September Cyber Roundup: AI Threats + Actions for your Sector

Electric Utilities & Critical Infrastructure (OT/ICS): Signals from August

INL’s TAIGR Launch
Idaho National Lab stood up Testing for AI Grid Resilience (TAIGR) to trial where AI genuinely helps grid ops (e.g., forecasting, fault detection) and where it hurts (hallucinations, adversarial prompts, poisoned training data). The goal is disciplined pilot-to-production paths with guardrails, not “AI everywhere” by default.
Read more.

Patch pressure
CISA pushed batches of ICS advisories across common OT vendors (Johnson Controls, Schneider Electric, Siemens, Rockwell, and more). Several issues are remotely exploitable or low-complexity—meaning they’re realistic footholds for actors pairing commodity exploits with AI-driven recon.
Read more.

What you should do now
- Triage the August advisories against asset inventory; pay attention to Johnson Controls FX80/FX90 (CVSS v4 8.4) and Rockwell FLEX 5000 I/O (remote/low complexity). Read more.

- If you’re piloting AI in SCADA/DER forecasting, align data pipelines to CISA’s AI Data Security guidance (provenance, anti-poisoning, and integrity checks). Read more.

- Start network micro segmentation work now (CISA’s new Part 1 is a planning playbook that actually helps you sequence the work). Read more.

State & Local Government: What happened — and why it matters

St. Paul, MN — confirmed ransomware + data leak
The city’s July 25 cyber incident was confirmed as ransomware on Aug 10. St. Paul required in-person password resets for ~3,500 employees to safely restore access—an unusual but prudent move when identity systems may be compromised. Days later, attackers posted about 43 GB of city data after officials refused to pay, with impacts touching public-facing services (e.g., library internet access). Reporting attributes the attack to the Interlock group. This is the full lifecycle: disruption → coercion via leak site → recovery under zero-trust assumptions.
Read more.

Mower County, MN — ransomware with HIPAA scope
Following a June 18 ransomware detection, the county issued a HIPAA web notice on Aug 15 confirming unauthorized access to protected health information tied to Health & Human Services. The county says individual notifications and creditmonitoring are planned. It’s a clear example of how a county IT event can become a regulated healthdata incident once PHI is involved.
Read more.

State & Local Cybersecurity Grant Program (SLCGP) — What to know now
CISA’s SLCGP funds SLTT (state, local, tribal, territorial) cybersecurity. FY25 funding is $91.75M, administered by CISA (program) and FEMA (grants). States/territories apply; 80% must pass through to local governments and 25% to rural areas within that pass-through, generally within 45 days of funds release. Cost share: 40% for single-entity projects, 30% for multi-entity projects (waived only for certain insular territories this year).  Cybersecurity Plans must be confirmed or revised by Jan 30, 2026. Subrecipients (except .edu institutions)
must migrate to .gov domains. Funds can support plan implementation, admin (capped at 5%), and projects aligned to program objectives; ransom payments, construction/renovation, and similar items are not allowed. Read more.

Why it matters
If you’re SLTT leadership, this is near-term budget to advance zero-trust controls, identity hardening, IR, and .gov migration without waiting on the next appropriations cycle. If you’re a vendor/partner, align offerings to eligible uses and help customers meet pass-through and cost-share constraints.
Do now
- States/territories: confirm Cybersecurity Plan status and prep pass-through mechanics (funds or in-kind) with rural allocation tracking.
- Locals/tribes: engage your SAA and Cybersecurity Planning Committee early; queue projects that map cleanly to the plan and can meet 40%/30% match.
- Everyone: plan the .gov domain migration path and capture cost-share documentation up front.

What State & Local Government leaders should do now
- Assume identity compromise after ransomware: require staged credential resets (admin → service → user), and enforce out-of-band verification for privileged actions—St. Paul’s in-person reset posture is a relevant model.
- Plan for “data-leak pressure.” If you refuse to pay, prepare comms, legal, and victim-support workflows (e.g., credit monitoring) before the leak post goes live. Use Mower County’s HIPAA notice structure as a template if PHI is involved.

FINANCE: Risk & Regulation: AI Threats Reshaping Finance

Deepfake fraud moved from “rare” to “routine controls”
The Wall Street Journal reports a surge in AI generated voice/video impersonations of executives used to authorize large, urgent transfers—mirroring a widely covered eight-figure UK case. Reported losses already exceed $200M in early 2025, and techniques now blend deepfaked Zoom calls with spoofed domains and compromised vendor threads.  U.S. banks are being urged to harden verification steps because traditional “recognize the voice”
controls no longer hold.
Read more

Regulatory anchor is shifting: FFIEC CAT retires Aug 31, 2025
FFIEC confirmed it will remove the Cybersecurity Assessment Tool on Aug 31, 2025 and not update it for newer resources.  Supervisors direct firms to align with NIST CSF 2.0 and CISA Cybersecurity Performance Goals (CPGs) instead—expect examiner conversations and board reporting to reference these frameworks going forward.
Read more.

Workday: vendor CRM data accessed via social engineering
Workday says attackers tricked people by phone/text; info pulled from a third-party CRM (names, emails, phone numbers). No evidence of access to customer tenants. 
Why it matters: Classic vendor-path exposure that can seed targeted phishing. 
Do now: retrain on voice/SMS phishing, lock down CRM access, review third-party authorization.
Read more

What to do now
- Bake deepfake-aware payment controls into treasury ops (mandatory callback on vendor banking changes; tiered holds; executive deepfake drills). Use FinCEN typologies to drive training.
- Re-baseline programs against NIST CSF 2.0/CRI ahead of the CAT sunset and brief Audit/Risk.

Enterprises are rapidly adopting AI and cloud technologies while neglecting to adequately strengthen their core defenses
According to a recent survey of 1,000 senior IT and business leaders, only 62% are implementing or planning zero trust, 45% have deployed or intend to deploy MDR, and just 42% use or plan to use digital identity/IAM—even though identity-related breaches are the leading cause of cloud incidents. Additionally, 85% acknowledge that their strategies are “too reactive.” While 78% plan to increase spending on GenAI, only 45% are “very satisfied” with the return on investment. Most respondents also feel unprepared for post-quantum threats; 71% say their defenses are inadequate, and only 14% believe their infrastructure can support post-quantum cryptography, despite U.S. recommendations to begin early migration planning. Read more 

Why it matters: Even if tenants are not compromised, having accurate names, emails, and phone numbers from a vendor CRM provides attackers with believable reasons to target your executives, finance/AP, HR, sales, and procurement teams. This can lead to business email compromise (BEC), invoice and bank-change fraud, vendor-portal takeovers, and customer impersonation, resulting in actual financial losses, contract risks, and damage to your supply chain’s reputation. 
Do now:  
Identity: Implement single sign-on (SSO) with phishing-resistant multi-factor authentication (MFA); eliminate legacy authentication; apply least-privilege and time-limited vendor access.  
CRM data: Limit exports and API access; set alerts for unusual data retrieval; rotate keys regularly; activate data loss prevention (DLP).  
Payments: Use out-of-band callbacks to verified numbers and require two-person approvals for bank or purchase order changes and high-value wire transfers.  
Email/brand: Employ SPF, DKIM, and DMARC to reject spoofed emails; monitor for look-alike domains; enable alerts for executive impersonation attempts.  
Humans: Conduct vishing and smishing training exercises; establish scripted phone and SMS verification processes for finance, help desk, and vendor management.  
Vendors: Mandate MFA and maintain audit logs; disable inactive accounts; include 24–48 hour incident notification clauses in contracts.  
Incident response: Maintain a playbook and prewritten communications for social engineering incidents originating from vendors.

HEALTHCARE: Signals and Next Moves

AI cuts both ways in hospitals
Health systems are adopting AI to triage alerts and detect anomalies faster; at the same time,threat groups use GenAI to write phishing lures, customize malware, and produce  deepfakes, increasing identity-assurance and HIPAA-process risk. Read more.