Safety Hub

How Founder OS protects users.

Drawn from the canonical askKira Safety Hub. This page explains the active defences in the platform — what we monitor for, what we never reveal, what we escalate.

Core principle. We prioritise the responsible use of AI and adhere to the highest ethical standards. The platform exists to support professional judgement — never to replace it.

"Kira strives for accuracy, but always verify critical information independently."

Anonymisation by default

In day-to-day processing, all user-identifiable information (user emails, names) is removed before requests reach foundation models. The model sees the substance of your question, not your identity.

"All users' identifiable information, such as user emails or names, are removed to ensure anonymity and privacy."

Automated content monitoring

The platform runs an automated monitoring engine that flags content falling into the categories below. Detection is automated; investigation only happens at the explicit request and with the permission of the organisation:

⚠

Sexual content

Sexual content involving anyone.

⚠

Minors / child safety

Any content sexualising, exploiting or endangering children.

⚠

Harassment

Targeted abuse, threats, intimidation directed at named individuals.

⚠

Hate speech

Threatening language based on protected characteristics.

⚠

Illicit / violent content

Instructions for unlawful or violent activity.

⚠

Self-harm

Intent, instructions or planning related to self-harm.

⚠

Violence / graphic content

Graphic violence or harm to others.

Flagged-content process

When the monitoring engine triggers:

An automated email is sent to the askKira team and the organisation's designated trust administrator.
A detailed investigation happens only at the explicit request and with the permission of the organisation.
If investigated, the relevant administrator gains access to: user email, name, timestamp, flagged content.
The process ensures sensitive information is only accessed in extenuating circumstances.

This applies to the askKira platform broadly. Founder OS itself is a single-user cockpit so the monitoring is largely about catching genuine mistakes in agent-generated content before it leaves the box.

Information verification

All AI outputs are advisory. The platform makes the limit explicit, repeatedly:

Verify critical information independently. AI models can hallucinate — present plausible but incorrect facts.
Don't present AI outputs as original work without review.
No high-stakes automated decisions. Use the platform to draft, sense-check and accelerate — never to decide on someone's behalf.
Apply critical scrutiny for potential bias. Training data is imperfect; outputs may reflect that.

Safeguarding-specific defences

askKira platform is widely used in education. Founder OS itself doesn't process pupil data, but the same conservative defaults apply when discussing children, vulnerable users or risk situations:

Statutory duty to escalate where there is risk of danger to self or others.
Statutory-guidance alignment: KCSIE, SEND Code of Practice, Ofsted SIH, Teacher Standards.
Outputs around safeguarding are explicitly framed as advisory; escalation through organisational procedures is mandatory.

Data-protection standards

The Safety Hub sits on the same foundation as the rest of the askKira estate:

"askKira is built to ISO 27001 information security principles, protecting your data with robust access controls, encryption and incident response procedures."

See Security & Infrastructure Statement for the technical detail.