AI Strainer roll-out: when AI agent autonomy needs Human auditable agency
5 Oct 2026
AI agents are getting more autonomous. Our controls need to become more explicit.
Recent incidents involving AI agents acting outside their intended scope are making one thing increasingly clear: telling an AI what it should do is not the same as being able to prove what it was actually allowed to do.
This what we designed our SDK for, to solve two sides of the same problem.
- For individuals: people need a way to express their AI preferences once and have those preferences respected consistently across AI providers and applications (rather than renegotiating privacy settings every time their data reaches another AI system).
- For organisations: companies need a simple, auditable source of truth for AI authorisation and data protection settings (one that cannot simply be rewritten after the fact by the application or the AI provider being audited).
Our Product is not about adding another dashboard. It is about moving from: “We have a policy saying what should happen.” to: “We can demonstrate, for this specific AI request, what was authorised, on what basis, and what setting was actually enforced.”
Whether the proposed GDPR Article 88B ultimately becomes law, whether another Hugging face hack becomes a major UN concern or not, organisations will increasingly need this kind of technical capability simply to comply with GDPR and the EU AI Act: a persistent, user-controlled and enforceable signal for individuals, and a trustworthy source of truth for organisations.
This small, verifiable control layer between an application and the AI services it calls is the technical service we designed as an "AI strainer" and we roll it out in Q4 2026.
Recent incidents involving AI agents acting outside their intended scope are making one thing increasingly clear: telling an AI what it should do is not the same as being able to prove what it was actually allowed to do.
This what we designed our SDK for, to solve two sides of the same problem.
- For individuals: people need a way to express their AI preferences once and have those preferences respected consistently across AI providers and applications (rather than renegotiating privacy settings every time their data reaches another AI system).
- For organisations: companies need a simple, auditable source of truth for AI authorisation and data protection settings (one that cannot simply be rewritten after the fact by the application or the AI provider being audited).
Our Product is not about adding another dashboard. It is about moving from: “We have a policy saying what should happen.” to: “We can demonstrate, for this specific AI request, what was authorised, on what basis, and what setting was actually enforced.”
Whether the proposed GDPR Article 88B ultimately becomes law, whether another Hugging face hack becomes a major UN concern or not, organisations will increasingly need this kind of technical capability simply to comply with GDPR and the EU AI Act: a persistent, user-controlled and enforceable signal for individuals, and a trustworthy source of truth for organisations.
This small, verifiable control layer between an application and the AI services it calls is the technical service we designed as an "AI strainer" and we roll it out in Q4 2026.