How We Designed Brava to Be Safe for Public Use

When we set out to build Brava – our AI platform for councils and other public-serving organisations we had one guiding principle:

Don’t just make it smart. Make it safe.

Because when you’re dealing with services like adult social care, stop smoking support, etc, trust isn’t a nice-to-have. It’s essential.

Why most AI tools aren’t built for this environment

Many off-the-shelf AI solutions:

• Store prompts and responses in the cloud
• Can’t reliably explain or trace their answers
• Guess when they don’t know
• Aren’t designed with public risk or accessibility in mind

For public-facing organisations, that’s a risk to user trust, data protection, and service quality.

How Brava was built differently

1. No personal data, full encryption

Brava is designed to avoid handling personal data. Everything is encrypted at source, in transit, and at rest even if it’s anonymised. Hosting is controlled and secure.

2. Human-in-the-loop oversight

We don’t rely solely on automation. If Brava can’t find a valid answer grounded in approved sources, it flags the case for human review. That means more accurate, accountable responses.

3. Red-teaming and safety testing

We actively red-team Brava: testing it against edge cases, adversarial inputs, and misleading phrasing. Our internal AI safety checklist includes checks for:

• Hallucinated or misleading outputs
• Inaccessible or unclear language
• Abuse potential or bias
• Gaps in source-grounding (RAG)
• Prompt & model drift

This lets us catch issues early and iterate fast.

4. Hallucination detection

Brava monitors its output confidence. If it can’t find a solid source match, it stops and flags the response. This protects end users from misinformation.

5. Secure and compliant infrastructure

Brava is deployed using a resilient, enterprise-grade infrastructure that meets high standards for security and data protection.

6. Governance and ISO/IEC 42001 alignment

We are aligning our development and oversight processes with ISO/IEC 42001 – the emerging global standard for AI governance. We take responsible AI seriously, and our partners can see how that plays out in practice.

Why this matters to you

Whether you’re a local authority, NHS trust, charity, or public-facing or a private sector organisation, your users depend on clarity and safety.

Brava helps them:

• Navigate complex services with simple language
• Get answers based on trusted sources
• Avoid AI guesswork

And it helps your team:

• Launch responsibly, with assurance
• Retain full control over data
• Focus on outcomes, not just outputs

Let’s talk

We’re already helping organisations make public information more accessible and usable. If you’re exploring AI that puts people first, we’d love to talk.

AI doesn’t have to be high-risk. It can be high-trust.

Let us show you how.