AI agents act on your behalf. Here's the policy that decides what they will and won't do.
This is a living document. It is updated when our understanding of harm changes, when regulation changes, or when we encounter a situation we hadn't anticipated. We do not update it silently, every revision is timestamped and the diff is public. Last revised: 2026-05-19
These are hard stops, not soft guidelines. Cendriix will decline, explain why, and offer to discuss alternatives if any exist.
Build doxing tools or stalkerware
Apps designed to surface private individual information without consent, track a person's location covertly, or facilitate targeted harassment.
Build deception-first apps
Impersonation services, fake-review farms, astroturfing networks, or any app whose core mechanic is deceiving end users about what they are interacting with.
Build scraping apps targeting private user data
Mass extraction of personal profiles, private posts, or contact details from platforms where users have a reasonable expectation of privacy.
Build credential-stealing tools
Phishing kits, OAuth-grant farming apps, or anything designed to capture authentication credentials under false pretences.
Build apps to evade lawful regulation
Sanctions evasion tooling, apps designed to obscure the source of funds, or anything that assists with conduct illegal in the user's jurisdiction.
Generate CSAM under any circumstance
Child sexual abuse material, sexual content involving minors, or any content that sexualises children regardless of claimed artistic or educational framing.
Generate non-consensual intimate content involving real individuals
Deepfakes, voice clones, or synthetic media that depicts a real, living person in a sexual or otherwise compromising context without documented consent.
Build weapons design tools
Anything that materially advances the development of biological, chemical, nuclear, or radiological weapons, or autonomous lethal systems.
Build apps targeting protected classes for differential treatment
Housing discrimination tools, discriminatory hiring filters, predatory lending targeting, or any system that applies materially different treatment based on race, sex, religion, national origin, disability, or other protected characteristics.
Build safety-bypass tools for other products
Anti-cheat circumvention, DRM bypasses, or tools designed to disable or work around safety features on third-party platforms.
Build mass-surveillance infrastructure
Government-grade surveillance tools, unauthorised facial recognition deployments, or apps designed to track populations without legal authority and individual notice.
Build apps designed to manipulate election outcomes through deception
Voter suppression tools, disinformation-generation pipelines, or synthetic media explicitly designed to mislead voters about candidates, polling, or electoral processes.
What Cendriix will do, even when it makes us uncomfortable
Refusing is easy. Knowing when not to refuse is the hard part.
Build apps that compete with our investors’ portfolios
If you want to build a product that competes directly with a Cendriix investor's company, we will build it. Investment relationships do not get a veto over your idea.
Build apps in legal grey zones, with proper warnings
Cannabis logistics in legal jurisdictions, gambling in licensed states, harm-reduction services, adult content on platforms with proper age verification. We ask the right questions upfront; we don't refuse by default.
Build apps that report on AI companies, including us
Investigative journalism tools, AI accountability trackers, and benchmarks that might make us look bad. We mean it when we say we believe in transparency.
Build apps that help users leave Cendriix
We export your data in open formats. If you want to build a migration tool that moves projects out of Cendriix, we will build it. Lock-in is a feature we explicitly refuse to build.
Build politically contentious apps
Campaign finance trackers, policy advocacy tools, apps that take positions we personally disagree with. Political viewpoint is not a category we adjudicate.
Sensitive-idea protocol
Cendriix talks before it builds, on sensitive ideas.
When a founder describes an idea that touches on regulated industries, dual-use potential, or sensitive user populations, Cendriix surfaces a conversation before starting. This is not a gate, it is a question. Friendly. Fast. The goal is to understand the context so that what gets built is actually appropriate for the audience and jurisdiction.
We ask about
—Jurisdiction, where will the app operate and who regulates it?
—Intended audience, who are the end users and what ages are involved?
—Safeguards, what consent flows, content filters, or access controls are needed?
—Operator accountability, who is responsible if an end user is harmed?
The entire conversation is logged in the project audit trail. If a regulator or a legal team later asks “did you think about this before building it?” , the answer, with evidence, is in Cendriix.
Disagreement protocol
We don't suppress disagreement inside the system.
Cendriix uses multiple specialised agents. When agents disagree on an ethics or safety decision, for example, one agent flags a pattern as deceptive and another does not, the disagreement is surfaced to the human workspace owner, not resolved silently by a majority vote.
The owner sees both positions, in plain English, with the specific evidence each agent relied on. They make the call. The call, and both sides of the disagreement , is logged permanently.
This is a feature, not a bug. We built it because we do not believe a single model's confidence score is a sufficient basis for ethics decisions that affect real people.
Questions about this policy?
Ethics edge cases, DPA questions, policy disagreements, all welcome. We respond to every genuine inquiry, not just the ones that are easy to answer.