Skip to main content

Security & Trust Center

Security is our foundation

Built for the regulated tender workflows our customers run. Honest about where data lives, what we use, and what we don't do.

Data residency

Production runs on Google Cloud in europe-west1 (Belgium). For Australian customers a dedicated stack in australia-southeast1 (Sydney) is available on request: documents, vector embeddings, application database, and audit logs stay in Sydney. AI inference currently uses Google's global Gemini endpoint pending regional Pro-model deployment, and Bid Library documents are also indexed in a document store in Google's Gemini API with no regional commitment.

Google Cloud's engineering blog published a technical write-up of this architecture: running a global tender platform on AlloyDB and MCP.

No foundation-model training

Customer documents are sent to Google Gemini for analysis only. Per Google's Generative AI terms, paid-tier prompts and responses are not used to train Google's foundation models. Outside Google, document text goes only to the Azure checks above, and never for a workspace whose admin turned them off or that requires Google-only processing.

Encryption

TLS 1.2 or newer in transit; older versions are refused. Encryption at rest provided by Google Cloud (AES-256) for AlloyDB and Cloud Storage. The Australian deployment's database additionally uses customer-managed encryption keys (CMEK) via Cloud KMS, and connects with IAM-authenticated short-lived tokens rather than a password.

Checks on Microsoft Azure

Azure checks each tender pack for planted instructions (Prompt Shields), gives the bid call a second check (Azure OpenAI in the EU Data Zone) and reads scanned PDFs (Document Intelligence). The checks are on by default for workspaces created from 5 November 2026 and for all workspaces from 5 December 2026, except workspaces that require Google-only processing. A workspace admin can turn them off at any time in Settings, Data & compliance. Details are in the sub-processor list.

574 of 574 calls to the Azure checks service succeeded in a load run on 23 public tender packs on 5 Nov 2026 (482 Prompt Shields checks, 69 document scans, 23 second checks).

How we handle your data

What we store

  • Account details (email, name, company) in AlloyDB
  • Analysis results, drafts, and metadata
  • Billing records (Stripe customer + subscription IDs only, never card numbers)
  • Uploaded documents in Google Cloud Storage, encrypted at rest

What we never do

  • Send your documents to a foundation-model provider for training
  • Sell or share customer data with third parties for marketing
  • Co-mingle one customer's Bid Library with another's
  • Retain documents indefinitely after account deletion

Application-layer controls

Live in production today. Nothing on this list is roadmap.

  • Workspace roles (owner, admin, editor, viewer), and a sign-off that needs an approver other than the person asking
  • Audit logging of user actions and document access, with daily anomaly alerts
  • Customer-defined retention: tenders and audit records older than the days you set are deleted every night
  • Workspace export: one archive of members, tender details and the audit trail
  • Draft version history with one-click restore
  • Automated database backups with point-in-time recovery
  • Email alerts on production errors and downtime
  • Login brute-force protection: per-IP and per-account rate limits with lockout
  • Inbound email ingestion checks DMARC, or SPF or DKIM, against sender spoofing

Security documents for procurement reviews are available on request: data processing agreement, subprocessor list, security overview and policies. Email contact@ailucius.com.

Report a vulnerability

If you think you have found a security problem in Lucius, email contact@ailucius.com with what you found, the steps to reproduce it and the page, address or account involved. We reply within 3 working days and keep you updated until it is fixed.

While you test:

  • Use only an account you created yourself. Never open, change or delete another customer's data. If you reach any, stop and tell us.
  • No load or denial-of-service tests, no high-volume automated scanning, no social engineering of our team or our customers, no physical tests.
  • Give us a reasonable time to fix the problem before you tell anyone else. We suggest 90 days.

If you follow these rules in good faith, we will not take legal action against you for your research, and with your permission we will thank you by name on this page.

Rewards

We pay for security problems in Lucius that are new to us, in scope and reported to us first.

SeverityExamplesReward
CriticalReading or changing another customer's documents or data; taking over an account; running code on our serversEUR 500
HighGetting past sign-in or two-step verification; reading another workspace's file names or settingsEUR 250
MediumStored cross-site scripting that needs one click; a sign-in or code limit that can be bypassedEUR 100
LowA leak of internal details with no customer data; a security setting whose absence you can show an effect forEUR 50

We decide the severity. One reward per problem, to the first report that lets us reproduce it. We pay within 30 days of confirming the problem, by bank transfer or PayPal, and may ask for the details we need to pay you. Rewards are capped at EUR 2,000 in total per calendar year; when the cap is reached, this page says so and reports are still welcome and thanked.

In scope

  • ailucius.com and api.ailucius.com
  • The Lucius MCP endpoint and its sign-in flow
  • The Lucius agent API

Out of scope

  • Services run by others (Google Cloud, Stripe, our email, analytics and error-tracking providers): report those to them.
  • Scanner output with no working example; missing headers or settings with no shown effect; opinions on email records.
  • Denial of service, load tests, social engineering, physical tests.
  • Problems that need a rooted device, an out-of-date browser or access to the victim's unlocked computer.
  • Problems you can only trigger against your own account with no effect on anyone else.

We cannot pay people in countries under EU or US sanctions, or anyone who works for or with Lucius AI.

Security FAQ

Compliance posture

We're a small team. We'd rather be honest about where we are than claim badges we haven't earned.

GDPR aligned

Personal data handling, lawful basis, data subject rights, and EU residency.

Hosted on certified Google Cloud services

The Google Cloud services Lucius runs on (Cloud Run, AlloyDB, Cloud Storage, Cloud KMS, Secret Manager, Cloud Load Balancing, Cloud Armor, Memorystore and Generative AI on Vertex AI) are in scope of Google's ISO/IEC 27001, 27017, 27018 and 27701 certifications and its SOC 1, 2 and 3 reports: see Google's list of services in scope. Google Cloud is also certified at the High category of Spain's Esquema Nacional de Seguridad (ENS certificate B0278, issued by BDO, valid to October 2027), which lists the same services and Google's Belgian data centre. Google AI Studio, which workspaces without EU-only processing also use, is not on either list. These certificates cover Google's infrastructure, not Lucius: the application layer is ours, and Lucius holds no certification of its own yet.

Architecture published on the Google Cloud blog

In September 2026 the Google Cloud blog published how Lucius runs on AlloyDB for PostgreSQL: two regional database clusters (the Australian one under customer-managed keys), least-privilege database access for our operations agent, and vector search inside the database. Google reviewed the piece before publication. It describes our architecture; it is not a security audit.

Code and dependency vulnerability scanning in CI

Every push to the backend runs a Trivy scan of the code and its dependencies, which fails on high and critical CVEs that have a fix. It runs beside the deploy rather than blocking it.

Secret scanning and Cloud Armor

gitleaks in CI and as a pre-commit hook. Cloud Armor edge rules block scanner traffic and rate-limit per IP.

Azure Marketplace listing, co-sell ready

Lucius AI passed Microsoft's commercial marketplace certification and is listed on Azure Marketplace, alongside an approved Google Cloud Marketplace listing. As of August 2026 the solution is also co-sell ready: Microsoft validated our customer materials and lists Lucius in its sellers' internal solution directory. Marketplace certification reviews the vendor, the offer and its claims; it is not a security audit, so we list it here as a diligence signal, not a certificate. Read the announcement of 20 October 2026.

Cyber Essentials: in preparation

The cloud-side readiness work is done; the device and account checks and the certification itself are still to do. We will update this page when it is granted, not before.

Continuous monitoring against SOC 2 and ISO 27001 frameworks

As of July 2026, our EU cloud environment is monitored around the clock against the SOC 2 (2017 Trust Services Criteria) and ISO/IEC 27001:2022 control frameworks via Google Cloud Security Command Center, with automated misconfiguration detection and evidence collection. Security reviewers can request current evidence on specific controls.

SOC 2 / ISO 27001 certification: not yet

Continuous monitoring is not a certificate. Independent audits are on the roadmap as the team grows, and this page will say "certified" only when an auditor does. In the meantime we're happy to walk procurement teams through our setup, with evidence, over a call.

Sub-processors

  • Google Cloud Platform

    Purpose
    Compute (Cloud Run), database (AlloyDB), object storage (GCS), KMS, Vertex AI Gemini endpoint
    Location
    europe-west1 (Belgium) primary; australia-southeast1 for AU customer data at rest; Vertex AI global endpoint for inference
    Data handled
    Encrypted documents, application database, user metadata. Inference requests transit to Google's nearest available region for the chosen Gemini model
  • Google Gemini API

    Purpose
    Document analysis (extraction, compliance, drafting), embeddings, and a per-company document store for Bid Library documents
    Location
    Vertex AI global Gemini endpoint and Google AI Studio; for a workspace with EU-only processing on, Vertex AI's EU multi-region endpoint only
    Data handled
    Document text sent at request time, which Google may cache, encrypted and per project, for up to 24 hours; PDFs uploaded to Google's file service for reading, removed by Google within 48 hours; Bid Library documents kept in the company's document store until deleted; not used to train Google's foundation models per Google's terms for paid use
  • Microsoft Azure (Document Intelligence, Content Safety, Azure OpenAI)

    Purpose
    Checks on tender documents: reading scanned PDFs, checking packs for planted instructions, a second check on the bid call. On by default for workspaces created from 5 November 2026 and for all workspaces from 5 December 2026. A workspace admin can turn them off at any time in Settings, Data & compliance. Never used for workspaces that require Google-only processing, which includes every workspace with EU-only processing
    Location
    Sweden Central, West Europe (Netherlands); North Europe (Ireland) for the checks audit log (no document text); Azure OpenAI within the EU Data Zone
    Data handled
    Document text, for workspaces where the checks are on
  • Stripe

    Purpose
    Payment processing
    Location
    EU (Dublin) + global
    Data handled
    Customer billing details and card data, handled by Stripe; we never see card numbers
  • Resend

    Purpose
    Email delivery: product notifications, and the monthly newsletter for people who subscribe
    Location
    EU/US
    Data handled
    Email addresses and message content for product notifications and the newsletter
  • Customer.io

    Purpose
    Product and onboarding emails, for example getting-started tips and trial reminders, each with an unsubscribe link. From 17 November 2026, after 7 days' notice to account admins. Never used for workspaces with EU-only processing
    Location
    US (Customer.io's US data centre)
    Data handled
    Account email address and name, workspace plan and trial dates, product usage events such as "tender analysed". No document content
  • Sentry

    Purpose
    Error tracking and performance monitoring
    Location
    EU (Frankfurt, de.sentry.io)
    Data handled
    Stack traces, request paths, user IDs (no document content)
  • PostHog

    Purpose
    Product analytics (frontend usage)
    Location
    US (PostHog Cloud US, reached through a first-party proxy)
    Data handled
    Usage events on UI interactions: counted without cookies or an identifier until a visitor accepts; for signed-in users and visitors who accept, linked to the account, with masked session replays. No document content
  • Google Analytics 4

    Purpose
    Website analytics
    Location
    Google (global)
    Data handled
    Page views and events, only after a visitor accepts on the cookie banner, some relayed from our server; no document content
  • Anthropic (Claude)

    Purpose
    AI assistant the founder uses for operations and support
    Location
    United States
    Data handled
    Records the founder opens for a support case, normally counts and masked fields. Today a paid Claude Max plan with model training switched off; moving to the company's Claude Team plan under Anthropic's commercial terms in November 2026
  • Google Workspace

    Purpose
    Company email, including email with customers
    Location
    Google (global)
    Data handled
    Business contact data and the emails customers send us
  • Microsoft Azure Marketplace, Google Cloud Marketplace

    Purpose
    Purchases made through a cloud marketplace
    Location
    The marketplace's regions
    Data handled
    Billing contact data of customers who buy there
  • Microsoft Graph, Slack, Google Drive

    Purpose
    Integrations a workspace chooses to connect, under its own agreements with those providers: SharePoint import, Slack alerts, Drive file picker
    Location
    The provider's own regions
    Data handled
    Only what the connected integration reads or posts: the files you pick or import, the alerts you send

Have a security or procurement question? contact@ailucius.com

Found a vulnerability? Email the same address with details. Reports are read by the founder, acknowledged quickly, and researchers who want credit get it.