success than prompt engineering. SLMs performed significantly better when generating high-resource Python code compared to low-resource Lua scripts.

arxiv.org → (opens in a new tab)

Encryption Is Getting Squeezed From All Sides (opens in a new tab)

WIRED Security ·Matt Burgess, Lily Hay Newman ·fetched 19 Oct 2026, 03:35 UTC

Why readTrack how expanding global surveillance and anti-encryption legislation is forcing services like Signal to exit regional markets like Canada.

Proposed legislation in Canada and other democratic nations requiring backdoor access or client-side scanning is driving encrypted service providers to alter operations or withdraw from markets. Signal announced plans to exit Canada over unacceptable demands on its open-source codebase, while Apple, Meta, DuckDuckGo, and NordVPN have raised formal objections.

wired.com → (opens in a new tab)

Exploited Vulnerabilities Typically Take Three Months to Patch, Moody’s Finds (opens in a new tab)

Infosecurity Magazine ·fetched 19 Oct 2026, 19:33 UTC

Why readMoody's report reveals a median 87-day window to remediate known exploited vulnerabilities across surveyed enterprises.

A study by Moody's benchmarked corporate patching velocity, finding that known exploited vulnerabilities remain unpatched for a median of 87 days, while non-exploited flaws take 107 days. The report notes that accelerating AI-driven vulnerability discovery is increasing pressure on patch management programs as vendor release volumes swell.

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

Bullhorn takes staffing platform offline to contain cyberattack (opens in a new tab)

BleepingComputer ·Bill Toulas ·fetched 19 Oct 2026, 19:33 UTC Must read

Why readMajor recruitment software provider Bullhorn took its platform offline globally following a cyberattack, impacting over 10,000 corporate clients.

Staffing software giant Bullhorn disabled access to its cloud platform to contain a cyber incident that disrupted job posting, candidate tracking, and recruitment operations worldwide. CrowdStrike has been engaged to assist with eradication and forensic restoration across the platform. The outage highlights supply-chain risks for enterprise HR and talent acquisition departments reliant on centralized SaaS management.

bleepingcomputer.com → (opens in a new tab)

Proofpoint lines up $6.3 billion deal for Varonis’ data security (opens in a new tab)

The Record ·fetched 19 Oct 2026, 11:36 UTC

Why readProofpoint acquires Varonis for $6.3 billion in cash to combine email, identity, and data security posture management capabilities.

Proofpoint has announced an all-cash $6.3 billion acquisition of data security firm Varonis. The combined company aims to offer unified visibility and data protection across enterprise environments and AI implementations.

therecord.media → (opens in a new tab)

Ahold Delhaize confirms 'malicious actions' behind cyber outages affecting grocery brands (opens in a new tab)

The Record ·fetched 19 Oct 2026, 19:33 UTC

Why readOutlines operational disruptions across US and European grocery brands following a confirmed cyberattack on Ahold Delhaize.

Multinational grocery retailer Ahold Delhaize confirmed that recent IT outages across its US and European brands were caused by a cyberattack. Online ordering and back-end systems were disrupted in the US, the Netherlands, and Belgium, leading the company to file a police report in the Netherlands. The retailer stated there is no current evidence of data theft, though forensic investigations remain ongoing.

therecord.media → (opens in a new tab)

Conduent Agrees to Pay $80m to Settle Data Breach Class Action (opens in a new tab)

Infosecurity Magazine ·fetched 19 Oct 2026, 23:35 UTC

Why readProvides benchmark settlement figures for major health and personal data breaches affecting millions of individuals.

Digital platform provider Conduent agreed to an $80 million class action settlement in federal court following a major 2024 breach affecting 62 million people. Court filings show the plaintiff class is divided into 40 million individuals whose data was stored in complex backup files and 22 million whose data was readily accessible. Funds will cover verified financial losses, credit monitoring services, and legal fees.

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

Ex-OpenAI worker accused of planting device had Saudi support, says court filing (opens in a new tab)

BBC Technology ·fetched 19 Oct 2026, 23:35 UTC

Why readCourt filings allege state-backed support in a physical bugging incident targeting OpenAI.

US prosecutors revealed in court that a former OpenAI employee accused of hiding an audio recording device in a company conference room received direct support from the Saudi government. The employee had frequent communication with a member of the Saudi royal family before his arrest. The incident highlights physical insider security threats targeting major artificial intelligence firms.

bbc.co.uk → (opens in a new tab)

Nikkei discloses third data breach in a month, 1.8M records exposed (opens in a new tab)

BleepingComputer ·Bill Toulas ·fetched 19 Oct 2026, 23:35 UTC

Why readTracks a third consecutive data breach disclosure at Nikkei Inc. exposing 1.83 million records via compromised dev credentials.

Nikkei Inc. disclosed its third cyber incident in a month after unauthorized access to a development system exposed 1.83 million records. Approximately 150,000 records contained personal identifying details including names, email addresses, and phone numbers. The intruder accessed the network on October 7 using stolen credentials.

bleepingcomputer.com → (opens in a new tab)

Huge DMV data breach impacted current and former Florida residents (opens in a new tab)

Google News: incidents · Ars Technica ·fetched 19 Oct 2026, 19:33 UTC

Why readA significant data breach at the Florida DMV exposed motor vehicle department records.

Ars Technica reported a data breach at the Florida Department of Highway Safety and Motor Vehicles impacting current and former state residents. The story covers public sector data exposure risk affecting government-issued identification records.

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

Judge denies WhatsApp’s bid to hold spyware maker in contempt for allegedly targeting users (opens in a new tab)

The Record ·fetched 19 Oct 2026, 23:35 UTC

Why readReview the outcome of WhatsApp's motion against NSO Group after a federal judge found insufficient proof of ongoing injunction violations.

US District Judge Phyllis Hamilton denied Meta-owned WhatsApp's attempt to hold spyware developer NSO Group in contempt of court. WhatsApp claimed NSO violated a 2025 injunction by conducting spearphishing attacks against users, but the court ruled WhatsApp failed to meet the required clear and convincing evidence standard. The ruling leaves NSO's prior $4.4 million penalty intact while rejecting additional sanctions.

therecord.media → (opens in a new tab)

California to recieve $6m settlement after 23andMe data breach (opens in a new tab)

Google News: incidents · The Guardian ·fetched 19 Oct 2026, 23:35 UTC

Why readDetails a $6 million state regulatory settlement following the 23andMe data breach.

California state authorities reached a $6 million settlement with 23andMe resolving regulatory investigations into its major data breach. The legal outcome establishes further financial and enforcement precedent for breach reporting and consumer data protection failures.

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

Cyberattack raised refrigerator temperatures in French supermarkets — Politico Europe (opens in a new tab)

Google News: incidents · UA.NEWS ·fetched 19 Oct 2026, 03:35 UTC

Why readReports on a cyberattack against French supermarket refrigeration infrastructure that impacted cold-chain supply operations.

A cyberattack targeted systems controlling refrigeration in French supermarkets, causing temperatures to rise and threatening food safety. The incident highlights operational technology vulnerability in retail supply chains and physical infrastructure.

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

Russian social media giant VK faces disruptions after data center drone strike (opens in a new tab)

The Record ·fetched 19 Oct 2026, 19:33 UTC

Why readDetails service disruptions to Russian platform VK following a physical drone strike on a Leningrad data center.

Users across Russia experienced widespread outages on social network VKontakte following an overnight drone strike on a data center in the Leningrad region. While VK denied service impact, network monitoring services recorded thousands of user disruption reports and diminished connectivity. The event highlights physical attack risks to data center infrastructure in active conflict zones.

therecord.media → (opens in a new tab)

iRhythm Data Breach Affects 8.2 Million Individuals (opens in a new tab)

Google News: incidents · The HIPAA Journal ·fetched 19 Oct 2026, 15:35 UTC

Why readA major data breach at cardiac monitoring vendor iRhythm has exposed the personal information of 8.2 million individuals.

Health tech company iRhythm disclosed a data breach impacting 8.2 million individuals. The incident represents one of the larger healthcare-related data exposures reported this year.

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