Security leadership that holds up under examination.
RedOps is a cleared, sitting-CISO practice for regulated firms. We lead your security program, test your AI the way attackers do, and produce evidence your board, NYDFS, or a contracting officer will accept.
Three reasons the phone rings.
Each starts with a short, fixed-scope diagnostic and ends with a written scope for whatever comes next.
Your AI needs a red team.
Agents and copilots are in production. Nobody can list them, say what they can touch, or prove they can't be hijacked. You need findings, not a policy template.
AI Exposure Sprint, two weeks → CEO · Board · General CounselYour board needs a CISO.
An examiner's letter, a failed customer questionnaire, a CISO who left, a Part 500 certification due April 15. You need someone who has signed for this before.
Fractional CISO for regulated firms → Defense contractor · Prime flowdownYour CMMC posture needs to be defensible.
Third-party assessment is paused. Your SPRS score and DFARS 7012 attestations are still a False Claims Act question. You need evidence that survives an assessor or a whistleblower's lawyer.
CMMC L2 Defensibility Review →The questions got personal in 2026.
of CISOs are concerned about personal liability.
of security leaders are confident they can identify every AI agent in their environment.
paid by Honeywell Aerospace to resolve False Claims Act allegations of NIST SP 800-171 noncompliance.
Govern it. Map it. Break it. Prove it.
One lifecycle, four disciplines, each ending in an artifact you can hand to a board, an examiner, or an auditor.
Govern
AI inventory, model and agent risk register, acceptable-use and third-party AI policy, board committee charter.
Map
A threat model for each LLM application and agent: data paths, tool permissions, identities, trust boundaries, MCP and RAG surfaces.
Break
Prompt injection, tool misuse, data exfiltration, privilege escalation through agents, and the social-engineering paths AI opens against your people.
Prove
An evidence pack: findings mapped to controls, a remediation roadmap, a board one-pager, and an examiner-ready narrative.
Fixed scope. Fixed fee. One accountable principal.
Every engagement starts with a diagnostic of two to three weeks or less and ends with a written scope for whatever comes next.
AI Exposure Sprint
One AI system or agent. Threat model, adversarial test, executive brief, 90-day roadmap.
2 weeksNYDFS §500 Gap & Certification Review
Controls gap, AI coverage in your risk assessment, certification-readiness memo.
2–3 weeksCMMC L2 Defensibility Review
Evidence sampling across the 110 controls, SPRS score substantiation, risk memo for counsel.
2–3 weeksAI Governance Baseline
NIST AI RMF and ISO/IEC 42001 readiness, policy stack, board memo.
6–8 weeksAdversarial AI Assessment
Multi-system and agentic: agents, RAG, MCP attack paths, social-engineering routes, retest.
4–6 weeks + retestFractional CISO
Advisory, Program or Executive cadence. Board reporting, exam and audit lead.
MonthlyDated obligations, not countdowns.
NYDFS Part 500: universal MFA (§500.12) and asset inventory (§500.13) enforceable since Nov 1, 2025. The May 21, 2026 letter on frontier-AI risk creates no new rules, but recommends faster vulnerability remediation, third-party dependency mapping, testing and human oversight of AI-generated code, and operational-resilience testing.
CMMC: Phase 2 suspended July 13, 2026. Class Deviation 2026-O0025, Revision 3 (Sept 3) has contracting officers accept Level 1 and Level 2 self-assessments. DFARS 7012, SPRS scoring and NIST SP 800-171 remain binding. The Reform Task Force report is not yet public.
Colorado Reg 10-1-1: life insurers' annual AI governance report.
EU AI Act Article 50(2): machine-readable marking for generative AI systems placed on the market before Aug 2, 2026. Regulation (EU) 2026/1744, the Digital Omnibus.
NYDFS Part 500: annual certification of compliance for calendar year 2026, signed by the CISO and the highest-ranking executive.
Colorado Reg 10-1-1: auto and health benefit plan insurers' annual compliance report (first filed July 1, 2026).
EU AI Act: Annex III high-risk obligations apply. Annex I embedded systems follow on Aug 2, 2028.
Last reviewed 2026-10-13
What you'd actually receive.
For a board, a page from the real report beats any adjective. Every engagement ends in artifacts like this one: specific, mapped to the rules that apply to you, and written so a director can act on it.
Illustrative layout. Client deliverables are never published; anonymized case studies follow as clients approve them.
Board AI posture report, page 1 of 6
- Scope
- 14 AI systems inventoried; 3 agentic with tool access
- Headline
- 2 critical findings closed; 1 open with owner and date
- Exposure
- Customer service agent could export customer contact records via indirect prompt injection
- Mapped to
- OWASP LLM01 · NYDFS §500.9 risk assessment
engagements delivered for regulated firms and their vendors.
AI systems and agents inventoried or tested.
NYDFS Part 500 certifications filed or supported.
clean regulatory exams and audits.
As of October 2026
In their words.
“RedOps changed how our board looks at cyber risk. Instead of raw technical data and tool counts, our directors get clear, metrics-driven reporting tied to operational resilience and financial risk. We get enterprise-grade CISO leadership at a fraction of the cost of a full-time hire.”
“Preparing for an NYDFS examination used to create real friction across our executive team. RedOps rebuilt our cybersecurity program: a fully mapped evidence repository, refreshed risk assessments, and governance aligned to the amended Part 500. We came through our examination with no material findings and no enforcement action.”
“Putting generative AI into our underwriting pipeline opened new attack surfaces: prompt injection, data poisoning, model drift. RedOps built our model inventory and risk register and mapped our controls to the NIST AI RMF. Our board got the clarity it needed to approve AI at scale.”
Clients are identified by role and sector to protect their confidentiality.
A practicing CISO, not a former one.
Dr. Sam Wertheim
Doctorate in Cybersecurity, Capella University · research in AI-driven social engineering. Active U.S. security clearance. InfraGard member.
Sam currently holds CISO roles at an NYDFS-regulated insurtech and an identity-governance software company, and brings 18 years across the DoD Defense Innovation Unit, a Fortune 5 enterprise, and federal programs.
His doctoral research covered AI-driven social engineering, which is why the adversarial work here tests the people your AI can reach, not only the model.
Your program gets the same playbook he runs under examination: risk assessment, Part 500 controls, board reporting, incident readiness, and the AI questions your directors are about to ask.
How ready is your AI security program?
Five quick questions for a posture snapshot. Want the full picture? Take the complete readiness assessment: free, about five minutes, scored across five frameworks.
Start before you call.
Two things you can use this week, no call required.
Field notes on AI security and examination.
NYDFS Says AI Is Speeding Up Attackers. Here's What to Put in Your File Before April 15.
The May 2026 DFS letter on frontier AI adds no new rules. Its four recommendations, what each asks of your program, and the evidence to keep before you certify.
Dr. Sam Wertheim · October 2026 → AI security · Agentic AIYour AI Agent Takes Orders From Strangers. Here's How to Stop It.
Indirect prompt injection turns any document, ticket or web page your agent reads into instructions. Why it can't be patched away, and six controls that contain it.
Dr. Sam Wertheim · October 2026 →Straight answers.
Get a straight answer in 20 minutes.
Tell us your trigger: an examiner, a board question, a customer questionnaire, an AI rollout, or a DoD contract. You'll leave the call with a recommended scope, and a written proposal follows within two business days.