TLP:GREEN  ·  States / Public Sector
AI-Powered Exploitation, Supply Chain Worms, and Three New KEVs:

What State Government CISOs Must Act On This Week

ELEVATED. Elevated from the prior cycle's sustained posture due to the convergence of three actively exploited CISA KEV vulnerabilities directly relevant to state government infrastructure, combined with the emergence of AI-enabled autonomous exploitation that fundamentally compresses the attacker's timeline. A Chinese-speaking threat actor is using an AI agent to autonomously research and exploit vulnerabilities at machine speed — targeting the same web application platforms running in state agency environments. Simultaneously, a self-propagating supply chain worm has compromised 440 NPM packages with 500 million combined weekly downloads, and CISA has issued three new KEV entries with a remediation deadline of August 7, 2026.

I am a
My sector

DateDevelopmentImpact
Aug 1, 2026CVE-2026-18577 exploitation confirmed against MSP environments; 9+ organizations compromisedDirect supply chain risk to state agencies using affected MSPs
Aug 3, 2026CISA adds CVE-2026-18577 to KEV catalog; federal patch deadline set for Aug 6Establishes mandatory remediation timeline
Aug 4, 2026ChainDrop NPM worm escalates from 56 to 440 compromised packages (2,212 malicious versions) in under 4 hoursState CI/CD pipelines consuming NPM packages at risk
Aug 4–5, 2026CISA adds CVE-2026-9198 (Langflow RCE, CVSS 9.8), CVE-2026-34486 (Tomcat), CVE-2026-18556 (N-central) to KEVThree actively exploited vulns with Aug 7 FCEB deadline
Aug 5, 2026Unit 42 discloses Chinese actor using DeepSeek AI as autonomous exploitation agent — 460+ targetsParadigm shift: AI compresses exploitation timeline from days to hours
Aug 5, 2026GitGuardian reveals 321 n8n automation instances accepting leaked API tokens from public GitHubWorkflow automation platforms exposing connected databases and credentials
Aug 5, 2026Researchers demonstrate "Pass-ta-key" attacks stealing Google synchronized passkeys via malwareUndermines passkey security assumptions for FIDO2 rollouts
Jul 25, 2026 (ongoing)DHS discloses Salt Typhoon & Volt Typhoon persistent access in state/National Guard networks — no remediation reportedChinese MSS espionage threat remains unresolved in affected environments
OngoingCactus, INC Ransomware, and Silent Ransom Group C2 infrastructure confirmed active; government remains in target profileRansomware threat posture unchanged; absence of new incidents is not evidence of reduced risk
OngoingBANISHED KITTEN (IRGC) targeting water/wastewater PLCs across 7+ U.S. statesCritical infrastructure under active Iranian threat

TimeframeEventThreat ActorRelevance to State Gov
Jul 25, 2026DHS discloses Salt Typhoon & Volt Typhoon persistent access in state/National Guard networksChinese MSS-affiliatedNo remediation reported; ongoing espionage risk
Aug 1, 2026N-central auth bypass exploitation begins (CVE-2026-18577)Unknown (multiple actors likely)MSP-managed state endpoints directly exposed
Aug 4, 2026ChainDrop worm reaches 440 packages / 500M weekly downloadsUnknown (supply chain)State developer CI/CD pipelines at risk
Aug 4–5, 2026Triple KEV addition (Langflow, Tomcat, N-central)Multiple — including knaithe/KnYuanInternet-facing state applications targeted
Aug 5, 2026AI-autonomous exploitation campaign disclosedknaithe/KnYuan (Zhuhai, China)Tomcat, VPN, workflow platforms in target set
OngoingCactus ransomware C2 infrastructure active (107.189.28[.]30, confidence 98)Cactus Ransomware GroupGovernment remains in target profile
OngoingBANISHED KITTEN targeting water/wastewater PLCs across 7+ U.S. statesIRGC-affiliatedCritical infrastructure under active threat

A Chinese-speaking threat actor operating under the aliases knaithe and KnYuan (based in Zhuhai, China) has been confirmed using DeepSeek AI via the Hermes Agent framework as an autonomous offensive operator. This is not theoretical — Unit 42 documented the actor targeting 460+ internet-exposed devices with an AI agent that independently researches new vulnerabilities when initial exploits fail, manages its own compute resources without human intervention, compresses hundreds of hours of manual targeting analysis into minutes, and autonomously pivots between attack vectors.

The actor's confirmed exploitation targets include CVE-2026-9198 (Langflow RCE), CVE-2026-34486 (Apache Tomcat), CVE-2026-3055 (Citrix NetScaler), and CVE-2026-33824 (IKE VPN) — all platforms commonly found in state government environments.

Why this matters for state agencies: Traditional 30-day patch windows assume human-speed adversaries. When an AI agent can discover, research, and exploit a new vulnerability within hours of disclosure, those windows are dangerously inadequate for internet-facing services.

T1190

CVE-2026-18556 and CVE-2026-18577 are authentication bypass and account takeover vulnerabilities in N-able N-central through version 2026.3.1. CVE-2026-18577 is an incomplete fix for CVE-2026-18556 — meaning organizations that patched the first vulnerability may still be exposed.

Both are now confirmed actively exploited, with 9+ organizations already compromised. Over 55% of N-central cloud servers remained unpatched as of last reporting.

The supply chain math is simple: If your MSP uses N-central to manage your endpoints, an attacker who compromises the MSP has authenticated access to push commands, deploy software, and move laterally across every client environment — including yours. This mirrors the 2023 MOVEit pattern where a single vendor compromise cascaded across hundreds of downstream victims.

Hotfix available: N-central 2026.3 HF1 addresses both CVEs.

T1078

The ChainDrop campaign has escalated dramatically. What began as 56 compromised packages has exploded to 440 packages with 2,212 malicious versions published in under 4 hours on August 4. The combined weekly download count of infected packages exceeds 500 million.

Key technical details: Initial vector — compromised maintainer GitHub account in the keyv and cacheable namespaces. Targets — NPM tokens, GitHub credentials, AWS keys, Kubernetes secrets, HashiCorp Vault credentials. C2 mechanism — Ethereum blockchain (EtherHiding technique), extremely difficult to take down. Propagation — self-replicates via stolen NPM tokens to republish poisoned package versions. Novel persistence — injects Claude AI and VS Code configuration files for developer-to-developer infection. Anti-forensics — dead-man's switch polls GitHub API every 60 seconds; self-destructs if token is revoked.

State government exposure: Any agency with custom citizen-facing web applications built on Node.js, or any CI/CD pipeline consuming NPM packages from the affected namespaces, is potentially compromised.

T1195.002T1059.007T1102.002

GitGuardian research found 4,576 unique n8n API tokens in 5,469 public GitHub commits. Of the reachable instances, 321 (36%) accepted at least one leaked token, granting access to workflow definitions and execution history, stored credentials for connected systems, database connections, and cloud environment access.

Additionally, 58% of scanned n8n instances are running versions with known security advisories, including CVE-2025-68613 (CVSS 9.9, expression injection, in CISA KEV).

State government exposure: Agencies using n8n or similar low-code workflow automation for inter-agency data pipelines may have tokens committed to repositories by developers or AI coding assistants. The .claude/settings.json credential leakage vector is particularly concerning — AI coding tools are creating exposure paths that didn't exist 12 months ago.

T1552.001

The following nation-state threats from prior cycles remain active with no reported remediation:

ActorAffiliationActivityStatus
Salt TyphoonChinese MSSPersistent access in state agency networksNo remediation reported (disclosed Jul 25)
Volt TyphoonChinese MSSPersistent access in National Guard networksNo remediation reported (disclosed Jul 25)
BANISHED KITTENIRGC (Iran)Targeting water/wastewater PLCs in 7+ statesActive
APT29 / Midnight BlizzardRussian SVRCredential harvesting via hotel Wi-Fi proximity attacksOperationally quiet this cycle
APT28Russian GRUInfrastructure IOC refresh (77.90.185[.]28)Active

No new ransomware incidents targeting state or local government were reported this cycle. However, three tracked ransomware operations remain active: Cactus Ransomware (C2 infrastructure at 107.189.28[.]30 confirmed active, confidence 98, last validated Aug 5); INC Ransomware / UTA0533 (exploiting SonicWall appliances — common in state government remote access architectures); Silent Ransom Group / Luna Moth (conducting encryption-free extortion targeting organizations with limited security maturity).

The absence of new victims is not evidence of reduced risk. These groups operate on their own timelines, and state agencies remain squarely in their target profiles.

ScenarioProbabilityBasis
Additional exploitation of CVE-2026-9198 (Langflow) at scaleHIGH (75–85%)AI-enabled scanning dramatically accelerates target discovery; broad internet exposure of Langflow instances
N-central exploitation leveraged for MSP-to-client lateral movementMODERATE (50–65%)Mirrors MOVEit 2023 pattern; 55%+ instances unpatched; 9+ orgs already compromised
ChainDrop worm variants targeting new NPM namespacesMODERATE (45–60%)Original packages being remediated; actor has demonstrated ability to pivot quickly
Cactus ransomware deployment against government targetLOW-MODERATE (25–40%)C2 infrastructure confirmed active; government in target profile; timing unpredictable
AI-autonomous exploitation tool adopted by additional threat actorsMODERATE (40–55%)DeepSeek is publicly available; Hermes framework is open-source; barrier to replication is low

ATT&CK TechniqueDetection FocusHunting Hypothesis
T1190 (Exploit Public-Facing Application)Monitor for exploitation attempts against Langflow, Tomcat, N-central, and VPN endpoints"If an AI agent is scanning our perimeter, we'll see rapid sequential exploitation attempts against multiple services from a single source IP within a compressed timeframe"
T1078 (Valid Accounts)Alert on N-central administrative actions from unexpected source IPs or at unusual hours"If our MSP's N-central instance is compromised, we'll see legitimate-looking management commands originating from infrastructure outside the MSP's known IP ranges"
T1195.002 (Supply Chain Compromise)Monitor CI/CD pipeline logs for unexpected package version changes in keyv and cacheable namespaces"If ChainDrop reached our pipelines, we'll see NPM install events pulling versions published after Aug 1 from compromised namespaces, followed by outbound connections to Ethereum nodes"
T1552.001 (Credentials in Files)Scan repositories for .claude/, .cursor/, .env, and n8n token patterns"If developers are using AI coding assistants, leaked credentials may exist in AI configuration files that bypass standard .gitignore rules"
T1059.007 (JavaScript Execution)Monitor for unexpected preinstall script execution in NPM packages"If a compromised NPM package executes during build, we'll see child process spawning from npm/node during CI pipeline runs with network callbacks to blockchain infrastructure"
T1102.002 (Web Service C2)Monitor for outbound connections to Ethereum RPC endpoints from build servers"ChainDrop uses EtherHiding — if our build infrastructure is compromised, we'll see JSON-RPC calls to Ethereum nodes for C2 resolution"
IOC Blocking Table:
77.90.185[.]28107.189.28[.]30139.162.118[.]88

Block the above at perimeter firewalls, proxies, and DNS. Additional IOCs available via Anomali ThreatStream Next-Gen and partner feeds.

Specific Detection Rules to Deploy:
RULE 01
AI-Speed Scanning Signature
Alert when a single source IP attempts exploitation of ≥5 distinct CVEs within 60 minutes. This pattern distinguishes AI-driven autonomous scanning from traditional vulnerability scanners.
RULE 02
N-central Anomaly Detection
Baseline your MSP's N-central management traffic (source IPs, time windows, command types). Alert on deviations — particularly new administrative account creation or mass policy deployment outside maintenance windows.
RULE 03
NPM Supply Chain Tripwire
In CI/CD pipelines, alert on any package-lock.json changes to packages in the keyv or cacheable namespaces. Verify integrity hashes against known-good versions published before August 1, 2026.
RULE 04
Credential Exposure Scanning
Add the following paths to secret scanning tools (GitGuardian, GitHub Advanced Security, or equivalent): .claude/settings.json, .claude/settings.local.json, .cursor/ directory contents, and any file matching pattern *n8n*token* or *N8N_API_KEY*.

Financial Services
State Treasury, Revenue, Tax Systems
Primary threat
Credential theft via ChainDrop (NPM supply chain) targeting AWS and cloud credentials used in payment processing and tax filing systems
Secondary threat
Cactus and Silent Ransom Group specifically target organizations with financial data for double-extortion leverage
Actions
  • Audit all Node.js applications processing financial data for dependencies in keyv/cacheable namespaces
  • Rotate AWS IAM keys for any service that ran npm install after August 1
  • If implementing FIDO2 for taxpayer authentication portals, ensure server-side user verification validation — do not rely on client-side UV flags alone
Energy
State-Regulated Utilities, Grid Operations
Primary threat
BANISHED KITTEN (IRGC) targeting water/wastewater PLCs across 7+ U.S. states; ICS advisories for Schneider IGSS SCADA systems issued this cycle
Secondary threat
Energy sector Tomcat-based SCADA web interfaces are in the target profile for AI-enabled autonomous scanning
Actions
  • Verify network segmentation between IT and OT environments
  • Confirm that N-central or other RMM tools do NOT have network paths to SCADA/ICS systems
  • Review Schneider IGSS deployments against CISA ICS-CERT advisories (ICSA-26-211 series)
Healthcare
State Health Agencies, Medicaid Systems
Primary threats
Ransomware (Cactus, INC) targeting healthcare data for extortion; supply chain compromise via MSP tooling managing clinic endpoints
Secondary threat
n8n or similar workflow automation connecting EHR systems, eligibility databases, and cloud services — audit for leaked tokens in any public repositories
Actions
  • Confirm MSP patch status for N-central immediately — healthcare MSPs are high-value targets
  • Ensure offline backups of Medicaid enrollment and claims data are current and tested
Government
Executive Branch Agencies, Courts, Legislature
Primary threats
Nation-state espionage (Salt Typhoon, Volt Typhoon persistent access confirmed); AI-enabled exploitation of internet-facing government applications
Secondary threat
Pass-ta-key research undermines passkey security — any FIDO2 pilot for employee authentication should include endpoint health attestation
Actions
  • Conduct emergency inventory of all internet-facing Apache Tomcat, Langflow, and VPN endpoints
  • Prioritize patching CVE-2026-9198 and CVE-2026-34486 within 24 hours for any public-facing instance
  • Engage DHS/CISA for Salt Typhoon/Volt Typhoon remediation guidance if not already underway
Aviation / Logistics
State DOT, Airports, Port Authorities
Primary threat
Volt Typhoon pre-positioning in transportation infrastructure; SonicWall exploitation (INC Ransomware/UTA0533) targeting remote access at distributed field offices
Secondary threat
Transportation agencies with custom logistics applications on Node.js should verify NPM dependency integrity
Actions
  • Audit all SonicWall SMA appliances for current firmware and known exploitation indicators
  • Verify that TP-Link Omada equipment at field offices has been updated for the 15 zero-touch provisioning vulnerabilities disclosed this cycle
  • Segment airport/port OT networks from enterprise IT

Contact your MSP and confirm N-central patch status. Both CVE-2026-18556 and CVE-2026-18577 are actively exploited. Hotfix 2026.3 HF1 must be applied. If your MSP cannot confirm patching, escalate to CISO for risk acceptance decision. FCEB deadline is August 7.
Incident ResponderCISO / Exec
Patch all Apache Tomcat instances below versions 11.0.21 / 10.1.54 / 9.0.117. CVE-2026-34486 is confirmed exploited by a Chinese actor using AI-enabled automation. Internet-facing instances are highest priority.
Incident Responder
Inventory all Langflow deployments including shadow IT and data science environments. Upgrade to 1.10.1+ or immediately disable public API access. CVE-2026-9198 is CVSS 9.8 and actively exploited.
Incident Responder
Add the three confirmed-active IPs to perimeter blocking: 77.90.185[.]28 (APT28), 107.189.28[.]30 (Cactus C2), 139.162.118[.]88 (APT infrastructure). For hash-based detection, query Anomali ThreatStream for indicators tagged to this cycle's campaigns.
SOC Analyst
No immediate actions for the selected roles.
Audit all state GitHub/GitLab repositories for leaked n8n API tokens, .claude/ directories, .env files, and automation platform credentials. Revoke and rotate any discovered tokens immediately.
Threat Hunter
Deploy AI-speed scanning detection: Alert on >5 unique CVE exploitation attempts from a single source IP within 60 minutes. This is the behavioral signature of autonomous AI-driven reconnaissance.
SOC Analyst
Pin all NPM dependencies to exact versions and enable npm audit in CI pipelines. Verify no packages from keyv/cacheable namespaces are at versions published after August 1. Rebuild CI runners if any compromised versions were installed.
Threat Hunter
Add AI assistant config paths to secret scanning: .claude/settings.json, .claude/settings.local.json, .cursor/ directories. These are new credential leakage vectors created by AI coding tools.
Threat Hunter
Baseline N-central management traffic (source IPs, time-of-day, command types) and configure alerts for anomalous administrative actions — especially new account creation or mass policy pushes outside maintenance windows.
Incident Responder
Brief the CIO on AI-enabled exploitation paradigm shift and patch SLA implications.
CISO / Exec
Confirm incident response retainer covers AI-speed exploitation scenarios.
CISO / ExecIncident Responder
Validate offline backup integrity for critical citizen-facing systems.
Incident Responder
No 7-day actions for the selected roles.
Reassess patch SLA policy. Current 30-day windows are inadequate against AI-speed exploitation. Recommend tiered approach: 24h for KEV/CVSS≥9.0 internet-facing, 72h for KEV/CVSS≥7.0, 14 days for remainder.
CISO / Exec
Assess passkey/FIDO2 deployment plans in light of Pass-ta-key research. Any passkey rollout must include server-side user verification validation and endpoint health attestation — do not rely solely on client-side UV flags.
CISO / Exec
Inventory all TP-Link Omada deployments at branch and field offices. Apply firmware updates for the 15 zero-touch provisioning vulnerabilities. Replace default provisioning passwords with unique credentials per site.
Incident Responder
Establish a unified supply chain risk register. Three distinct supply chain vectors are active simultaneously (MSP tooling, developer ecosystem, workflow automation) — each owned by a different team. Assign single executive owner for cross-cutting supply chain risk.
CISO / Exec
Engage CISA for Salt Typhoon/Volt Typhoon remediation if not already underway. DHS disclosed persistent Chinese MSS access in state and National Guard networks on July 25 with no remediation reported.
CISO / Exec
Tabletop exercise: MSP compromise scenario (N-central as attack vector).
CISO / ExecIncident Responder
Review cyber insurance policy for supply chain compromise coverage.
CISO / Exec
No 30-day actions for the selected roles.
The Bottom Line

The threat landscape facing state government IT has shifted this week in a fundamental way. We are no longer defending against human-speed adversaries alone. A confirmed threat actor is using AI to autonomously discover, research, and exploit vulnerabilities faster than most organizations can read a CISA advisory — let alone patch. At the same time, the supply chain attack surface continues to expand across three simultaneous vectors: your MSP's remote management tools, your developers' package dependencies, and your automation platforms' leaked credentials. Each of these represents a different team's blind spot, and adversaries are exploiting all three concurrently. The convergence of AI-speed exploitation, expanding supply chain attack surfaces, and persistent nation-state access demands that state government IT leadership move from reactive patching to proactive threat management. The adversary's clock is now measured in minutes. Ours must be measured in hours — not days.

1
Has your MSP patched N-central? If you cannot get a confirmed answer within 24 hours, you have a supply chain risk that requires executive-level attention.
2
Do you have internet-facing Tomcat or Langflow instances? An AI agent is actively scanning for them. Your patch window is measured in hours, not weeks.
3
Do your developers use AI coding assistants? If yes, you likely have a new class of credential exposure in your repositories that your current secret scanning tools don't detect.
No items found.