Digital Employees · Principles
What One3 will never promise, and the rules that keep it honest
Digital Employees work for your team under human supervision. They are measured against your previous process, disclosed as AI, and improved only through tested releases.
The positioning rule
Digital Employees work for the client's team, under human supervision, measured against the client's previous process. They are never positioned as replacing or competing with people. Public claims follow written rules, and every external claim needs an evidence link.
What One3 never promises
A hard list.
- 1
Guaranteed results
No guaranteed revenue, leads, rankings or ROI.
- 2
Ownerless growth
No fully autonomous business growth without human owners.
- 3
Silent learning
No agents that change behaviour in production without a versioned release.
- 4
Cloning
No cloning of any person without verified, revocable consent.
- 5
Workforce replacement
No replacement of a workforce, and no comparisons that pressure named employees.
- 6
Professional advice
No legal, tax, medical or investment advice from agents. They route to licensed humans through the Human Expert Cloud.
How One3 talks about AI
| Say | Never say |
|---|---|
| AI Digital Employee under your team's supervision | Replace your team / fire your staff |
| Measured against your previous process | Better than humans |
| Investment and return shown with the evidence behind it | Guaranteed revenue / guaranteed ROI |
| Improves through tested, versioned releases | Self-learning / learns on its own in production |
| AI avatar, always disclosed as AI | Indistinguishable from a real person / clone anyone |
| Acts within the limits and approvals you set | Fully autonomous business / runs your company |
| Recommendation with evidence and confidence level | The AI knows the best decision |
| Reverse or compensate, depending on the action | Undo anything |
Ethics and labour commitments
| Topic | Commitment |
|---|---|
| Disclosure | People always know when they interact with AI |
| No deceptive personas | No fake human identities, fake testimonials or undisclosed likeness use |
| Human dignity | Agents don't monitor, rank or discipline employees covertly; no emotion surveillance of employees |
| Human decisions on people | Hiring, firing, promotion, compensation, credit, clinical and legal outcomes are decided by humans; AI is advisory with explanations |
| Bias audits | Screening, matching and review assistants are audited for disparate impact before release and periodically |
| Reskilling | When agents absorb tasks, affected employees get learning paths toward supervising and higher-value work |
| Worker voice | Deployment checklists include consultation where works councils, unions or collective agreements apply (verify per jurisdiction) |
| Transparency to employees | Employees can see which agents work in their team and what data they access |
| Accountability | A named human is accountable for each agent |
Advisory roles
HR, legal and compliance, credit and clinical roles are advisory only. Their autonomy ceiling is A1/A2 with R0 for people-affecting decisions, and the UI requires a human rationale. Legal conclusions are advisory; lawyers decide.
Autonomy is earned
| Level | Name | What it does |
|---|---|---|
| A0 | Observe | Watches events, produces insights only |
| A1 | Suggest (shadow) | Proposes Actions; a human executes; agreement is scored |
| A2 | Draft | Prepares the Action; one-click approval |
| A3 | Act & notify | Executes within budget and policy; a human can reverse or compensate |
| A4 | Autonomous | Executes and optimises within the goal boundary; only R1–R3; R4 never above A2 |
Risk tiers
| Tier | Meaning | Default gate |
|---|---|---|
| R0 | Read-only | Permission check only |
| R1 | Reversible internal change | Allowed; reversal available |
| R2 | Reversible, visible to others | Allowed with notification |
| R3 | External or customer-facing | Approval unless autonomy has been earned |
| R4 | Irreversible, financial or legal | Always human approval plus a second factor |
Guardrails in order
Applied to every AI call and agent step.
Input
Tenant and actor resolution, PII masking, prompt-injection checks, tainted-source tagging.
Planning
Tool allow-list, budget check, autonomy level check.
Output
Schema and business-rule validation, safety classifier, groundedness check, brand and Constitution check.
Action
Risk tier gate, approval if needed, audit, reversal or compensation token.
Post
Sampling for human review, drift monitors, incident workflow, kill switch.
Your data stays yours
Cross-tenant isolation rules for rented agents.
Templates hold no tenant data
Only role logic, generic examples, synthetic scenarios and authored playbooks.
Everything else stays inside your boundary
Memory, playbook overrides, approved examples, adapters and ledger data — with row-level security and tenant-keyed encryption.
Shared templates learn only from safe sources
Synthetic data, One3's own usage, and opted-in tenants' anonymised, aggregated signals — never raw content.
No cross-tenant caches
Semantic, embedding and tool-result caches are tenant-partitioned.
Opt out of aggregates
Any tenant can exclude itself from all aggregate signals, regardless of plan.
Audited continuously
Cross-tenant leak tests on every relevant change, weekly canary-token audits, deletion verification on every offboarding.
Questions about responsible AI
Will One3 replace my staff?
No. Digital Employees work for your team under supervision, measured against your previous process.
Is my data used to train shared models?
Not by default. Opt-in is per tenant with a contract clause, and shared templates learn only from anonymised, aggregated signals above minimum thresholds.
Can an agent sign a contract or make a payment?
No. Those are R4 Actions and always need a human approval with a second factor.
What if an agent says something wrong to a customer?
The incident flow contains it, compensates where possible, replays the run to find the cause and adds new eval cases.
Are your compliance claims certified?
One3 names only certifications it actually holds. SOC 2 and ISO 27001 work is planned during W2; none is claimed today.
Read the rules before you rent
Ask us for the full claim rules, guardrails and isolation tests.