Fraud, dispute, and governed-AI services for financial firms and the agencies that oversee them. Cyber Uphold helps firms pause risky money movement, scale the pause to the risk, and document every reasonable-belief decision — with an accountable person making the call.
Part of the public record. The Hicks & Loeffel comment letter on FINRA Regulatory Notice 26-02 (filed March 9, 2026) is cited by FINRA in its filing with the SEC, SR-FINRA-2026-018 — on reasonable belief, federal reporting destinations, and extension documentation.
Federal Register ↗For twenty years, financial technology competed on speed. Adversaries turned that speed into a weapon. Authorized push payment scams, account takeover, and AI-enhanced social engineering all exploit how fast and final modern payments are. A short pause, away from the scammer's influence, is one of the most effective defenses.
FINRA's hold rules are permissive safe harbors. The safe harbor applies only when the rule's conditions are met — so the work is operational and evidentiary: show reasonable belief, run unforgiving business-day clocks, give notice on time, and keep a defensible file.
Protective friction is a feature, not a failure.
Intelligent Friction is the program behind Sentinel Desk. When credible risk signals line up, a firm pauses money movement, scales the pause to the risk, explains the decision in plain language, and leaves the reasonable-belief call with an accountable person.
We automate the paperwork — clocks, notice drafts, rationale, exam packs. We never automate the decision.
A low-risk anomaly gets a step-up prompt that takes seconds. A high-risk case gets the full hold or delay with escalated review.
Models gather signals, score, and draft. An authorized person attests to the hold, extension, decline, or release.
Every decision produces a versioned reasonable-belief package. Later events never rewrite earlier ones.
The customer hears what is happening, why, for how long, and how to reach the firm — without revealing detection methods.
Clocks, notices, attestations, and records are part of the workflow, not a bolt-on.
Illustrative only. In Sentinel Desk, confidence is a documentation weight — explicitly not a hold threshold.
03 · Vertical stacked lockup
The Checkpoint emblem translates Intelligent Friction into a single institutional mark. Nothing in it is decoration.
Together they form a protective shield — the firm's supervisory structure: policy, procedure, and record.
It interrupts the left flank. It stands for an accountable human principal reviewing and authorizing a regulated action before it executes.
Friction placed deliberately — a pause in the structure, exactly where a person decides.
Sentinel Desk runs FINRA Rule 2165 exploitation holds and proposed Rule 2166 fraud delays from intake to exam pack: a signal checklist, an attested human decision, a business-day clock, notice tracking, communication drafts, and an append-only audit timeline. Built by Mind Mesh AI.
Cyber Uphold Sentinel DeskProduct preview with a fictional firm and synthetic cases. Harborline Securities, its people, and all accounts are invented for demonstration.
New York business days with NYSE holidays. Rule 2165 at 15, 25, and 55 days, with 85, 115, and 145 labeled proposed. Rule 2166 profiles at 5 or 10 days.
icon-friction-clockExplicit attestation text for placing a hold or delay, declining, extending, and resolving. Roles map to the titles your WSPs authorize.
icon-human-principalSpecified Adult age, secrecy or isolation pressure, suspected coaching, new payee, unverified destination, and more. No customer-name field; a PII backstop rejects SSNs, emails, and phone numbers in free text.
icon-belief-packPathway-specific notice rules. The 2-business-day clock is satisfied only by a recorded notice, never by a draft. Email and SMS drafts are marked not sent.
icon-copy-kitHTML, Markdown, or JSON: signals, clock, notice obligations, decisions and attestations, extensions, rationale versions, and the immutable timeline.
icon-exam-exportAI agents can propose actions on a case. A supervisory principal approves or denies. Approval records the decision; it never runs the action itself.
icon-agent-gateIts non-goals are its best feature. A human principal places every hold.
The same codebase that runs Sentinel Desk exposes its evidence layer as versioned REST APIs, so your detection vendor, case system, or AI agents can produce the same defensible record — downstream of whatever detects the risk.
| Agent Gate risk class | Example actions | Default policy |
|---|---|---|
| Irreversible | Place or extend hold, place delay, resolve, decline, mark notice sent | Always needs a person. Never auto-allowed. |
| Consequential | Generate belief pack, edit checklist or narrative, create case | Needs a person. A named trusted agent may skip approval for belief packs only. |
| Reversible | Read or list cases, export exam or belief pack, draft notice | Allowed. A draft is not a send. |
| Unknown | Anything else | Denied. |
When the SEC acts on SR-FINRA-2026-018, retail broker-dealers will need to revisit written supervisory procedures, training, customer notices, and records. We help firms get ready now, with Sentinel Desk carrying the work.
A standing retainer that combines regulatory operations, AI, and cybersecurity expertise — scaled to your firm's size, with a Sentinel Desk pilot included.
Scope a retainer →Calibrate one disbursement path, map your alerts to friction levels, and walk a sample reasonable-belief file end to end.
Consumer-finance fraud and dispute expertise for financial firms and law firms.
The lowest-friction way in: versioned customer, advisor, and exam language, with updates as the rules change.
Scam typologies and hold/delay simulations built on synthetic cases.
Human approval before an AI agent moves money or takes a regulated action — the same Agent Gate discipline built into Sentinel Desk, applied to your own AI programs.
Rule 2165 applies today; Rule 2166 and the extensions are coming.
Consistency across branches and exam scrutiny at volume.
Instant money movement, APP and account-takeover scams.
Fiduciary duty and state senior-protection laws.
Elder-exploitation SAR volume and state hold laws.
APP scams, mule accounts, scam-liability regimes abroad.
Guardrails before agents take regulated actions.
Protecting beneficiaries and service members from scams and exploitation.
Mind Mesh AIPractice of Cyber Uphold LLC
Cyber Uphold LLC, through its Mind Mesh AI practice, builds the software, carries the thought leadership, and explains governed AI in plain language. Mind Mesh AI built Sentinel Desk and its APIs.
Visit mindmeshai.org ↗Every AI deployment has a named owner and a person in charge of the decision.
Firms keep custody of the code, prompts, and records their AI depends on.
Human approval layers aligned with emerging agent-payment standards from Google, Mastercard, and Visa.
Compliance buyers discount claims they cannot verify. So we separate what is built into Sentinel Desk today from what is on the production roadmap — and we list an attestation here only once it is held.
Pilots run on synthetic data until production-hardening milestones are met. Our founder holds CISO and CEH credentials and applies them to our own stack first.
The firm's credibility is the combination: one principal who has built and run technology companies and now works on governed AI, and one who has spent his career inside consumer-finance fraud and disputes.
Philip Loeffel is a three-time founder with successful exits, now focused on applied AI. He holds a Bachelor of Science in Computer Engineering and a Master of Science in Computer Science, and earned Chief Information Security Officer and Certified Ethical Hacker credentials in 2024, alongside the FINRA Securities Industry Essentials qualification.
At Cyber Uphold he leads the Mind Mesh AI practice and the build of Sentinel Desk, applying one rule to every AI deployment: own what you deploy, and keep a person in charge of the decision.
Chad Hicks is a service-disabled veteran of the United States Marine Corps and a consumer-finance fraud and disputes specialist. His expertise centers on Regulation E and the Electronic Fund Transfer Act, error-resolution and fraud operations, and serving as a named expert resource in arbitration.
His work sits on the operational side of the rules Sentinel Desk is built around: what a firm must do, by when, and what it must be able to show an examiner afterwards.
Co-authors of the Hicks & Loeffel comment letter on FINRA Regulatory Notice 26-02. Titles reflect the planned ownership change; the post-close structure takes effect at closing.
Cyber Uphold LLC is the legal contracting entity. Its SAM.gov registration, capability statement, proposals, invoices, and this website use the same names in the same order. Cyber Uphold claims only its own work.
Company data · Nov 2026| Legal entity | Cyber Uphold LLC, a Florida limited liability company |
|---|---|
| Location | St. Petersburg, Florida |
| UEI | WLUAHNA6QDW4 |
| CAGE | 9UJL6 |
| Primary NAICS | 541511 · Custom Computer Programming Services |
| SAM.gov | Registration active |
| Core capabilities | Fraud and hold-case operations · consumer-finance dispute and expert services · governed-AI and agentic guardrails · secure software development |
| Practice | Mind Mesh AI — the AI practice of Cyber Uphold LLC |
| Certifications | Socioeconomic certifications are listed here only once granted. |
| Point of contact | Philip Loeffel, Founder |
Tell us about your firm and the disbursement path you want to calibrate. We'll come back with a scoped briefing.