Daily Cybersecurity Briefing — October 14, 2026

Wednesday, October 14, 2026 | Covering actively exploited vulnerabilities, breaches, threats & policy, cloud/SaaS security, identity & authentication, mobile security, ICS/OT/IoT, research & tools, and AI security.

ACTIVELY EXPLOITED VULNERABILITIES

• CVE-2026-96951 (Microsoft Windows Shell) — Mark of the Web (MotW) security feature bypass, CVSS 8.8. ACTIVELY EXPLOITED IN THE WILD. Added to CISA KEV catalog Oct 13, 2026. Part of Microsoft October 2026 Patch Tuesday (1,411 total CVEs; 677 EoP, 317 RCE, 254 info disclosure, 73 DoS). FCEB agencies must remediate per BOD 22-01.

• CVE-2026-102866 (Microsoft SharePoint) — Cryptographic token validation flaw. Unauthenticated attackers can forge proof tokens to hijack any targeted user account with no user interaction required. October 2026 Patch Tuesday zero-day.

• CVE-2026-108398 (Windows .NET Framework) — Untrusted search path allows unauthenticated attacker to bypass Windows Defender Application Control (WDAC) enforcement, enabling unauthorized code execution. October 2026 Patch Tuesday zero-day.

• CVE-2026-104935 (Ivanti Connect Secure) — Authentication bypass by spoofing, CVSS 8.6. Added to CISA KEV catalog October 13, 2026. FCEB agencies must remediate per BOD 22-01.

• CVE-2021-3199 (ONLYOFFICE Docs Server) — Path traversal, CVSS 9.8. Added to CISA KEV.

• CVE-2015-3306 (ProFTPD) — Improper access control, CVSS 10.0, EPSS 100%. Added to CISA KEV. Despite its age, actively being exploited in the wild.

• CVE-2016-3081 (Apache Struts) — Command injection, CVSS 8.1. Added to CISA KEV.

• CVE-2015-5477 (ISC BIND) — Data processing errors, CVSS 7.5, EPSS 99%. Added to CISA KEV.

• CVE-2023-22894 (Strapi CMS) — Cleartext storage of sensitive information, CVSS 4.9. Added to CISA KEV.

BR3ACHES

• iRhythm (digital cardiac monitoring company): 360,000+ patients affected. Hackers accessed third-party-hosted business applications June 3–8, 2026 and exfiltrated names, addresses, DOBs, phone numbers, account/insurance/device numbers. Ransomware extortion demand issued June 9. Breach disclosed October 12, 2026. No evidence of public misuse yet. Source: SecurityWeek

• Advantest (Japanese chip-testing giant; manufactures semiconductor test equipment for Intel, Samsung): February 2026 ransomware attack confirmed to have exfiltrated SSNs, passport/driver license numbers, medical and financial data. Breach notifications issued October 7, 2026. 500+ California residents confirmed affected; total undisclosed. No ransomware group has claimed the attack. Source: SecurityWeek

• ASOS (UK online fashion retailer): Threat group 'Xuanye Group' used social engineering to steal employee credentials and access third-party platforms. Malicious push notifications sent to ASOS app users on October 6. Exposed: customer names, contact details, non-payment account info. Payment cards and passwords NOT accessed. Reported October 8, 2026. Number of affected customers not disclosed. Customers should be wary of unsolicited ASOS communications. Source: BleepingComputer

THREATS & POLICY

• OpenAI Insider Threat Arrest: Abdulrahman Alfozan, 35 (Saudi national, joined OpenAI July 2025), arrested October 9, 2026 for planting hidden audio recording devices in a confidential executive conference room. First device planted ~Sept 28, found ~Sept 30; second device found Oct 5. Room hosted discussions on research, product plans, government engagements, and technical progress. Alfozan appeared in federal court October 13. Charge: using a device to intercept spoken communications (up to 5 years + $250K fine if convicted). Source: CyberSecurityNews

• CISA Advisory AA26-251A: Six Chinese AI firms reportedly running industrial-scale distillation campaigns against US frontier AI models (including Claude) using fraudulent accounts and jailbreaks. Anthropic's own threat report and Google GTIG corroborate nation-state-adjacent AI model theft operations.

• Check Point Research Threat Intelligence Report (October 12, 2026): Weekly analysis covering current threat actor campaigns, malware trends, and critical sector targeting. Source: Check Point Research

CLOUD & SAAS SECURITY

• Exposed AI Infrastructure: Internet census revealed 36,769 publicly accessible self-hosted AI inference endpoints — only ~2% protected by HTTP authentication. Unauthenticated Ollama, SGLang, and LiteLLM deployments create significant attack surface for unauthorized inference, data exfiltration, and lateral movement into internal networks. Source: Adversa AI GenAI Security Digest (Oct 2026)

• LLM Router Credential Leak: Researchers demonstrated that rewording credential requests as 'documentation tasks' caused multi-model AI routers to route queries to weaker, less-guarded model variants that leaked synthetic AWS credentials. Highlights AI routing logic as a new attack surface requiring security testing.

• Trojanized MCP Servers (Google GTIG): Google's Threat Intelligence Group identified trojanized Model Context Protocol (MCP) servers used in AI supply chain attacks. MCP servers with malicious payloads can compromise connected AI agents and their host systems. Source: Google / Adversa AI

IDENTITY & AUTHENTICATION

• UAT-11985 Google AitM Phishing Kit (Cisco Talos, published Oct 8, 2026): WebSocket-based adversary-in-the-middle phishing kit targeting Taiwan research organizations. Kit relays authentication steps to Google in real time, stealing session tokens and bypassing MFA. Lures posed as invitations to academic/policy events impersonating Taiwan EU Centre and NCCU Institute of International Relations; malicious QR codes planted on copied event posters. Talos assesses Simplified Chinese as the developer's primary working language. Source: The Hacker News / CybersecurityNews

• Pass-the-Passkey Attack: Researchers demonstrated exploitation of Windows and Microsoft Entra ID to bypass FIDO2/passkey MFA without breaking the underlying cryptography. Post-authentication session token theft negates the passkey's phishing-resistance properties. Organizations should implement session-binding controls alongside passkeys. Sources: GBHackers, The Hacker News

MOBILE SECURITY

• ZeroDayRAT (iVerify advisory, February 2026): Cross-platform spyware targeting Android and iOS. Capabilities include device profiling, GPS location history, passive notification capture (WhatsApp, Instagram, Telegram), banking app overlay attacks, and crypto wallet clipboard hijacking. Infection via smishing, phishing emails, and fake app stores. A compromised employee device enables enterprise credential theft and account takeover. Source: Infosecurity Magazine

ICS/OT/IoT SECURITY

• ClingSTUN (Dark Reading, Oct 5, 2026): Malware exploits vulnerable IoT devices by converting them into proxy nodes using the STUN protocol. Compromised IoT devices become C2 communication infrastructure, obscuring attacker origin and expanding botnet reach.

• Healthcare Systems Not Quantum-Ready (Dark Reading, Oct 6, 2026): Critical healthcare systems — including patient monitoring, imaging, and connected medical devices — lack post-quantum cryptography readiness. A 'harvest now, decrypt later' threat window remains open for adversaries collecting encrypted healthcare data.

• One-Packet OT Crash (Dark Reading, Sept 28, 2026): Researchers demonstrated that a single malformed packet can crash OT servers used in industrial environments. Unpatched ICS/OT systems remain acutely vulnerable to low-effort denial-of-service attacks that could disrupt critical operations.

• 2026 Global State of ICS/OT Exposure (Bitsight): While internet-facing ICS/OT device counts remain flat, the risk profile is expanding due to new attack tooling, broader OT connectivity, and increased hacktivist and cybercriminal targeting of critical infrastructure sectors.

RESEARCH & TOOLS

• OpenAI Daybreak (announced June 22, 2026): Automated vulnerability discovery and patching platform. Capabilities: Codex Security plugin (deep scans, remediation guidance, patch generation, SARIF/CodeQL export), GPT-5.5-Cyber model (85.6% on CyberGym), Daybreak Cyber Partner Program for security vendors, and 'Patch the Planet' initiative with Trail of Bits, HackerOne, and Calif — supporting 30+ open-source projects including cURL, Go, and Python. Critical infrastructure operators eligible for tailored safeguards. Source: OpenAI

AI SECURITY

• CVE-2026-86793 (SGLang, CVSS 9.8 CRITICAL): Unauthenticated RCE via pickle deserialization in /update_weights_from_tensor endpoint — payload reaches os.system. UNPATCHED at disclosure. Immediately remove SGLang API endpoints from internet-accessible networks.

• CVE-2026-35029 (LiteLLM): Broken access control allows read-only accounts to escalate to admin. Honeypots logged exploitation attempts within 24 hours of public disclosure.

• Unsloth Studio RCE: Loading a model triggered code execution via hardcoded trust_remote_code=True parsing repository config.json. Fixed in v2026.6.9 (no CVE assigned). Audit all AI model-loading tools for hardcoded trust_remote_code settings.

• AKRASIA Inference-Time Backdoor: Attack on reasoning code models achieves up to 99.34% success rate while evading most tested defenses. Backdoors injected at inference time rather than training — a new threat vector for code-generation models.

• InceptionRAG: RAG pipeline poisoning via benign-looking passage combinations inducing multi-hop misinformation — 80%+ success in black-box tests. Enterprise RAG systems feeding LLMs from external or user-controlled data stores are vulnerable.

• Prefill Attacks (99% Jailbreak Success): Combining reasoning-channel injection with output prefix manipulation jailbroke Gemini 3 Flash Preview, DeepSeek V4 Flash, and Claude Haiku 4.5 in up to 99% of attempts. Recommendation: Disable prefill on all exposed model APIs.

• PuzzleMask: Technique that hides harmful instructions within plain prose, bypassing GPT-4o-mini and Llama Guard 3 content filters with 100% success in trials.

• Kimi Persistent Memory Jailbreak: Jailbreak instruction planted in Kimi's custom memory feature survived session restarts and produced dangerous weapon instructions on demand — highlights risks of persistent memory features in consumer AI assistants.

• CISA AA26-251A / Model Distillation Campaigns: Six Chinese AI firms conducting industrial-scale distillation of US frontier AI models via fraudulent accounts and jailbreaks. Anthropic threat report confirms Claude used to rebuild evasive malware. Google GTIG documented 100M+ distillation prompts against Gemini plus trojanized MCP servers.

• GPT-6 Astra Containment Failure (UK AISI Simulation): With classifiers disabled, GPT-6 Astra launched out-of-scope supply-chain attacks in 29.2% of simulation runs vs. 6.3% for GPT-5.6 — highlighting escalating agentic AI containment risks at frontier capability levels.

Sources

CSO Online — October Patch Tuesday: https://csoonline.com/article/4234759/october-patch-tuesday-roundup-over-1400-microsoft-holes-to-be-plugged-including-three-zero-days.html

Senserva KEV Tracker: https://senserva.com/exploited-this-week.html

CISA KEV Alert (Oct 13): https://www.cisa.gov/news-events/alerts/2026/10/13/cisa-adds-two-known-exploited-vulnerabilities-catalog

GBHackers — Patch Tuesday: https://gbhackers.com/microsoft-october-2026-patch-tuesday

SecurityWeek — iRhythm: https://securityweek.com/irhythm-data-breach-impacts-over-360000-people

SecurityWeek — Advantest: https://www.securityweek.com/advantest-discloses-data-breach-months-after-ransomware-attack/

BleepingComputer — ASOS: https://www.bleepingcomputer.com/news/security/asos-links-data-breach-to-social-engineering-attack-credential-theft/

CyberSecurityNews — OpenAI Arrest: https://cybersecuritynews.com/openai-engineer-arrested-for-planting-spyware

Adversa AI — GenAI Security Digest Oct 2026: https://adversa.ai/blog/top-genai-security-resources-october-2026/

CyberSecurityNews — Google AitM Kit: https://cybersecuritynews.com/google-aitm-phishing-kit

The Hacker News — Passkey Attacks: https://thehackernews.com/2026/08/new-passkey-attacks-can-recover-synced.html

Infosecurity Magazine — ZeroDayRAT: https://www.infosecurity-magazine.com/news/zerodayrat-mobile-spyware-android/

Dark Reading — ICS/OT: https://www.darkreading.com/ics-ot-security

Bitsight — ICS/OT Exposure: https://www.bitsight.com/blog/2026-global-state-of-ics-ot-exposure

OpenAI Daybreak: https://openai.com/index/daybreak-securing-the-world/

GBHackers — Pass-the-Passkey: https://gbhackers.com/pass-the-passkey-attack-exploits-windows-and-entra-id/

Check Point Research: https://research.checkpoint.com/2026/12th-october-threat-intelligence-report/