Security & Trust
A security product should be the most trustworthy thing in your stack.
These are design rules, not features that can be switched off by a sales promise.
On-premise first
ApexAI ships as an on-premise or private-cloud product, not SaaS. Your alerts, cases and models stay in your environment.
- Outbound connections only for security work: threat-intel feeds, CVE / PQC lookups and an optional, opt-in cloud LLM
- Tunnel agents connect private data sources over mTLS without exposing them to the internet
- Fonts and assets are self-hosted: nothing loads from a CDN
Local LLM by default
The default AI backend is a local, Ollama-compatible model that you host. No prompt or alert text leaves your network unless you turn a cloud LLM on.
- Cloud LLM only by explicit opt-in
- Model choice stays with you
AI gateway
If a cloud LLM is turned on, its calls go through Core’s server-side AI Gateway.
- Capability-scoped: each caller gets only what it is allowed to ask for
- The API key stays on the server, never in the browser
Post-quantum signing
Post-quantum cryptography is the default, not a roadmap item.
- Licences signed with ML-DSA-87
- Key exchange with ML-KEM-1024
- SOC decision policies signed with ML-DSA-65 and bound to their workspace
The human approval gate
Nothing that changes a real system — block, isolate, revoke, rotate — runs without a person’s approval. This is a fixed design rule.
- The most critical cases need two approvers
- Two-person rule configurable per action type
- A kill switch stops automated proposals
- Promoting a policy never widens what may run without approval
Honest by construction
An action with no connected backend says “not applied” instead of reporting success. Nothing in an API response is invented, and concept work is labelled as such.
- Every action result says whether it was really applied
- Evidence is kept for work people do by hand
- Concept previews are labelled and kept apart from product features
Responsible AI
The AI is measured against people, not the other way round.
- Every AI decision is recorded and can be reviewed
- Accuracy is counted against people’s verdicts, with reproducible scorecards
- Drift checks roll a policy back automatically
- Each tenant sees only its own data; isolation is tested
Compliance
ApexAI claims no certification on this site. Where we describe how the platform supports a framework such as [framework], we say what it is designed to support, not that it is certified. [Certification status — to be confirmed before publication.]
Reporting a vulnerability
Please report security issues to [security]@[domain]. We acknowledge reports within [N] business days.
Vulnerability disclosureBring your security team’s hardest questions.
We walk through deployment, data flows and the approval gate with your architects.