This interview is with Ash Sobhe, CEO, R6S.
For Featured readers, how would you introduce your role as CEO of R6S in the computer software space and the lens it gives you on contracts, IP, compliance, and finance?
I run R6S, a private AI systems company. We build intelligent software that lives on our clients’ own hardware, trained on their own data, with zero cloud dependency. Our clients are business owners running $1M to $300M operations who need AI that actually works for them, not the other way around.
What makes my perspective different on contracts, IP, compliance, and finance is that I sit at the intersection of all four every single day. When you are building private AI systems for clients, you are handling their most sensitive business data, their proprietary processes, their client lists, and their financial models. That forces you to think about IP ownership, data governance, and compliance from day one, not as an afterthought.
With over 20 years in technology, I have negotiated and reviewed hundreds of software contracts, licensing agreements, and service agreements. I have seen how badly things go when IP ownership is ambiguous or when compliance language is treated as boilerplate. In the AI space specifically, the questions around who owns the model, who owns the training data, and who is liable when the system makes a decision are still being figured out in real time. I deal with these questions with every client engagement.
On the finance side, running a company that builds custom AI systems means understanding both the technology economics and the client’s business economics. I am constantly evaluating build-versus-buy decisions, total cost of ownership, and ROI frameworks that help business owners understand what they are actually paying for and what they are getting back.
What key experience in your shift from digital marketing to leading private, on‑prem AI most shaped how you approach legal risk and financial decisions today?
The shift happened gradually, then all at once. Over a decade ago, I got an early look at IBM Watson when it was still a mainframe system focused on analytics and marketing intelligence. At the time, I was deep in digital marketing, working with Fortune 100 and Fortune 500 clients on campaigns where every dollar had to be justified.
We were already doing what I would now call primitive AI. We built custom logic using programmatic rules to A/B test everything from background colors to messaging sequences, then automatically adjusted the next touchpoint based on the results. The goal was always the same: higher return on ad spend, better conversion rates, and full compliance with each client’s brand and regulatory requirements.
That work forced me to understand two things simultaneously. First, the financial pressure. When your clients are Fortune 100 companies demanding measurable ROI on every campaign, you learn very quickly how to evaluate risk, justify spend, and structure agreements that protect both sides. Every contract had to account for data ownership, performance guarantees, and liability if a campaign underperformed or crossed a compliance line.
Second, the human element. I independently studied psychology through Yale’s program because I realized that the technology was only as good as our understanding of why people make decisions. That combination of technical capability, financial accountability, and behavioral science became the foundation for everything I do today.
When I made the full transition to private, on-premise AI systems, those instincts carried over directly. Now instead of optimizing ad campaigns, I am building intelligent systems that handle clients’ most sensitive data, their contracts, their financial models, and their proprietary processes. The legal and financial stakes are exponentially higher, but the core discipline is the same: understand the risk, structure the agreement properly, measure the outcome, and never lose sight of the human being on the other end of the decision.
Starting with contracts, when you structure a strategic partnership or enterprise AI engagement, what single clause or mechanism do you insist on because it most protects long‑term value?
The one clause I never compromise on is full intellectual property and data ownership staying with the client from day one. Not after a transition period, not after final payment, not after some licensing window. From the moment we begin building, everything we create on their infrastructure belongs to them.
This matters because the entire value proposition of private, on-premise AI is sovereignty. If I build a system that runs on a client’s hardware, processes their proprietary data, and learns their business logic, but I retain any claim to that IP, I have fundamentally undermined the reason they hired me in the first place. They came to me because they do not want a third party holding the keys to their intelligence layer. That includes me.
What I retain is the methodology: how I approach the architecture, the research process, and the deployment framework. That is my craft and my competitive advantage. But the output— the trained models, the integrations, the workflows, the data— all of it is theirs. They can fire me tomorrow, and everything still runs.
This clause does two things strategically. First, it builds an enormous amount of trust upfront. When a business owner realizes you are not trying to create dependency, they lean in harder. They give you more access, more data, and more latitude to build something truly powerful. Second, it protects long-term value for both sides. The client is never locked in, which paradoxically makes them more likely to stay and expand the engagement. And I am never liable for how they use their own system after handoff, because it is their system.
Most vendors in this space do the opposite. They want recurring SaaS revenue, they want to host your data, and they want you dependent on their platform. I structure my contracts to make myself replaceable on purpose. That is the single biggest trust signal you can send to a sophisticated buyer, and it is the reason my clients refer other clients.
On IP specifically, how do you define and document ownership across data sets, fine‑tuned models, prompts, and generated assets so innovation can scale without future disputes?
I break IP ownership into four distinct categories in every engagement, and each one gets its own section in the contract. This is not something you can leave ambiguous, because the moment a dispute arises, vague language around “the AI stuff” will destroy a business relationship.
The first category is raw data. Any data the client provides or that the system collects from their operations belongs entirely to the client. Full stop. I never touch it, I never copy it, and I never retain it after the engagement. This is non-negotiable and it is the foundation of trust in private AI work.
The second category is fine-tuned models and configurations. When we train or configure a model on the client’s data using their hardware, that resulting model is theirs. It was shaped by their proprietary information, it runs on their infrastructure, and it reflects their business logic. I document this explicitly because this is where most vendors get greedy. They will argue that because they did the training, they have some claim to the output. I reject that completely. If a carpenter builds you a custom bookshelf in your house with your wood, they do not own the bookshelf.
The third category is prompts, workflows, and system architecture. This is where it gets nuanced. The specific prompts and workflows I build for a client are theirs; they are custom to their business and paid for under the engagement. However, the general methodology and architectural patterns I use across engagements remain mine. I am transparent about this distinction upfront. A client gets full ownership of their specific implementation, but I am not signing away my ability to use similar approaches for other clients in different industries.
The fourth category is generated assets: anything the AI produces during operation, such as reports, analyses, communications, and creative output. All of it belongs to the client. This seems obvious, but you would be surprised how many AI service agreements include clauses that give the vendor rights to “aggregate” or “anonymize” generated outputs for their own use. I do not do that.
Every engagement includes a simple IP schedule that lists these four categories with clear ownership assignments. Both sides sign it. There is no room for interpretation. When you build on this kind of clarity from day one, innovation scales naturally because nobody is looking over their shoulder wondering who owns what.
In regulated deployments, what does your pre‑launch compliance playbook look like to ensure security, privacy, and auditability hold up once the system hits real operations?
Our compliance playbook starts with a decision that eliminates most of the problems other companies spend months trying to solve: we deploy everything on hardware the client physically owns. No cloud, no third-party data processors, and no vendor chain to audit. When a regulator asks, “Where does the data live?” the answer is, “In that machine, in that room, in that building.” That conversation is over in five minutes instead of five months.
The playbook itself has three phases. Before we write a single line of code, we map every data flow the system will touch: what goes in, what comes out, what gets stored, and what gets discarded. In regulated industries, the biggest compliance failures come from data that was never supposed to persist but ended up cached somewhere nobody thought to check. We design for data minimization from day one. The system processes information in memory, returns the result, and the raw input does not hit long-term storage unless there is a documented regulatory reason for retention.
During the build, we enforce AES-256 encryption at rest and in transit, role-based access controls, and comprehensive audit logging. Every interaction with the AI system is logged with timestamps, user identity, and the nature of the query. Not the content of sensitive queries themselves, but enough metadata that an auditor can reconstruct who accessed what capability and when. We run local large language models so inference happens entirely on-premise, with no API calls sending regulated data to external servers.
Before launch, we run what I call an adversarial audit. We try to break our own system. Can a user with standard permissions access data outside their scope? Can the AI be prompted to surface information it should not have access to? Does the audit trail hold up if we simulate a compliance review six months from now? We document every finding, fix what needs fixing, and the client’s compliance team signs off before we go live.
The key insight most companies miss is that compliance is not a layer you add on top; it is an architectural decision you make at the foundation. When the infrastructure is private, encrypted, and air-gapped from external services, you have already eliminated the majority of compliance risks before the conversation even starts. The companies struggling with AI compliance are almost always the ones who deployed on shared cloud infrastructure and are now trying to retrofit privacy controls onto a system that was never designed for them.
Translating that to go‑to‑market, what single safeguard do you now bake into every performance marketing or SEO program to avoid IP and advertising‑law pitfalls across platforms?
Every piece of outbound marketing at R6S runs through what I call a “provability filter” before it goes live. If we cannot prove the claim with a specific, verifiable example, it does not ship. No vanity metrics we cannot back up. No client results we cannot document. No implied guarantees about AI performance that we cannot demonstrate on demand. This sounds simple, but it eliminates the majority of advertising-law exposure that AI companies stumble into.
The industry is full of marketing that says things like “our AI increases revenue by 40%” or “deploy in two weeks and see immediate ROI.” Those claims are lawsuit magnets because AI is inherently unpredictable. A model that performs brilliantly on test data can hallucinate in production. A system that works for one client’s data structure may fail on another’s. Promising outcomes you cannot guarantee is how companies end up with FTC complaints and client lawsuits simultaneously.
Our marketing uses the word “research” deliberately. We position our service as research and development of private AI systems, not as guaranteed business transformation. This framing is honest because that is genuinely what we do, and it insulates us legally because research carries an inherent acknowledgment that outcomes vary. Our contracts reflect this too: no guaranteed results, no performance warranties on AI output, and clear documentation that large language models can and do hallucinate, and that the client accepts this as an inherent characteristic of the technology.
On the SEO and content side, we never use competitor names in paid advertising or comparison pages. We never make claims about competitors’ products that we have not independently verified. And we never fabricate case studies, testimonials, or client stories for marketing purposes. Every example we reference in public content is real, verifiable, and used with appropriate anonymization to protect client confidentiality.
The single safeguard that makes all of this work: we treat marketing the same way we treat client deployments. Everything must be auditable. If someone challenges a claim, we can produce the evidence behind it. If we cannot, the claim does not exist in our marketing. This approach costs us some flashy headlines, but it means we have never had a compliance issue, a takedown request, or a client dispute over misrepresented capabilities.
On negotiations, what preparation routine consistently improves your outcomes in high‑stakes agreements and large purchases?
I walk into every high-stakes negotiation with an unfair advantage: a team of AI employees that has been preparing for weeks before the meeting even starts. My AI system listens to every client call, reads every forwarded email, tracks my calendar, and maintains a living dossier on every relationship and deal in progress.
By the time I sit down for a negotiation, the AI has already analyzed every prior conversation with that person, surfaced relevant market data, identified leverage points I might have missed, and prepared a briefing document tailored to that specific meeting. I do not prepare for negotiations; my AI team prepares for me, continuously, in the background.
But it goes further than that. I wear Meta smart glasses with AI integration. With the client’s permission, my conversations are captured in real time. The AI sees what I see: documents on a table, whiteboards, and body language cues. During a meeting, I can quietly ask my AI system to pull up a specific contract clause, verify a claim someone just made, or calculate the financial impact of a term being proposed.
The other side is negotiating with what they brought to the table. I am negotiating with the entire intelligence infrastructure of my business sitting behind my eyes. This changes the preparation routine from an event into a continuous process. Traditional negotiation advice suggests spending two hours researching before a big meeting. My system has been researching since the first interaction with that contact. Every email, every call transcript, and every piece of market intelligence is synthesized and available the moment I need it.
When someone across the table references a conversation from three months ago, I have the exact context instantly. The result: negotiations feel less like adversarial exchanges and more like informed conversations where I can focus entirely on the human relationship because the data work is already done. The person with better information sets the terms. I have a full-time intelligence team that never sleeps, never forgets, and has been building context on every deal since day one.
Thanks for sharing your knowledge and expertise. Is there anything else you'd like to add?
The one thing I wish more AI founders would be honest about is that most businesses do not need AI; they need better systems. AI is a multiplier, and multiplying zero is still zero. If your operations are chaotic, your data is a mess, and your team does not have clear workflows, deploying AI will amplify that chaos at machine speed. The companies getting real value from AI are the ones that had strong fundamentals first.
That said, when AI is deployed correctly, on clean data, for specific workflows, with clear success metrics, it is the most transformative technology I have seen in over 20 years of building businesses. I have watched a single AI deployment compress what used to take a team of five people an entire week into something that happens automatically before anyone starts their morning. Not because the people were replaceable, but because 80% of what they were doing was collecting, organizing, and summarizing information that a well-built AI system handles instantly. Those people now spend their time on judgment, relationships, and strategy—the work that actually moves a business forward.
The biggest opportunity right now is not in building AI products; it is in deploying AI infrastructure that businesses own permanently. The subscription model that dominates the industry, where you pay monthly for access to someone else’s AI running on someone else’s servers using your data, has a fundamental flaw: you are building value for the vendor, not for yourself. Every prompt you send, every workflow you build, and every piece of institutional knowledge you feed into a cloud-based AI system belongs to someone else’s ecosystem.
The founders who will build the most durable businesses in this era are the ones helping companies own their AI the way they own their buildings, their equipment, and their intellectual property: private hardware, private models, private data. When you own the infrastructure, every improvement compounds in your favor, not your vendor’s. I built R6S on that conviction, and every deployment reinforces it. The future of AI is not access—everyone has access. The future is ownership.