The AI That Forgets: Can Mix Solve AI’s Privacy Crisis?

SCOTTSDALE, AZ – August 18, 2026 – We are increasingly turning to artificial intelligence as a silent confidant. We ask it to draft sensitive business strategies, analyze legal disputes, and offer advice on our health and relationships. Yet, this digital confidant has a perfect, permanent memory. Every query, every vulnerability, is logged and stored, creating a personal data trove that is a magnet for security breaches, legal discovery, and corporate surveillance. A new service unveiled today by Modulus AI, Inc. aims to sever this link between utility and vulnerability. It’s called Mix, and its core design principle is to forget.

A Radical Approach to AI Amnesia

In a market dominated by models that retain conversation histories, Modulus AI's Mix is built on the opposite premise. "Some questions should not become part of your permanent AI history," the company states. This isn't just a policy setting; it's an architectural commitment. According to the company, Mix is designed to systematically anonymize user data before it is ever sent to a large language model for processing. The service evaluates queries to remove or substitute personally identifying information, effectively decoupling the user from their question at the earliest possible stage.

This stands in contrast to the standard enterprise privacy features offered by major AI providers, which typically focus on not using customer data for model training and providing data encryption. While valuable, those protections often still involve data retention unless explicitly configured otherwise. Mix’s approach is more aggressive. "If the service does not need to keep the question, it should not keep the question," said Richard Gardner, the founder of Mix, in today's announcement.

The company's commitment to privacy extends to its physical infrastructure. Mix keeps its billing and reasoning systems on entirely separate hardware. The billing system knows a user has purchased credits, but the reasoning system—the part that processes the sensitive query—has no knowledge of the user's identity. A temporary authorization token acts as the bridge, allowing the query to be processed without passing payment or personal identity into the reasoning environment. For industries like healthcare, where data breaches have become rampant, or finance, where confidentiality is paramount, this design could shift the calculus of risk for AI adoption.

More Brains, Better Answers

Privacy is only half of the equation. A confidential AI that gives flawed or nonsensical answers is of little use. Here, Mix introduces its second major innovation: a multi-model reasoning engine. Most AI products provide an answer from a single large language model. This has a known, critical weakness: a model can be confidently and persuasively wrong, a phenomenon known as hallucination. This has been a major barrier to deploying AI in mission-critical applications where accuracy is non-negotiable.

Mix tackles this by acting as a conductor for an orchestra of AIs. It can route a single problem to several frontier models, such as Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6 Sol Pro, assigning them independent roles. According to the company, these models analyze the problem, challenge each other's assumptions, critique weak conclusions, and reconsider evidence. Mix then examines any conflicts and synthesizes the strongest possible result. "One model can sound completely certain and still be wrong, which is one additional reason why we leverage multiple AI models," Gardner noted.

This ensemble method is well-supported by AI research, which has shown that multi-agent consensus frameworks can significantly improve factual accuracy and reduce hallucination rates. By being model-independent, Mix isn't tied to a single provider's ecosystem. It can flexibly route tasks to the best-performing models on the market, creating a system that is not only more private but also potentially more reliable and robust than its single-brained competitors.

Disrupting the Subscription Model

Modulus AI is also challenging the industry’s prevailing business model. Instead of a mandatory monthly subscription, Mix operates on prepaid usage credits. The introductory offer is $20 for one million tokens, which do not expire. This pay-as-you-go approach offers a level of cost predictability and flexibility that is often missing from the complex, tiered pricing structures of major AI platforms.

This model could prove highly attractive to a wide swath of professional users—from lawyers and consultants to small business owners—who may require powerful AI for intermittent, high-stakes tasks but are hesitant to commit to a recurring fee. For enterprises in finance and healthcare, the transparent, non-expiring credit system simplifies budgeting and eliminates the pressure to "use it or lose it," aligning costs directly with actual need.

A Foundation of Trust

While Mix is a new service, the company behind it is not. Modulus AI, Inc. boasts a technology heritage dating back to 1997. Founded by Gardner, the company has spent decades building high-performance, mission-critical systems for some of the most demanding clients in the world, including Nasdaq, JPMorgan Chase, and Goldman Sachs. Its technology is reported to help process a significant fraction of daily U.S. stock-market volume.

This deep experience in the trenches of financial technology and high-performance computing lends significant credibility to its ambitious claims for Mix. The company holds a robust portfolio of patents in AI and financial technology, cited by a who's who of tech and finance giants from Google to IBM. This is not a fledgling startup making promises; it's an established technology firm leveraging nearly three decades of expertise to solve a modern problem.

With AI becoming more powerful and integrated into our lives, the question of data trust has moved from a niche concern to a central challenge. By building a service on the principles of amnesia and distributed reasoning, Modulus AI is making a bold statement. "People should not have to choose between powerful AI, thoughtful answers, and privacy," Gardner said. "That's why we built Mix."