Cyber News

What’s Going on in Cyber (29 OCT 2026)

Posted on by Jason Kramer in Cyber News

The Human Take

Every headline this week says “AI.” Read past the headlines, though, and almost every story turns on a person: someone who left a key in a repo, someone who reused a credential, someone stretched too thin to patch in time.

October 2026 has already earned the nickname “Black October,” with 79 dated attacks and breaches tracked for the month, 13 of them described as AI-run. That number is real and it matters. But if you work in security, or just worry about it, the more useful question isn’t “how smart are the machines now?” It’s “where are the humans in this story, and what can they actually do?”

One operator, sixteen banks

The story that will define this month is ARTEX. CrowdStrike tied a China-based actor to a campaign against South Korean financial firms that ran from late September into early October. Reporting puts it at 16 firms and roughly 68,000 affected customers, and South Korea’s Financial Services Commission has opened a probe. Japanese firms were hit next.

The uncomfortable detail: ARTEX is an open-source penetration-testing framework. It was built so defenders could red-team their own networks. Attackers picked it up, wired it to commercial language models (Claude Code among them, per the reporting) for planning and phishing drafts, and used it to do work that normally takes a small team.

That’s the shift worth sitting with. Not that AI “hacked a bank” on its own, but that one person can now run a multi-organization campaign. Headcount used to be a natural limit on attackers. It isn’t anymore.

The human response isn’t panic, and it isn’t banning the tools. Red-team frameworks have always been dual-use; so were port scanners and Metasploit. What changes is tempo. If one operator can probe sixteen institutions at once, defenders can’t rely on noticing a slow, manual intruder. Detection has to be automated too, and the people on the team have to spend their time on judgment calls the automation can’t make.

It’s also a reminder for AI vendors: misuse detection and account enforcement are part of the security perimeter now, whether anyone planned for that or not.

The real zero-day is a leaked key

Line up the rest of the month’s incidents and a pattern jumps out. Almost none of them start with a clever exploit. They start with a credential in the wrong place.

  • Storm-0940 and ConnectWise. Microsoft and CrowdStrike link this group to a compromise of ConnectWise Cloud Backup that reached more than 2,250 tenants. One compromised app credential could sign in across many Microsoft 365 tenants.
  • Railway. The platform disclosed unauthorized access that exposed customer secrets and AI provider keys, which the attacker then used for model distillation.
  • LiteLLM 1.107.0. The malicious release traces back months, to an internal master key accidentally committed to a public repository.
  • Indian fintech. CERT-In warned that attackers are stealing API credentials from lenders and payment firms and pointing AI agents at confidential records.

Every one of those is a human moment. Someone committed a file without checking it. Someone granted an integration broad access because scoping it was tedious. Someone never rotated a key because nothing seemed wrong.

I don’t say that to blame individuals. These are system failures that show up as individual mistakes. If a single forgotten key can unlock a supply chain, the problem is the design, not the developer. The fix is boring and effective: secret scanning on every commit, short-lived credentials, least-privilege scopes for apps and agents, and an inventory of every non-human identity you run. AI agents are just the newest kind of employee with a badge; they need the same offboarding and access reviews as everyone else.

1,415 patches and one tired sysadmin

Microsoft’s October Patch Tuesday reportedly shipped 1,415 CVEs, a record, including one zero-day under active attack: CVE-2026-96951, a Windows Shell bypass that lets attackers dodge Mark of the Web protections. Microsoft grouped the release into new “risk cohorts” to help teams sort it.

That grouping is an admission. Nobody can read 1,415 advisories. The number alone tells you that patching is no longer a checklist task; it’s a triage problem, and triage is a human skill.

The same month brought PHP2Shell (CVE-2026-109618/19), memory-corruption flaws in the PHP runtime that can lead to remote code execution, and a CVSS 10.0 unauthenticated RCE in self-managed GitLab (CVE-2026-110264). If you run either, those jump the queue.

The practical take: patch the actively exploited and the internet-facing first, and give your team explicit permission to defer the rest. Burnout is a security risk too. A team that tries to treat every CVE as urgent ends up treating none of them that way.

Frameworks are only as good as the people running them

While attackers were busy, the governance world kept building. The EU AI Act is now a binding regulation with fines up to €35 million. NIST’s AI Risk Management Framework stays voluntary, extended by a Generative AI Profile (NIST AI 600-1) with 13 GenAI-specific risks and more than 400 suggested actions. ISO/IEC 42001 offers a certifiable management system on top. NIST also updated Cybersecurity Framework mappings to tie checklist settings to CSF 2.0 outcomes. Meanwhile, privacy rules keep shifting across 50-plus jurisdictions, and US states are layering their own AI obligations on top.

Many organizations are settling into a three-layer stack: NIST as the operating framework, ISO 42001 for certification, and EU AI Act conformity where it applies. That’s sensible. It’s also a lot of paperwork.

Here’s the human catch. None of this week’s breaches would have been stopped by a well-formatted policy binder. They’d have been stopped by someone who knew where their keys lived, who could see what their agents were accessing, and who had time to act on it. The NIST framework’s own “Govern” function says as much: it’s about culture, roles, accountability and workforce readiness, not documents.

Use the frameworks to decide what matters. Then make sure a real person owns each piece of it.

What a person can actually do this week

Machine-speed attacks are real, and AI-driven detection is part of the answer. But the leverage is still mostly human. Five things worth doing before next Patch Tuesday:

  1. Find your secrets. Turn on secret scanning for every repo, including old ones, and rotate anything it finds.
  2. Inventory your non-human identities. List every service account, API key and AI agent, what it can reach, and who owns it.
  3. Shrink the blast radius. Scope app credentials to one tenant and the least privilege they need; prefer short-lived tokens.
  4. Triage, don’t drown. Patch CVE-2026-96951 and any internet-facing PHP or self-managed GitLab now; schedule the rest.
  5. Name an owner. For each framework control you claim, write down the human responsible for it.

The tools changed this month. The people who keep things running didn’t. They’re still the best defense we have, as long as we give them the time and visibility to do the job.

Cybersecurity threats represented in Black October

Weekly Cyber News Summary 2026-10-29

This week in cyber:
  • ARTEX AI agent hacks South Korean banks, exfiltrating customer data
  • Microsoft Patch Tuesday releases 1,415 CVEs, including one zero-day
  • AI-driven attacks surge as machine-speed threats redefine defense
  • EU AI Act and NIST AI RMF comparison guides enterprise compliance
  • PHP2Shell vulnerabilities expose web apps to remote code execution
  • Black October 2026 sees 79 dated cyberattacks and breaches
  • Railway breach exposes customer secrets and AI provider keys
  • Global data privacy laws evolve with 50+ jurisdictions updating rules

State-Sponsored Attacks

ARTEX AI Agent Hacks South Korean Banks CrowdStrike has identified a China-based threat actor using the ARTEX AI agent to hack South Korean financial organizations. The campaign ran from late September to early October 2026, exfiltrating customer data. The actor used Claude Code and other large language models to plan moves and draft social engineering material. The actor used ARTEX to automate reconnaissance and exploitation steps that would normally require a small team, then leaned on commercial chatbots to plan next moves.
Sources:
Shattered.io: https://shattered.io/crowdstrike-artex-ai-agent-korea-bank-hack-2026/
Cynet: https://www.cynet.com/security-foundations/ai-machine-learning/ai-cyberattacks/
Storm-0940 ConnectWise Compromise Microsoft and CrowdStrike link the Storm-0940 threat group to a massive ConnectWise Cloud Backup compromise. The group used stolen credentials to access over 2,250 tenants, demonstrating the power of AI-driven attacks against MSP tooling. The campaign highlights how a single compromised app credential can sign in to multiple Microsoft 365 tenants, allowing the group to operate across a wide range of environments.
Sources:
Explainx.ai: https://explainx.ai/blog/october-2026-cyberattacks-list-breaches-exploits-ai-run-attacks-tracker-2026
Unosecur: https://www.unosecur.com/blog/october-2026-cybersecurity-incidents
150% Surge in Nation-State Attacks A 150% surge in nation-state cyberattacks is reported for 2026, with China, Russia, Iran, and North Korea sponsoring 77% of operations. The focus is on long-term strategic persistence and critical infrastructure. The report highlights that state-sponsored actors are increasingly prioritizing strategic persistence over short-term disruption, targeting energy, financial, and defense networks.
Sources:
Bugitrix: https://www.bugitrix.com/blog/news-trends-10/nation-state-cyberattacks-150-percent-surge-2026-44
CyberProof: https://www.cyberproof.com/cyberproof-2026-global-threat-intelligence-report/

Regional Campaigns

South Korea Financial Sector Targets South Korea’s financial sector faces a new era of AI-linked bank hacks. The ARTEX campaign affected 16 firms, with 68,000 customers impacted, prompting a probe by the Financial Services Commission. The campaign is described as quiet data exfiltration rather than destructive disruption, consistent with the broader 2026 industry shift toward data-theft-first criminal operations.
Sources:
Shattered.io: https://shattered.io/crowdstrike-artex-ai-agent-korea-bank-hack-2026/
CyberProof: https://www.cyberproof.com/cyberproof-2026-global-threat-intelligence-report/
Japan Firms Hit by ARTEX Japanese firms are also targeted by the ARTEX tool, following the South Korea campaign. The open-source agentic framework is proving effective for financially motivated operators. The tool’s ability to route around whichever vendor’s filters are weakest makes it resilient and attractive to attackers running multi-organization campaigns.
Sources:
Shattered.io: https://shattered.io/crowdstrike-artex-ai-agent-korea-bank-hack-2026/
CyberProof: https://www.cyberproof.com/cyberproof-2026-global-threat-intelligence-report/
Indian Fintech Attacks Indian fintech companies face attacks where attackers steal API credentials at lenders and payment firms. CERT-In warns about AI agents using leaked keys to read confidential records. The advisory highlights the risks of non-human identities and the need for visibility into agent access to sensitive data.
Sources:
Explainx.ai: https://explainx.ai/blog/october-2026-cyberattacks-list-breaches-exploits-ai-run-attacks-tracker-2026
Unosecur: https://www.unosecur.com/blog/october-2026-cybersecurity-incidents

AI Integration

AI-Driven Attack Surge AI cyberattacks have evolved into fully automated attack chains, combining phishing, malware, and lateral movement. 2026 sees machine-speed attacks where systems react to each other in real time. The rise of AI-driven malware allows it to modify itself mid-execution, shift its behavior to avoid detection signatures, and respond dynamically to defensive actions.
Sources:
Cynet: https://www.cynet.com/security-foundations/ai-machine-learning/ai-cyberattacks/
Ridge IT: https://www.ridgeit.com/ai-cybersecurity-defense-2026/
ARTEX Agentic Tool ARTEX, an open-source agentic penetration-testing tool, allows a single operator to run a multi-organization campaign. It orchestrates tasks and calls out to LLMs for reasoning-heavy steps. The tool was built for defensive red-teaming but has been repurposed by criminals to automate reconnaissance and exploitation, lowering the barrier for entry.
Sources:
Shattered.io: https://shattered.io/crowdstrike-artex-ai-agent-korea-bank-hack-2026/
Unosecur: https://www.unosecur.com/blog/october-2026-cybersecurity-incidents
AI Threat Detection Guide AI threat detection is evolving to counter AI-powered attacks. SMBs are adopting AI-powered security to achieve 98%+ threat prevention rates. Modern EDR platforms use AI to monitor all endpoint activity continuously, identifying malicious behavior patterns across the entire attack lifecycle.
Sources:
Ridge IT: https://www.ridgeit.com/ai-cybersecurity-defense-2026/
Cynet: https://www.cynet.com/security-foundations/ai-machine-learning/ai-cyberattacks/

AI Policy and Posture

EU AI Act vs NIST AI RMF The EU AI Act and NIST AI RMF are the two most influential AI governance instruments. The EU Act is a binding regulation, while NIST is a voluntary framework. Organizations are using a three-layer governance stack: NIST AI RMF as the operating framework, ISO/IEC 42001 as the certifiable management system, and EU AI Act conformity where jurisdictional scope applies.
Sources:
Risk Publishing: https://riskpublishing.com/eu-ai-act-vs-nist-ai-rmf-comparing-ai-risk/
Unosecur: https://www.unosecur.com/blog/october-2026-cybersecurity-incidents
AI Governance Roles AI governance requires active controls like tracking bias drift and aligning AI with compliance. Frameworks like NIST AI RMF provide a structure for risk management. The Govern function in NIST AI RMF is a cross-cutting organizational layer covering culture, policy, roles, accountability, third-party risk, and workforce readiness.
Sources:
Swfte: https://www.swfte.com/ai-governance
Risk Publishing: https://riskpublishing.com/eu-ai-act-vs-nist-ai-rmf-comparing-ai-risk/
AI Compliance Best Practices AI compliance in 2026 involves navigating the EU AI Act, NIST AI RMF, and global regulations. Organizations are building audit-ready programs for AI systems. Key practices include implementing AI asset discovery, enforcing least privilege, and monitoring suspicious identity activity at runtime.
Sources:
Anaconda: https://www.anaconda.com/guides/ai-compliance
Risk Publishing: https://riskpublishing.com/eu-ai-act-vs-nist-ai-rmf-comparing-ai-risk/

Zero-Day Exploits/CVEs

Microsoft Patch Tuesday Microsoft released 1,415 CVEs in October 2026 Patch Tuesday, including one exploited zero-day. The release featured record volume and new “risk cohorts” for organization. One bug under active attack is CVE-2026-96951, a Windows Shell Security Feature Bypass Vulnerability that allows attackers to avoid Mark of the Web protections.
Sources:
Zero Day Initiative: https://www.zerodayinitiative.com/blog/2026/10/13/the-october-2026-security-update-review
Unosecur: https://www.unosecur.com/blog/october-2026-cybersecurity-incidents
PHP2Shell Vulnerabilities PHP2Shell vulnerabilities (CVE-2026-109618/19) expose web applications to remote code execution. The flaws affect the underlying PHP runtime used by many apps. The vulnerabilities involve memory-management failures during request processing and cleanup, allowing attackers to corrupt memory, access process data, or execute arbitrary code.
Sources:
Unosecur: https://www.unosecur.com/blog/october-2026-cybersecurity-incidents
Explainx.ai: https://explainx.ai/blog/october-2026-cyberattacks-list-breaches-exploits-ai-run-attacks-tracker-2026
GitLab Unauthenticated RCE GitLab addressed a critical unauthenticated RCE (CVE-2026-110264) in CI/CD regex compilation. The vulnerability could allow code execution under affected conditions. The flaw is rated CVSS 10.0 and affects self-managed GitLab installations, allowing unauthenticated attackers to execute code via crafted requests.
Sources:
Unosecur: https://www.unosecur.com/blog/october-2026-cybersecurity-incidents
Explainx.ai: https://explainx.ai/blog/october-2026-cyberattacks-list-breaches-exploits-ai-run-attacks-tracker-2026

Incident Reports

Black October 2026 “Black October” 2026 saw 79 dated cyberattacks, including 14 breaches and 13 AI-run attacks. The month highlighted the risks of stolen credentials and trusted channels. The incidents included the disclosure of the PHP2Shell vulnerabilities, a critical GitLab bug, and a compromised LiteLLM release, demonstrating the speed of modern attacks.
Sources:
Explainx.ai: https://explainx.ai/blog/october-2026-cyberattacks-list-breaches-exploits-ai-run-attacks-tracker-2026
Unosecur: https://www.unosecur.com/blog/october-2026-cybersecurity-incidents
Railway Breach Railway disclosed unauthorized access to internal systems, exposing customer secrets and AI provider keys. The attacker used the keys for model distillation. The breach began with abnormal infrastructure activity and developed into a sustained campaign targeting the platform’s build infrastructure.
Sources:
Unosecur: https://www.unosecur.com/blog/october-2026-cybersecurity-incidents
Explainx.ai: https://explainx.ai/blog/october-2026-cyberattacks-list-breaches-exploits-ai-run-attacks-tracker-2026
LiteLLM Supply Chain LiteLLM 1.107.0 was compromised, with malicious packages stolen via a GitHub token. The malware targeted credentials and supported additional code execution. The attack began months before the malicious package appeared, when an internal master key was accidentally committed to a public repository.
Sources:
Unosecur: https://www.unosecur.com/blog/october-2026-cybersecurity-incidents
Explainx.ai: https://explainx.ai/blog/october-2026-cyberattacks-list-breaches-exploits-ai-run-attacks-tracker-2026

Framework Changes

NIST AI RMF and GenAI Profile NIST AI RMF 1.0 and its GenAI Profile structure AI risk management. The framework is widely adopted by Fortune 500 companies. The NIST Generative AI Profile (NIST AI 600-1) extends the framework with 13 GenAI-specific risks and more than 400 suggested actions.
Sources:
Risk Publishing: https://riskpublishing.com/eu-ai-act-vs-nist-ai-rmf-comparing-ai-risk/
NIST: https://www.nist.gov/cyberframework
ISO/IEC 42001 Certification ISO/IEC 42001 is a certifiable AI management system that complements EU AI Act and NIST AI RMF. It provides third-party-auditable evidence. The standard helps organizations build a single AI inventory with EU AI Act risk-tier tags and one control library mapped to NIST AI RMF subcategories.
Sources:
Risk Publishing: https://riskpublishing.com/eu-ai-act-vs-nist-ai-rmf-comparing-ai-risk/
Unosecur: https://www.unosecur.com/blog/october-2026-cybersecurity-incidents
NIST Cybersecurity Framework Updates NIST Cybersecurity Framework updates include enhanced mapping concepts for evidence-ready automation. The updates align with SP 800-53 controls. The update includes an appendix with enhanced mapping concepts between checklist settings, NIST Cybersecurity Framework (CSF) 2.0 outcomes, and Common Configuration Enumeration (CCE) identifiers.
Sources:
NIST: https://www.nist.gov/cyberframework
Unosecur: https://www.unosecur.com/blog/october-2026-cybersecurity-incidents

New Legislation

EU AI Act Enforcement The EU AI Act becomes fully applicable on August 2, 2026, with GPAI model rules applying from August 2025. The Act imposes fines up to €35 million. The second half of 2026 will be the first real stress test of EU AI Act and NIST AI RMF as a combined governance system.
Sources:
Risk Publishing: https://riskpublishing.com/eu-ai-act-vs-nist-ai-rmf-comparing-ai-risk/
Unosecur: https://www.unosecur.com/blog/october-2026-cybersecurity-incidents
Global Privacy Laws 2026 Global data privacy laws in 2026 cover over 50 jurisdictions. The guide includes GDPR, CCPA/CPRA, and cross-border transfer mechanisms. The comprehensive guide to data privacy and protection laws across more than 50 global jurisdictions in 2026 covers enforcement trends, compliance checklists, and penalty benchmarks.
Sources:
Global Law Lists: https://globallawlists.org/insights/international-lawyers-guide-data-privacy-laws-2026-navigating-50-plus-jurisdictions
Cdslegal: https://cdslegal.com/insights/global-data-privacy-laws-in-2026-mid-year-update/
US State AI Laws US state laws are layering obligations onto NIST-based programs. Colorado’s AI Act and California’s SB 1047 are key examples. The trajectory is profile proliferation rather than statutory change, with state-level AI laws creating a patchwork of obligations for enterprises.
Sources:
Cdslegal: https://cdslegal.com/insights/global-data-privacy-laws-in-2026-mid-year-update/
Risk Publishing: https://riskpublishing.com/eu-ai-act-vs-nist-ai-rmf-comparing-ai-risk/