Responsible AI
Capable where it helps. Bounded where it counts.
Mireto puts AI in front of real decisions about strategy, investment and people. That only works if the output is trustworthy. This page sets out the principles we hold ourselves to, how we keep the analysis honest, how your data is handled and, just as importantly, what we do not claim.
The engineering detail, the seven quality layers, the deterministic scoring and the citation checks, lives on our AI Quality page. This is the plainer statement of how we work and why you can rely on it.
Last updated: June 2026
Our statement
We use AI to do in days what used to take a consulting team months. We do not use it to guess. Numbers are calculated or sourced, never invented. Findings are grounded in live evidence and labeled by confidence. And every Complete AI Strategy deliverable is read and signed off by the delivering partner before it reaches you, because a person, not a model, should stand behind the advice you act on.
Commitments
Six things we hold ourselves to
Grounded in evidence
Findings come from live research and your own inputs, not from a model’s memory. Every figure is sourced, calculated or clearly labeled as an estimate.
A person is accountable
On Complete AI Strategy, the delivering partner reviews and signs off before release. The AI does the work; a person stands behind it.
Fair to every organization
The same analysis runs for a council, a charity, a school and a corporate. The language and metrics adapt to your entity type. The rigor does not.
Transparent about confidence
Every finding is tagged Evidenced, Inferred or Hypothesis, so you always know what is verified and what still needs validation.
Private by default
Your data is scoped to your account, stored in your jurisdiction and never used to train AI models.
Honest about limits
We tell you what the analysis does not cover, and we never present Mireto as a substitute for professional advice.
Accountability
A human signs off the advice you act on
The most important guardrail in Mireto is not code. It is a person. Every Complete AI Strategy deliverable is reviewed by the delivering partner before it is released. They read what the AI has produced, check it against your context and stand behind it with a signed cover letter. The AI does the analysis. A qualified human is accountable for what you receive.
On every scan, the AI Disruption Snapshot is delivered with a data quality grade so you can see how much evidence the analysis had to work with. Scans that ran on thin data are flagged, so you can weigh the findings accordingly.
Honest output
Grounded in evidence, not guessed
A general-purpose AI asked to assess your organization will write confident text, invent statistics and cite sources that do not exist. Mireto is built to prevent that.
Live research, not memory. Every assessment starts with real-time web research into your organization, its peers and your industry. Peer companies come from live search results rather than recalled training data, and business registration details are verified against the relevant national register.
Scores are computed, not improvised. Claude judges each of eight dimensions against a calibrated rubric with published numeric boundaries. The AI Disruption Score itself is then arithmetic: a fixed weight table applied identically to every organization, so the same dimension profile always produces the same score. Where your own documents refine a dimension, the change is capped at ±2 and must cite specific evidence.
Sources are checked. Every citation is cross-referenced against the research the scan actually gathered. Unverified sources are flagged before they reach you.
Confidence is labeled. Every finding is tagged Evidenced, Inferred or Hypothesis, so you can tell grounded fact from reasoned assessment at a glance.
These are a few of the controls. The full system, seven independent quality layers between the AI and you, is set out on our AI Quality page.
Fairness
Fair and neutral by design
Mireto analyzes councils, charities, schools, small businesses and corporates on the same basis. The rigor does not change with the type of organization. The language does.
Entity-aware language. A government agency sees citizens and outcomes, not customers and revenue. A not-for-profit sees impact. The metrics adapt to what your organization is actually there to do.
Discipline about people. The platform does not make negative assertions about named individuals without a sourced citation. Where evidence is absent, it says “no public evidence of” rather than stating a definitive negative.
Anchored to the right jurisdiction. Australian scans are held to Australian regulatory and market context. US scans use US regulatory context and US data sources, so the analysis matches where the organization operates.
Privacy
Your data stays yours
Scoped to your account. Your scans, uploads and advisor conversations are visible only to you. Row-level security enforces that isolation at the database.
Stored in your jurisdiction. Scan data is stored in the jurisdiction of the organization being assessed, determined automatically: Australian scans in Sydney (AWS ap-southeast-2), US scans in a dedicated US region. Encrypted in transit and at rest.
Not used to train AI models. Data sent to the Anthropic Claude API is not used to train models, under Anthropic’s commercial terms. Your assessment is never shared with other users, and there is no third-party tracking or advertising on the platform.
Filtered before processing. Document uploads and workshop conversations pass through a filter for Australian identifiers (ABN, TFN, Medicare) and universal patterns such as card and bank account numbers before any AI call.
Governance
Human oversight and governance
Models are pinned. We run specific, version-controlled Claude models chosen for each task. They are not swapped silently, because the choice affects accuracy and behavior.
Reference material is curated, not scraped. The frameworks, and for government scans the policy library the advisors draw on, are assembled by hand and reviewed by people. The AI reasons with that material. It does not invent it.
We pressure-test and monitor. We review real output for weaknesses, tighten the underlying guardrails when we find one and re-check that the fix holds. Errors are captured through Sentry, so a failed feed or generation surfaces to us rather than quietly degrading your result.
Jurisdiction
Built for the rules you actually operate under
Mireto is built for the rules your organization actually operates under, not a one-size-fits-all default. Australian scans are handled in line with the Australian Privacy Principles under the Privacy Act 1988 (Cth) and anchored to Australian regulatory and market context; US scans are handled under US privacy law and anchored to US regulatory and market context. For government scans, the advisors reference the live policy library the relevant jurisdiction’s agencies are required to comply with.
Our approach reflects the principles of trustworthy AI, wherever you operate. It is reliable and safe, transparent about its reasoning and its limits, fair to every organization it assesses and accountable to the people who rely on it.
Limits
What we do not claim
No AI system is infallible, and we will not pretend ours is. The controls on this page are designed to make a wrong answer rare and a fabricated figure rarer still. They are not a guarantee. That is why every finding carries its confidence, why estimates are labeled as estimates and why the delivering partner signs off Complete work.
Mireto provides strategic analysis and decision support, not regulated financial, legal or accounting advice. The assessment is a considered starting point for a board conversation, not a substitute for professional advice or your own due diligence. Verify what matters before you act on it.
See our disclaimer and terms for the full position.
Questions
Who is responsible
Mireto is operated by High Impact Group Pty Ltd (ABN 40 682 923 128). If you have a question about how we use AI, how a finding was reached or how your data is handled, get in touch.