Built by Aurix AI Solutions. Profitable from day one.
Caspian persistent memory
An assistant that remembers projects across sessions—private and local.
Policy as code
YAML policies, prebuilt packs, and delta previews make governance auditable and repeatable.
Instant peace of mind
Flags AI-generated images, text, and voice as you scroll.
Founders: Those of Aurix AI Solutions
The Frontier Security Architect and his VP of Strategy.
Eric
Founder, Chief AI and Data Architecteric@aurixai.solutionsTwenty years building security and platform infra at companies you'd recognize and a few you wouldn't. Wrote most of Defog's detection core. Believes the best privacy tool is one your mom will actually keep installed.
Nicki
VP of Strategy and Partnershipsnicki@aurixai.solutionsProduct, design, and the human side of the platform. Came up through trust-and-safety and creator tooling. The reason Fluidity reads like English instead of audit-log.
Get in touch
We answer email ourselvesTell us what you're working on. The two of us read every note.
We built the company we wished existed when we were the ones being pitched.
— Aurix founders
For people
The world is getting harder to read. Feeds are optimized, voices are cloned, and the people most exposed — the young, the grieving, the simply busy — are handed no defense at all. We started here because trust is the thing chaos takes first. Fluidity exists to give an ordinary person back the half-second that manipulation is engineered to steal: quiet, default-on, and free, because the market will never build it for them.
For enterprises
Inside companies the same shift broke the old controls. Employees paste secrets into prompts your DLP cannot see; executives get wire requests in cloned voices. Shield is the missing observability layer — security, transparency, and interoperability by design. You keep your stack, hold your own keys, export your data, and add a layer that finally makes the invisible governable. We tell you what we cannot do, in writing, before you buy.
We built the company we wished existed when we were the ones being pitched.
What This Means
The AI industry has built a trust problem for itself, and Aurix refuses to participate in it. Most platforms treat their decision-making as proprietary magic — data flows in, outputs emerge, and users are expected to simply believe the black box. We built Defog on a fundamentally different premise: that you have the right to know what is synthetic and what is real, and that we have the obligation to tell you plainly, technically, and without evasion.
Transparency at Aurix is not a compliance posture or a marketing term. It is the architectural foundation. Every model we ship comes with documented capability boundaries, stated false-positive and false-negative rates, and a public changelog of detection accuracy changes. We do not quietly improve. We do not quietly degrade.
How We Enforce It
- Decision Audit Trails
- Every detection, every flag, every block is logged with full provenance — not just a timestamp, but the complete signal chain: which pattern fired, at what confidence level, against which heuristic version. Your compliance team can replay any event.
- Stated Limits
- We publish our detection boundaries. Where our models are uncertain, we say so in the product. Where coverage has gaps by content type, platform, or language, we name them publicly. Implied omniscience is not something we will sell.
- Public Mistake Process
- When our models produce significant false positive or negative rates affecting users at scale, we publish a summary of what happened, what changed, and what we are actively monitoring. We do not wait for a researcher to surface it for us.
- Honest Marketing
- We will not use benchmark scores from conditions that do not represent real-world usage. We will not describe our products with capabilities they do not currently have. If a feature is on a roadmap, we say so. If it is in beta, we say so.
What This Means
Privacy is not a feature we add after the product ships. It is a constraint that shapes the architecture from the first line of code. Every design decision at Aurix begins with the question: what is the minimum data required to accomplish this goal? The answer determines what we build — not the other way around.
Defog Fluidity is built on the principle that your device knows more about your browsing than we should. Client-side inference means the content you view, the pages you visit, and the prompts you write are processed locally — on your hardware — before they ever leave your control. We receive detection signals, not raw inputs.
Security is not a layer we bolt on after release. For Shield, it is embedded at the architectural level from day one: every data pipeline, every API endpoint, and every agent interaction is designed with the assumption that it will be audited, attacked, and scrutinised.
How We Enforce It
- Zero Payload Retention
- Detected content patterns are evaluated in memory and wiped from the C-heap after parse. We do not store what you typed, what you were about to send, or what feed content triggered a flag. Patterns fire and disappear.
- Customer-Managed Keys (BYOK)
- Enterprise customers on Business+ tier can bring their own encryption keys via AWS KMS, GCP KMS, or Azure Key Vault. At no point does Aurix infrastructure hold keys capable of decrypting your audit logs or event data.
- Third-Party Verification
- Our security posture is verified by an external penetration testing firm on a quarterly rotating vendor basis. The public summary of findings is available without NDA. The full technical report is available to customers under NDA.
- 72-Hour Breach Disclosure
- In the event of a confirmed breach affecting customer data, we will notify affected parties within 72 hours of confirmation — not 72 hours after we decide the investigation is complete. This is a commitment, not an aspiration.
- Hardware-Bound Device Identity
- Shield's device trust anchors to TPM 2.0 and Secure Enclave attestation. Per-device keys are not portable, not transferable, and revocable only through the control plane you operate.
What This Means
We build AI. We ship AI. We use it every day. We are not afraid of it. But we are deeply skeptical of any argument that AI governance should be invisible, that optimisation should outpace human comprehension, or that the speed of deployment should exceed the ability of people to understand what they are running.
Our products are built to augment human judgment — not to automate around it. Defog Fluidity gives individuals a second opinion on what they read and interact with. Defog Shield gives security teams the context to make better decisions faster — not to remove them from the loop. Human expertise and human accountability remain the final word at every consequential decision point.
How We Enforce It
- Core Protection Stays Free
- Basic AI detection, the manipulation lens, and the prompt firewall will remain available at no cost to individuals indefinitely. We will never paywall the ability to tell what is real. Where a paid tier exists, it adds capability — it does not remove what was already available.
- No Advertising, No Data Sales
- Defog is paid for by the people who use it. There is no advertising model, no data partner arrangement, and no insights sold to third parties. Our customers are the people we are trying to protect. That sentence cannot coexist with an ad-supported business model, so we chose not to have one.
- Interoperability Over Lock-In
- We add a layer on top of your existing stack. We do not replace your SIEM, your IDP, your endpoint agent, or your ticketing system — we federate to them. Export from Defog is a right, exercised through open OCSF, CEF, and JSON formats. Your data does not live here exclusively.
- Human Review on Consequential Actions
- No Defog Shield action that affects a user's access level, flags a file for deletion, or generates a compliance report to an auditor executes without a human confirmation step. This is not configurable. It is structural, enforced at the API layer, not the UI layer.
- The Right to Be Wrong
- Our models will make mistakes. Detection of AI-generated content is not a solved problem, and we will not pretend otherwise. We commit to publishing our error rates, maintaining an accuracy changelog, and never claiming certainty we do not have. When we are wrong publicly, we say so publicly.
Have any questions?
Talk to the team that built Defog — security and product engineers, not a sales desk.