A security product, secured accordingly.
NoFly sits close to your code and your AI agents, so we hold ourselves to the standard we help you enforce. This page is an honest account of where our compliance program stands, how the platform is built to protect your data, and who we rely on.
Where an audit is not yet complete, we say so. We would rather earn trust than imply a certification we don’t hold.
- 01SOC 2 Type IIn preparation
Controls documented; report in preparation. Not yet audited.
- 02SOC 2 Type IIPlanned
Observation window follows Type I. Timeline shared on request.
- 03ISO/IEC 27001Planned
Information-security management system on the roadmap.
- 04ISO/IEC 42001Planned
AI management system, fitting for an AI-security vendor.
- 05GDPRAligned
Data-processing practices aligned; DPA available on request.
- 06HIPAAPlanned
For healthcare deployments, available to discuss for design partners.
How we secure the platform.
These are architectural properties of the product today, not aspirations: they are how NoFly is built.
Strict tenant isolation
Every query is scoped to your organization through a single tenant guard that refuses to run without a bound org: the guardrail against a missing tenant silently widening a read. One customer’s data is never reachable from another’s session.
Untrusted code never runs in-process
The dangerous part of scanning, connecting to an untrusted MCP server, happens only inside a disposable, hardened container: no network, read-only filesystem, all Linux capabilities dropped, memory and PID limits. The application itself never executes what it scans.
Keys are hashed, shown once
API keys are 24 bytes of CSPRNG entropy. We store only their SHA-256 hash; the plaintext is shown a single time and is unrecoverable afterward. A leaked database yields no usable credential.
Encryption in transit & at rest
All traffic is TLS. Data rests in managed PostgreSQL with encryption at rest. Secrets are supplied through environment configuration, never committed to source.
RBAC, MFA, SSO & SCIM
Role-based access control with a per-org editable permission matrix, enforced on the server, not just in the UI. TOTP two-factor is available, and enterprise SAML SSO with SCIM provisioning deactivates a user’s access on the next request.
Least-privilege agent identity
Non-human agents get first-class identities with capability policies (which models, tools and servers each may touch), authorized at runtime, with a live kill switch to revoke them instantly.
Tamper-evident audit log
Governance actions (approvals, policy changes, key issuance, member changes) are written to an append-only audit trail, exportable for your own records and reviews.
Runtime enforcement, fail-safe
The runtime firewall decides the dangerous majority on the machine with zero network, so protection survives a slow or unreachable backend and adds no latency to ordinary calls.
We run NoFly on NoFly
Our own repositories and AI agents are gated and firewalled by NoFly. Our detections are tested against a benign corpus so a rule can only ever get tighter, never noisier.
How we handle your data.
What leaves your machine, what we keep, and what never happens to it. The full detail is in the Privacy Policy.
- NOFLY01
Local-first by design
The developer CLI runs on your machine. Local checks (dangerous shell, prompt injection, secrets, exfiltration sinks) produce a real verdict with no backend and no key, so your source doesn’t leave the machine to be scanned. Enrolling only layers your org policy on top.
- NOFLY02
Your data never trains a model
Nothing you send us is used to train or fine-tune any model, ours or anyone else’s, and we do not sell or share it. AI augmentation is optional and inert until you enable it; when it does run, it calls Anthropic under commercial API terms, which carry no training rights. Leave it off and every verdict still stands: the deterministic engine is the floor of the product, not a fallback.
- NOFLY03
AI inputs are treated as untrusted
When AI augmentation is on, the metadata we send to the model is framed strictly as data; embedded instructions are treated as injection, not commands. With AI off, the product degrades cleanly to its deterministic rule engine: a scan never fails because AI is unavailable.
- NOFLY04
Data minimization & redaction
Findings scrub raw bodies and redact secrets before they are persisted. We keep what is needed to show your posture and history, not your whole codebase.
- NOFLY05
Retention, export & deletion
Scan history and findings persist until you delete them or close your account; deletions cascade to the mirrored records. GDPR access and portability are self-serve exports, and erasure (anonymization) is available to your org admins, with no support ticket to us.
Infrastructure & subprocessors.
The third parties that process data on our behalf. We keep this list current, and a machine-readable version is available to customers under a DPA.
We notify customers of material subprocessor changes in advance where required by your agreement. How we use them is set out in the Terms of Service and the Privacy Policy.
Found a security issue?
We want to hear from you, and we credit the people who report to us.
Try it without handing over anything.
The fastest way to evaluate our security posture is to run the CLI locally: no account, and nothing leaves your machine.