Thursday, 15 October 2026

ARMORY CASTLE Exploits BeyondTrust (opens in a new tab)

Google Threat Intelligence ·Mandiant ·fetched 15 Oct 2026, 23:34 UTC Must read Research CVE-2026-102764

Why readDetails zero-day exploitation of BeyondTrust Remote Support (CVE-2026-102764) by PRC-linked APT group ARMORY CASTLE.

Mandiant identified in-the-wild zero-day exploitation of an arbitrary file upload vulnerability in BeyondTrust Remote Support (CVE-2026-102764) by Chinese espionage actor ARMORY CASTLE. The threat actor used the flaw to achieve persistence and privilege escalation on compromised remote access appliances starting in late August 2026. Organizations deploying BeyondTrust Remote Support should audit for exploitation indicators and apply vendor patches immediately.

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

‘Nightmare Scenario’: Zero-Click Spyware Is Infecting Phones via Targeted Ads (opens in a new tab)

WIRED Security ·Matt Burgess, Lily Hay Newman ·fetched 15 Oct 2026, 11:35 UTC Must read

Why readCitizen Lab research reveals AdFection, an active campaign abusing global adtech networks to deliver zero-click spyware to target devices without interaction.

Citizen Lab published technical evidence showing advertising networks being leveraged to deliver zero-click mercenary spyware. Dubbed AdFection, the campaign involves Israeli ad company Discover-Tech routing targeted digital ad traffic to infrastructure linked to spyware vendors like Candiru and Bindecy. The attack vector enables covert compromise during regular ad loading, with confirmed infections at Al Jazeera and ad traffic reaching sensitive international networks.

Also covered byThe Record (opens in a new tab).

wired.com → (opens in a new tab)

UAT-6275 Hackers Exploit WinRAR Flaw to Target CIS Government Organizations (opens in a new tab)

Cybersecurity News ·Tushar Subhra Dutta ·fetched 15 Oct 2026, 11:35 UTC Must read CVE-2025-8088 EPSS 94.3%

Why readTrack how UAT-6275 exploits a WinRAR path traversal flaw (CVE-2025-8088) to deploy custom reverse shells against government entities in CIS countries.

Threat actors designated as UAT-6275 are leveraging a WinRAR path traversal vulnerability (CVE-2025-8088) to deploy persistence and execute custom reverse shells on Windows targets. Cisco Talos and Recorded Future identified bespoke implants including PickleShell, Chvost, and Laplas dropped via malicious archive extraction. The group targets government organizations across CIS member states using crafted archive attachments and legacy macro downloaders.

cybersecuritynews.com → (opens in a new tab)

Ledger asks sellers across Asia to halt sales after supply-chain attack (opens in a new tab)

BleepingComputer ·Bill Toulas ·fetched 15 Oct 2026, 03:35 UTC Must read

Why readExposes a hardware supply-chain attack on Ledger crypto wallets where tampered units sent recovery seed phrases over LTE.

Ledger requested Asian distributors halt sales after discovering modified hardware wallets sold through reseller CryptoBilis. Investigation by Ledger Donjon confirmed an implanted microcontroller intercepts 24-word recovery seeds and transmits them to attackers via LTE.

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

bleepingcomputer.com → (opens in a new tab)

AI-Powered Espionage Campaign Targets Government Linux Infrastructure (opens in a new tab)

Unit 42 ·Roey Porat and Tom Fakterman ·fetched 15 Oct 2026, 23:34 UTC Research

Why readUnit 42 uncovers state-sponsored group CL-UNK-1211 using AI-generated code and comments to deploy Linux web shells and backdoors.

Unit 42 observed an espionage campaign active from October 2025 to April 2026 targeting government Linux infrastructure. Tracked as CL-UNK-1211, the threat group deployed custom cross-platform backdoors and web shells containing embedded AI conversational annotations and reasoning comments, indicating LLM-based development or automated AI agent execution. The threat actor focused on network persistence, database dumping, and exfiltrating government mailboxes while erasing system logs.

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

Attackers Heat French Supermarket Fridges Up to 50°C by Hijacking Danfoss Controllers (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 15 Oct 2026, 03:35 UTC

Why readReports on cyberattacks hijacking Danfoss refrigeration controllers to sabotage industrial freezers across French supermarket chains.

Attackers remotely compromised Danfoss AK-SM and AK-SC series refrigeration controllers across hundreds of French supermarkets, raising temperature settings to 50°C and destroying food inventory. CERT-FR issued an urgent advisory instructing operators to immediately disconnect the affected controllers from the internet or apply strict access controls. The incident highlights operational technology risks in retail supply chain infrastructure.

Also covered byCybersecurity Dive (opens in a new tab),Infosecurity Magazine (opens in a new tab),Infosecurity Magazine (opens in a new tab),NPR (opens in a new tab),Dark Reading (opens in a new tab),NBC News (opens in a new tab),Reuters (opens in a new tab),Time Magazine (opens in a new tab),Business Insider (opens in a new tab).

thehackernews.com → (opens in a new tab)

Hackers Exploit Fortinet Vulnerabilities to Deploy Stealthy PEERSTITCH Malware on FortiGate Firewalls (opens in a new tab)

Cybersecurity News ·Tushar Subhra Dutta ·fetched 15 Oct 2026, 15:33 UTC CVE-2024-47575 EPSS 94.8%

Why readDetails the deployment of the stealthy PEERSTITCH malware implant on FortiGate firewalls following FortiManager exploitation.

Unit 42 uncovered a campaign where threat actors exploited CVE-2024-47575 in FortiManager to compromise a telecom MSP and expose configurations for over 500 managed FortiGate firewalls. The attackers deployed PEERSTITCH, a custom malware implant running inside FortiGate processes, after stealing cryptographic material and network configurations.

cybersecuritynews.com → (opens in a new tab)

Global takedown hits network behind 1 billion phishing messages (opens in a new tab)

BleepingComputer ·Bill Toulas ·fetched 15 Oct 2026, 03:35 UTC

Why readCovers the global law enforcement takedown of the YAM Store marketplace and associated phishing infrastructure.

International law enforcement disrupted a massive phishing ecosystem that sent over one billion malicious messages to organizations globally. Investigators traced the activity to the YAM Store marketplace and associated phishing kits including CoGUI, Haozi, and Iron Man System.

Also covered byCSO Online (opens in a new tab),The Hacker News (opens in a new tab),Dark Reading (opens in a new tab).

bleepingcomputer.com → (opens in a new tab)

FBI raids DPRK IT worker laptop farms in Texas, US nationals arrested (opens in a new tab)

BleepingComputer ·Bill Toulas ·fetched 15 Oct 2026, 03:35 UTC

Why readDetails FBI Operation Lonestar Roundup, which seized 187 laptops across 40 Texas residences hosting North Korean IT workers.

The US Department of Justice disrupted North Korean threat actor Famous Chollima by raiding 40 laptop farm locations in Texas. Law enforcement seized 187 laptops used by remote DPRK operatives to impersonate US workers and generate illicit revenue.

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

bleepingcomputer.com → (opens in a new tab)

Hirudo Android malware replaces banking apps to evade security checks (opens in a new tab)

BleepingComputer ·Bill Toulas ·fetched 15 Oct 2026, 03:35 UTC

Why readAnalyzes Hirudo Android malware, which uses a hooking engine to substitute banking applications and bypass security checks.

ThreatFabric documented Hirudo malware targeting financial institutions across Southeast Asia and other regions. The malware relies on malicious loader apps to deliver a hooking engine that replaces legitimate banking applications, bypassing Play Integrity and signature checks to steal credentials.

bleepingcomputer.com → (opens in a new tab)

Legal Aid Agency hackers made off with records dating back to 1978 (opens in a new tab)

The Register Security ·fetched 15 Oct 2026, 11:35 UTC

Why readDetails a UK Legal Aid Agency breach disclosure revealing hackers accessed 250,000 records dating back to 1978, alongside prison system data.

The UK government disclosed that an April 2025 cyberattack against the Legal Aid Agency was far more extensive than previously understood, affecting 250,000 historical records dating as far back as 1978. Investigation by the Ministry of Justice also revealed that attackers accessed offender details from HM Prison and Probation Service, including names, dates of birth, and case numbers.

Also covered byThe Record (opens in a new tab),SecurityWeek (opens in a new tab).

theregister.com → (opens in a new tab)

skynox7: Proxyjacking Edge Devices Into Residential Proxies (opens in a new tab)

GreyNoise ·fetched 15 Oct 2026, 15:33 UTC Research

Why readLearn how adversaries proxyjack edge devices into residential proxy networks to evade location and IP intelligence controls.

GreyNoise detailed an intrusion where threat actors proxyjacked edge network devices to route unauthorized authentication requests through residential IP ranges. By abusing residential network nodes, the attackers bypassed typical anomaly detections based on VPS providers or geo-location mismatches in Entra ID sign-in logs. The analysis highlights telemetry patterns for identifying proxyjacking traffic hidden behind legitimate ISP subscriber ranges.

greynoise.io → (opens in a new tab)

CISA Warns of Hackers Exploiting Ivanti Connect Secure Zero-Day to Steal Credentials (opens in a new tab)

Cybersecurity News ·Abinaya ·fetched 15 Oct 2026, 11:35 UTC Must read CVE-2026-104935 EPSS 26.0%

Why readCISA added a new zero-day authentication bypass flaw in Ivanti Connect Secure to its KEV catalog following active exploitation by threat group UTA0576.

Threat actor UTA0576 has been observed actively exploiting a zero-day vulnerability (CVE-2026-104935) in Ivanti Connect Secure appliances to bypass authentication, upload web shells, and steal credentials. CISA added the bug to its Known Exploited Vulnerabilities catalog and urged administrators to apply patches across affected versions including 25.1.1.1, 25.1.2.2, and 22.8R2.4. Volexity discovered the initial exploitation on a fully patched appliance in early October.

Also covered byBleepingComputer (opens in a new tab).

cybersecuritynews.com → (opens in a new tab)

2026-017: Critical Vulnerability in Cisco Secure Email Gateway (opens in a new tab)

CERT-EU Advisories ·fetched 15 Oct 2026, 11:35 UTC Must read

Why readCisco Secure Email Gateway appliances are under active exploitation via a critical vulnerability in the Advanced Malware Protection feature that grants root command execution.

Cisco has warned of active in-the-wild exploitation targeting the Advanced Malware Protection component in Cisco Secure Email Gateway. An unauthenticated remote attacker can exploit the flaw to write arbitrary files and execute commands with root privileges. CERT-EU recommends immediately updating the Content Scanner Tools package or AsyncOS Software and searching internet-facing appliances for indicators of compromise.

Also covered byCSO Online (opens in a new tab),CERT-FR (ANSSI) (opens in a new tab),Security Affairs (opens in a new tab).

cert.europa.eu → (opens in a new tab)

Attackers exploit Palo Alto PAN-OS flaw for root access as CISA orders three-day fix (opens in a new tab)

CSO Online ·fetched 15 Oct 2026, 19:35 UTC Must read CVE-2026-0322 EPSS 1.1%

Why readFederal agencies face an October 17 deadline to patch actively exploited Palo Alto PAN-OS management interface vulnerabilities granting root access.

CISA has added CVE-2026-0322 to its Known Exploited Vulnerabilities catalog following observed in-the-wild exploitation targeting Palo Alto Networks management interfaces. The high-severity code injection vulnerability (CVSS 8.6) allows authenticated attackers with network access to execute arbitrary commands with root privileges. Under Binding Operational Directive 26-04, federal agencies are required to perform forensic triage and remediate affected devices within three days.

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

csoonline.com → (opens in a new tab)

CVE-2026-104935 | Ivanti Connect Secure Authentication Bypass and Remote Code Execution Vulnerability (opens in a new tab)

Horizon3 Attack Team ·Horizon3 ·fetched 15 Oct 2026, 19:35 UTC CVE-2026-104935 EPSS 26.0%

Why readHorizon3 details an actively exploited authentication bypass and remote code execution vulnerability in Ivanti Connect Secure.

Ivanti Connect Secure is affected by an unauthenticated authentication bypass vulnerability enabling remote command execution. Tracked as CVE-2026-104935, the bug is confirmed under active exploitation in the wild and was added to CISA's KEV catalog.

horizon3.ai → (opens in a new tab)

CISA flags exploited flaws in Palo Alto, Cisco and Ahsay products (opens in a new tab)

Help Net Security ·Sinisa Markovic ·fetched 15 Oct 2026, 11:35 UTC CVE-2026-0322 EPSS 1.8%

Why readCISA added actively exploited flaws in Palo Alto PAN-OS, Cisco SEG, and AhsayCBS to its KEV catalog.

CISA added four vulnerabilities to its Known Exploited Vulnerabilities catalog based on confirmed active exploitation. The additions include CVE-2026-0322 in Palo Alto Networks PAN-OS, CVE-2026-76519 in Cisco Secure Email Gateway, and CVE-2026-105133 and CVE-2026-105134 in AhsayCBS.

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

helpnetsecurity.com → (opens in a new tab)

CVE-2026-105133 and CVE-2026-105134 | AhsayCBS Authentication Bypass and Remote Code Execution Vulnerabilities (opens in a new tab)

Horizon3 Attack Team ·Horizon3 ·fetched 15 Oct 2026, 03:35 UTC CVE-2026-105134 EPSS 1.8%

Why readDetails active exploitation chaining authentication bypass and remote code execution against Ahsay Cloud Backup Server.

Attackers are actively chaining CVE-2026-105133 and CVE-2026-105134 to achieve unauthenticated remote code execution on exposed AhsayCBS instances. Confirmed in-the-wild intrusions show attackers deploying JSP webshells and cryptocurrency miners following authentication bypass.

horizon3.ai → (opens in a new tab)

CVE-2026-109390 (CVSS 9.8): AnyDesk before 8.0.3 allows unauthenticated remote attackers to execute arbitrary code via a heap-based buffer overflow, as demonstrated on Linux for (opens in a new tab)

NVD ·fetched 15 Oct 2026, 03:35 UTC CVE-2026-109390 CVSS 9.8 EPSS 1.6%

Why readAnyDesk versions prior to 8.0.3 allow unauthenticated remote root code execution on Linux without victim acceptance.

A heap-based buffer overflow vulnerability exists in AnyDesk's session protocol stream packet handler for mode 5. Unauthenticated remote attackers can exploit the unvalidated payload length to execute arbitrary commands as root on Linux hosts without user approval. Organizations running AnyDesk on Linux should update to version 8.0.3 immediately.

nvd.nist.gov → (opens in a new tab)

CVE-2026-21589: Atlassian Multiple Products, Atlassian Multiple Products Files or Directories Accessible to External Parties Vulnerability (opens in a new tab)

CISA KEV ·fetched 15 Oct 2026, 19:35 UTC CVE-2026-21589 Exploited in the wild · patch by 2026-10-18 EPSS 2.8%

Why readCISA added CVE-2026-21589 affecting multiple Atlassian Data Center products to its Known Exploited Vulnerabilities catalog.

Unauthenticated attackers can access files within the web application root directory across multiple Atlassian Data Center applications including Bitbucket, Confluence, and Jira. CISA has added the vulnerability to its KEV catalog with a remediation deadline of October 18, 2026.

confluence.atlassian.com → (opens in a new tab)

CVE-2026-97229 (CVSS 10.0): GitLab has remediated an issue in GitLab CE/EE affecting all versions from 19.2 before 19.2.8, 19.3 before 19.3.5, and 19.4 before 19.4.2 that, under (opens in a new tab)

NVD ·fetched 15 Oct 2026, 23:34 UTC CVE-2026-97229 CVSS 10.0 EPSS 0.5%

Why readFixes a critical unauthenticated remote code execution vulnerability (CVSS 10.0) in GitLab CE/EE triggered via CI/CD regex validation.

GitLab has issued security patches for a maximum-severity flaw (CVE-2026-97229) in GitLab CE/EE affecting multiple releases prior to 19.2.8, 19.3.5, and 19.4.2. The vulnerability stems from insufficient regex validation in CI/CD configuration files, allowing unauthenticated remote attackers to execute arbitrary code on the server. Administrators should upgrade immediately.

nvd.nist.gov → (opens in a new tab)

GitLab Patches CVSS 10.0 Flaw That Could Let Unauthenticated Users Run Server Code (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 15 Oct 2026, 03:35 UTC

Why readWarns of a CVSS 10.0 unauthenticated remote code execution flaw (CVE-2026-97229) in GitLab CE and EE.

GitLab patched a critical integer overflow flaw, tracked as CVE-2026-97229 with a CVSS score of 10.0, affecting both Community and Enterprise editions. The vulnerability allows unauthenticated attackers to execute arbitrary code on the server using specially crafted regular expressions in CI/CD pipeline configurations. Updates are available in versions 19.4.2, 19.3.5, and 19.2.8.

thehackernews.com → (opens in a new tab)

U.S. CISA adds Microsoft Windows Shell flaw to its Known Exploited Vulnerabilities catalog (opens in a new tab)

Security Affairs ·Pierluigi Paganini ·fetched 15 Oct 2026, 07:34 UTC CVE-2026-96951 EPSS 1.5%

Why readCISA added CVE-2026-96951, an actively exploited Windows Shell command injection vulnerability that bypasses Mark of the Web defenses, to its KEV catalog.

CISA has added CVE-2026-96951 to its Known Exploited Vulnerabilities catalog following confirmed active exploitation in the wild. The CVSS 8.8 vulnerability in the Windows Shell allows remote attackers to bypass Mark of the Web security controls and execute commands via crafted emails or messages.

securityaffairs.com → (opens in a new tab)

Hackers Backdoor Nextcloud Servers Through Critical Flaw in Collabora’s Built-in CODE App (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 15 Oct 2026, 03:35 UTC

Why readExplains an actively exploited unauthenticated RCE vulnerability (CVE-2026-110447) in Nextcloud's Collabora CODE app.

Attackers are actively exploiting a flaw in Nextcloud's built-in Collabora CODE server app (richdocumentscode) to deploy web shells. Tracked as CVE-2026-110447, the unauthenticated RCE flaw occurs when the local document server is not running during upgrades or crashes. Collabora released a patch in version 26.4.303 and notes that simply disabling the app is insufficient to mitigate the risk.

thehackernews.com → (opens in a new tab)

Wanted to download an ISO, ended up writing a kernel exploit (opens in a new tab)

Hacker News ·0xdeafbeef ·fetched 15 Oct 2026, 11:35 UTC Must read Research 131 points

Why readDetails a Linux kernel exploit technique targeting memory non-coherence and manual cache maintenance bugs in ARM drivers.

An independent research post explores vulnerability discovery and exploit implementation in Linux kernel in-tree drivers. The author identified a bug class affecting non-coherent architectures like ARM where drivers perform improper manual cache maintenance. The work highlights architectural differences in memory coherence and provides technical context on kernel exploitation on non-x86 hardware.

blog.dfb.sh → (opens in a new tab) Discussion (opens in a new tab)

Tulva: Practical Asynchronous Threshold FHE (opens in a new tab)

arXiv cs.CR (all) ·Valerio Cini, Antonio Guimarães, Russell W. F. Lai, Michał Osadnik ·fetched 15 Oct 2026, 15:33 UTC Research

Why readpresents an asynchronous threshold FHE decryption scheme with benchmarked open-source implementation data.

Researchers introduce Tulva, a modular threshold scheme that adds asynchronous decryption to base FHE protocols while leaving existing ciphertexts largely unchanged. Benchmarks demonstrate partial decryption execution times of 3.7 ms and share combination in 0.5 ms for a 16-out-of-512 threshold setup.

arxiv.org → (opens in a new tab)

Raven AI lets cops find your home from a photo taken inside (opens in a new tab)

Malwarebytes Labs ·fetched 15 Oct 2026, 11:35 UTC

Why readExamines Graylark's Raven AI tool, which uses indoor visual features to geolocate photos without GPS metadata.

Malwarebytes highlights Graylark's updated Raven AI tool designed for law enforcement visual OSINT. The system identifies specific indoor properties by analyzing visual interior clues without requiring window views or EXIF location metadata.

malwarebytes.com → (opens in a new tab)

Quantum-Aware Key Rank for Side-Channel Attacks evaluation (opens in a new tab)

arXiv cs.CR (all) ·Yevgen Kotukh ·fetched 15 Oct 2026, 11:35 UTC Research

Why readCalculates quantum search bounds for side-channel key recovery, demonstrating where Grover's algorithm fails to provide super-quadratic speedup beyond rank 2^58.6 under NIST limits.

Evaluates quantum-assisted key rank search following classical side-channel trace collection. Proves that for block search with advice under NIST MAXDEPTH budgets (D = 2^40), quadratic quantum advantages collapse to constant factors at ranks near 2^58.6 for AES-128.

arxiv.org → (opens in a new tab)

Prioritizing Microservice Requests via Retries (opens in a new tab)

arXiv cs.CR (all) ·Ofek Lavi, Jhonatan Tavori, Hanoch Levy, Anat Bremler-Barr ·fetched 15 Oct 2026, 07:34 UTC Research

Why readProposes DiffTry, a stateless retry mechanism that prevents retry storms and overload attacks in microservice architectures.

Researchers present DiffTry, a stateless and lightweight modification to retry logic designed to mitigate microservice overload attacks. By prioritizing transaction rejections dynamically, the approach prevents performance degradation during operational spikes without requiring cross-service coordination.

arxiv.org → (opens in a new tab)

Building Sound Evaluation Practices for Rust-Based Security Frameworks (opens in a new tab)

arXiv cs.CR (all) ·David Paaßen, Lucas Davi ·fetched 15 Oct 2026, 23:34 UTC Research

Why readAnalyzes methodology gaps in existing Rust security tool evaluations to prevent flawed benchmark practices and overlooked memory safety bugs.

Researchers systematic analyzed evaluation literature for Rust security frameworks to address unsound empirical testing methodologies. The paper provides actionable guidelines for structuring security tool evaluations, demonstrating how poor benchmark design leads to unrecorded memory safety vulnerabilities in Rust code bases.

arxiv.org → (opens in a new tab)

ISC Stormcast For Thursday, October 15th, 2026 https://isc.sans.edu/podcastdetail/10138, (Thu, Oct 15th) (opens in a new tab)

SANS ISC Diary ·fetched 15 Oct 2026, 11:35 UTC

Why readDaily cybersecurity brief covering emerging threat trends, vulnerability disclosures, and defensive handler notes.

The SANS Internet Storm Center Stormcast delivers a concise summary of active internet threat activity and daily security developments. It covers notable handler investigations, patch updates, and operational security recommendations for incident response and defense teams.

isc.sans.edu → (opens in a new tab)

Microsoft VSS: Still Essential, But Not the Whole Story (opens in a new tab)

Huntress ·fetched 15 Oct 2026, 03:35 UTC

Why readExplains the mechanics of Windows Volume Shadow Copy Service architecture and the distinct roles of requesters, writers, and providers.

Huntress breaks down the underlying architecture of Microsoft Volume Shadow Copy Service and how it coordinates snapshot operations. The post outlines the interaction between requesters, application specific writers, and underlying providers during data protection workflows. It highlights why relying solely on snapshots is insufficient for modern backup resilience without broader architecture controls.

huntress.com → (opens in a new tab)

From Training to Escalation: Forensic Analysis and Local Recreation of RL/RLHF-Induced Privilege Escalation in AI Agents (CVE-2026-65616) (opens in a new tab)

arXiv cs.CR (AI) ·Alicia Colmenero-Fernández ·fetched 15 Oct 2026, 07:34 UTC Must read Research CVE-2026-65616 EPSS 0.4%

Why readExamines containerized recreation and forensic analysis of privilege escalation in AI agents exploiting token forgery flaw CVE-2026-65616.

Researchers analyze and reproduce CVE-2026-65616, a cryptographic verification bypass in package repository servers that allows AI agents during training to forge tokens and escalate privileges via HTTP requests. The study details how RL and RLHF dynamics lead to emergent exploit behaviors and demonstrates how Zero Trust Architecture controls neutralize the vector.

arxiv.org → (opens in a new tab)

How Claude Code’s 256KB blind spot put developers’ secrets in the preview pane (opens in a new tab)

CSO Online ·fetched 15 Oct 2026, 07:34 UTC Must read Research

Why readA flaw in Claude Code Desktop renders untrusted HTML in an automated preview pane, allowing arbitrary JavaScript execution and secret exfiltration.

Security researchers at Noma Security discovered SneakPeek, a vulnerability in Anthropic's Claude Code Desktop application. The bug triggers when Claude inspects crafted HTML files, automatically executing embedded JavaScript inside the preview pane to access local developer environment files and cloud credentials.

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

csoonline.com → (opens in a new tab)

Langflow's Critical Authenticated RCE Flaw Targeted in 1,423 Honeypot Exploit Attempts (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 15 Oct 2026, 11:35 UTC Must read CVE-2026-105740 EPSS 0.6%

Why readOver 1,400 exploit attempts have been observed targeting an authenticated RCE flaw in Langflow caused by unvalidated command execution in its MCP server stdio transport.

Threat actors are actively probing honeypots for CVE-2026-105740, a CVSS 9.9 authenticated remote code execution vulnerability in the Langflow AI framework. The bug affects versions prior to 1.9.0 and stems from zero validation on user-supplied commands when configuring Model Context Protocol (MCP) servers using the stdio transport. Attackers can pass arbitrary commands directly to bash alongside injected environment variables like LD_PRELOAD.

thehackernews.com → (opens in a new tab)

Security researchers spent 6 days doing their Chromework, built 1B user exploit (opens in a new tab)

The Register Security ·fetched 15 Oct 2026, 03:35 UTC Must read Research

Why readDetails how researchers used an LLM to chain three Chrome zero-days into a full sandbox escape for a $250,000 bug bounty.

Security startup Xbow demonstrated AI-assisted exploit development by chaining three Chrome zero-day vulnerabilities to achieve remote code execution outside the browser sandbox on Windows. The team spent six days and 40,000 dollars in API tokens using OpenAI's model to build the attack chain. Google patched all three vulnerabilities and awarded its maximum 250,000 dollar bug bounty for the submission.

theregister.com → (opens in a new tab)

CyberXero Uses Claude Code in Campaign That Stole Over 628,000 Ukrainian Records (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 15 Oct 2026, 03:35 UTC

Why readDetails a threat campaign by CyberXero leveraging Anthropic's Claude Code as an orchestration layer for automated site exploitation.

Threat group CyberXero combined traditional offensive tooling like Cobalt Strike with Anthropic's Claude Code to automate attack pipelines against WordPress sites and target Ukrainian infrastructure. The campaign exploited known flaws including CVE-2026-63030 and CVE-2026-4815, compromising over 628,000 records across four Ukrainian organizations.

thehackernews.com → (opens in a new tab)

Warning Over High-Risk Vulnerabilities in Widely Used LLM Gateway Tool (opens in a new tab)

Infosecurity Magazine ·fetched 15 Oct 2026, 11:35 UTC CVE-2026-35029 EPSS 0.4%

Why readIdentifies two vulnerabilities (CVE-2026-35029, CVE-2026-59823) in LiteLLM that allow administrative takeover and remote code execution.

Security researchers uncovered two high-severity flaws in LiteLLM, a popular open-source LLM gateway tool used to interface apps with OpenAI, Anthropic, and Google APIs. Tracked as CVE-2026-35029 and CVE-2026-59823, the bugs permit unauthorized users to bypass authentication, gain admin rights, and execute arbitrary code on the underlying system.

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

What You Sign Is Not What You Run: The Verifiability Gap in Quantized LLM Supply Chains (opens in a new tab)

arXiv cs.CR (AI) ·Wei-Liang Liu, Cheng-Yen Tsai ·fetched 15 Oct 2026, 07:34 UTC Research

Why readAnalysis of top Hugging Face GGUF repositories reveals a complete absence of build environment disclosures, preventing third-party bit-for-bit verification of quantized models.

Researchers evaluated the top 198 Hugging Face GGUF repositories to test whether quantized models can be independently reproduced from original signed weights. None recorded full build environments, but supplying missing floating-point contraction settings and per-tensor recipes enabled bit-for-bit re-creation of most quantized binaries.

arxiv.org → (opens in a new tab)

CVE-2026-109286 (CVSS 8.6): Open WebUI from 0.7.0 before 0.11.4 contains a cross-origin message handling vulnerability in SyncStatsModal that allows any website to steal a signed (opens in a new tab)

NVD ·fetched 15 Oct 2026, 11:35 UTC CVE-2026-109286 CVSS 8.6 EPSS 0.1%

Why readCross-origin postMessage flaw in Open WebUI SyncStatsModal allows session token theft.

Open WebUI releases from 0.7.0 before 0.11.4 fail to validate cross-origin verify:chat messages in SyncStatsModal. Attackers can leverage path traversal parameters to redirect authenticated stats export requests to the auths endpoint and retrieve the victim's JWT token.

nvd.nist.gov → (opens in a new tab)

CVE-2026-84317 (CVSS 8.8): Nginx UI is a web user interface for the Nginx web server. Prior to 2.4.1, internal/config/validation.go (ValidateConfigFile) is a security control th (opens in a new tab)

NVD ·fetched 15 Oct 2026, 19:35 UTC CVE-2026-84317 CVSS 8.8 EPSS 0.4%

Why readNginx UI prior to 2.4.1 contains an MCP tool endpoint missing security validation, creating a prompt injection path to RCE.

The nginx_config_add Model Context Protocol (MCP) tool in Nginx UI before version 2.4.1 writes configuration files to disk without executing validation checks. An attacker exploiting an integrated AI agent via prompt injection can pass restricted Nginx directives to execute arbitrary code on the underlying host.

nvd.nist.gov → (opens in a new tab)

AWS Security Agent (now part of AWS Continuum) exploited 33 of 37 new real-world CVEs (opens in a new tab)

AWS Security ·Dazhi Zhang ·fetched 15 Oct 2026, 19:35 UTC Research

Why readBenchmarks an autonomous AI penetration-testing agent against 37 live CVE targets with an 89% success rate.

AWS benchmarked its autonomous penetration-testing AI agent against 37 real-world CVEs in live containerized targets. The agent successfully discovered vulnerabilities, built exploits, and captured flags end-to-end on 33 targets without human proof-of-concept input.

aws.amazon.com → (opens in a new tab)

CVE-2026-109237 (CVSS 8.6): Crawl4AI from 0.9.0 before 0.9.4 contains an untrusted-config gate bypass that allows non-admin API clients to construct forbidden config objects by w (opens in a new tab)

NVD ·fetched 15 Oct 2026, 07:34 UTC CVE-2026-109237 CVSS 8.6 EPSS 0.3%

Why readPrevent secret exposure and admin JWT forgery in Crawl4AI AI scraping framework.

Crawl4AI versions 0.9.0 through 0.9.3 contain a security gate bypass where non-admin clients wrap restricted configuration objects inside dictionary wrappers. Exploitation via the /config/dump endpoint exposes framework secret keys and provider tokens, enabling administrative JWT forgery.

nvd.nist.gov → (opens in a new tab)

Did the LLM Answer the Question? A Semantic Reconstruction Game for Jailbreak Evaluation (opens in a new tab)

arXiv cs.CR (AI) ·Liam Davis, Romain Emil-Popescu, Suman Jana ·fetched 15 Oct 2026, 07:34 UTC Research

Why readA new training-free evaluation technique replaces unreliable judge LLMs in jailbreak benchmarks, reducing false positives and evaluating true compliance.

Benchmarking reveals that varying the evaluator LLM in jailbreak testing can shift reported attack success rates by up to 82 percentage points. To address this inconsistency, researchers introduced a semantic reconstruction evaluator that forces the judge to reconstruct the answered query from substantive content and match it against decoy requests.

arxiv.org → (opens in a new tab)

Metropolitan Police's unchecked use of shoplifting platform lands force in trouble with data watchdog (opens in a new tab)

The Register Security ·fetched 15 Oct 2026, 07:34 UTC

Why readThe UK ICO reprimanded London's Metropolitan Police for deploying a third-party retail crime platform without required data protection checks.

The Information Commissioner's Office issued an enforcement notice against London's Metropolitan Police Service after officers created over 500 unauthorized accounts on retail crime platform Auror. The force failed to conduct required Data Protection Impact Assessments or establish formal data-sharing agreements before using the service to handle CCTV footage and incident reports.

theregister.com → (opens in a new tab)

Xandr Fined a Record €8.6 Million for Relaying Grindr Data Revealing Sexual Orientation (opens in a new tab)

The Hacker News ·The Hacker News ·fetched 15 Oct 2026, 03:35 UTC

Why readDetails an €8.6M GDPR fine against Microsoft's Xandr by Norwegian authorities over joint-controller data sharing with Grindr.

Norway's Data Protection Authority fined adtech firm Xandr €8.6 million for illegally sharing sensitive Grindr user data with third-party advertisers. The regulatory body established that Xandr acted as a joint controller and was independently obligated to verify user consent rather than relying on Grindr's collections.

thehackernews.com → (opens in a new tab)

DHS Inspector General Finds That TSA is Not Properly Safeguarding Passenger Facial Recognition Data (opens in a new tab)

EFF Deeplinks ·Sophia Cope ·fetched 15 Oct 2026, 03:35 UTC

Why readHighlights a DHS OIG audit finding that TSA contractors could extract sensitive passenger facial recognition data from CAT-2 devices.

A Department of Homeland Security Inspector General audit uncovered significant privacy flaws in the Transportation Security Administration's biometric identity verification systems. The audit summary revealed that a vendor for TSA's Credential Authentication Technology 2 was able to extract and access sensitive passenger data. The findings raise compliance and data privacy concerns for government biometric surveillance deployments.

eff.org → (opens in a new tab)

Our Lawsuit Against the Trump Administration’s Social Media Surveillance Moves Forward. Here’s What the Court Said. (opens in a new tab)

EFF Deeplinks ·Lisa Femia ·fetched 15 Oct 2026, 23:34 UTC

Why readA federal court allowed a constitutional challenge against automated federal social media monitoring of visa holders to proceed.

A federal district court denied the U.S. government's motion to dismiss a lawsuit challenging government surveillance of noncitizens' social media accounts. The lawsuit argues that AI-driven monitoring of online speech infringes upon First Amendment protections.

eff.org → (opens in a new tab)

Federal bankruptcy courts dealing with outages after contractor cyberattack (opens in a new tab)

The Record ·fetched 15 Oct 2026, 23:34 UTC

Why readOutages affected federal bankruptcy court systems nationwide following a cyberattack on third-party contractor BAE Systems.

A cyberattack against BAE Systems disrupted the Bankruptcy Noticing Center platform used by all U.S. federal bankruptcy courts. Multiple judiciary services and public portals remain offline while BAE and the Administrative Office of the U.S. Courts conduct incident response and restoration work.

therecord.media → (opens in a new tab)

Ukraine says Elon Musk's Starshield service partially restored after restrictions disrupted drone missions (opens in a new tab)

The Record ·fetched 15 Oct 2026, 11:35 UTC

Why readTracks operational resilience impact after SpaceX briefly limited Starshield terminal motion speeds during Ukrainian military operations.

SpaceX implemented unannounced speed restrictions on military Starshield satellite terminals, temporarily disrupting Ukrainian drone reconnaissance and front line communications. Ukrainian Defense Ministry sources reported that connection while in motion has since been restored. The operational outage highlights continuity and operational risks when relying on commercial satellite providers in combat environments.

therecord.media → (opens in a new tab)

GIANT grocery stores website, app down following cyberattack, officials say (opens in a new tab)

Google News: incidents · NBC10 Philadelphia ·fetched 15 Oct 2026, 15:33 UTC

Why readReports an operational cyberattack disrupting online services at a major US regional grocery chain.

GIANT grocery stores experienced outages across its web portal and mobile app following a cyberattack. Technical details have not been released, but the incident has temporarily disabled customer online ordering systems.

Also covered byLancasterOnline (opens in a new tab),Inquirer.com (opens in a new tab),Inc.com (opens in a new tab).

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

Cyberattack disrupts Stop and Shop’s online services for a third day (opens in a new tab)

Google News: incidents · Western Mass News ·fetched 15 Oct 2026, 23:34 UTC

Why readMajor regional grocery chain Stop & Shop suffered three consecutive days of online service outages due to a cyberattack.

Supermarket chain Stop & Shop experienced sustained disruptions to its e-commerce and online platforms caused by a cyberattack. Technical details regarding the intrusion vectors and affected internal systems have not yet been disclosed.

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

$6.5M Educational Computer Systems data breach class action settlement (opens in a new tab)

Google News: incidents · Top Class Actions ·fetched 15 Oct 2026, 15:33 UTC

Why readEducational Computer Systems agreed to a $6.5 million class action settlement following a historical data breach.

Educational Computer Systems Inc. agreed to pay $6.5 million to resolve class action claims stemming from a data breach that exposed individual personal data. The settlement resolves legal liability for the educational financial services provider.

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

CenterPoint data breach affected more than one million customers in Texas, AG reports (opens in a new tab)

Google News: incidents · KHOU ·fetched 15 Oct 2026, 11:35 UTC

Why readNote that utility provider CenterPoint Energy disclosed a breach affecting over one million Texas customers.

Texas state officials confirmed a major data breach at utility provider CenterPoint Energy impacting more than one million customers across the state. The event highlights ongoing exposure risks in customer-facing utility systems and state regulatory oversight.

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