TRUST · RESPONSIBLE AI
Useful, controlled, and used with judgment.
Edrak gives organisations access to powerful models through a governed platform, with safeguards, visibility and clear accountability for how AI is used in real business environments.Powerful models through a governed platform, with safeguards, visibility and clear accountability for how AI is used.
AI should support human work. Where judgment matters, the judgment stays with people: organisations and users remain responsible for the decisions they make, including decisions informed by AI outputs.Where judgment matters, the judgment stays with people.
Four principles
How the platform is designed to be used
01
Useful
AI should deliver real value in workflows where speed, scale and productivity matter.Real value in workflows where speed, scale and productivity matter.
02
Controlled
Use takes place within a managed environment with visibility, permissions and operational safeguards.A managed environment with visibility, permissions and operational safeguards.
03
Reviewable
Outputs can be reviewed, challenged and validated by the people using them, with their sources attached.Outputs can be reviewed, challenged and validated, with sources attached.
04
Accountable
Organisations and users remain responsible for how the platform is used and how outputs inform consequential decisions.Organisations and users remain responsible for how outputs inform decisions.
What responsible use means in practice.In practice
Edrak is built for responsible enterprise use, with oversight designed in rather than added later.
01
Access to AI takes place through a governed platform layer, with Foundry applying permissions and masking before any model sees the dataAccess through a governed platform layer; Foundry applies permissions and masking before any model sees the data
02
Organisations manage users, permissions and workspace activity from one placeUsers, permissions and workspace activity managed from one place
03
Atlas acts only within the authority a workflow was given; consequential steps wait for a named approver
04
Every output carries its citations so it can be checked before it is relied on
05
High-impact uses require human judgment and the organisation's own internal controlsHigh-impact uses require human judgment and internal controls
Output limitations
Outputs can be wrong. Review them.
AI systems produce useful outputs quickly and can also produce errors. Outputs may be incomplete, inaccurate, misleading, out of date, or unsuitable for a specific operational, legal, financial or regulatory use.
Edrak expects customers to apply human review before relying on outputs in high-impact contexts. AI-generated content is not a substitute for professional judgment, legal review, compliance review, financial sign-off or other required decision processes.
Outputs may be incomplete, inaccurate, misleading, out of date or unsuitable for a specific use. Human review applies before relying on them in high-impact contexts. AI content is not a substitute for professional, legal, compliance or financial sign-off.
Safeguards and platform integritySafeguards
Built to resist misuse.
- Access controls and permission structures
- Monitoring for misuse, abuse or suspicious activityMonitoring for misuse and suspicious activity
- Restrictions on attempts to bypass safeguardsRestrictions on bypassing safeguards
- Restrictions on harmful or prohibited use casesRestrictions on harmful or prohibited uses
- Enforcement where needed to protect customers, the platform or legal complianceEnforcement to protect customers and the platform
Prohibited uses
Users may not use Edrak to facilitate malware, phishing, credential theft, unauthorised surveillance, social engineering or other harmful conduct. Users may not attempt to bypass platform safeguards, manipulate the system through jailbreaking or prompt injection, or use the platform to build competing AI systems without authorisation. See the Usage policy for the full terms.No malware, phishing, credential theft, unauthorised surveillance, social engineering or other harmful conduct. No bypassing safeguards, jailbreaking or prompt injection, and no building competing AI systems without authorisation. See the Usage policy.