The instrument
The assessment instrument is publicly available, allowing you to read it, use it, and assess yourself against its criteria without engaging us directly.
ACES stands for the AI Control Efficacy Standard. It comprises two instruments: one focused on internal workforce AI governance, and the other on external, customer-facing AI release assurance. Both instruments assess the same ten domains using a five-level maturity scale.
Instrument A:
Instrument B:
Both instruments score the same ten domains, so profiles compare across teams and products. The diagnostic question behind each one changes with the exposure. Data boundaries asks about egress to external model providers under governance assurance, and about customer PII and prompt injection under release assurance.
Where regulators look first. The accountability chain (who owns the system, what the AI decided, who approved it, and whether that can be reconstructed) and transparency (whether people are told AI is involved). A gap in any of these is reported as a priority.
01: Decision rights and constitution (Regulatory Focus)
Who the AI answers to, and what it may decide alone, enforced in code rather than written down.
02: Data boundaries
What confidential, customer or personal data the AI can reach, and what stops it leaking out.
03: Tool boundaries
What systems and actions the AI can trigger, restricted to the minimum it needs.
04: Standard procedures
Whether the prompts and workflows people rely on are version-controlled and tested, or reinvented each time.
05: Output verification
What catches a wrong, fabricated or malformed answer before anyone acts on it.
06: Operational resilience
Whether a runaway or misbehaving AI can be stopped fast, and whether harm gets reported and followed up.
07: Economic control
Whether spend is capped by a technical limit before it runs away, not discovered on an invoice.
08: Human oversight
Whether the people meant to catch AI mistakes have the time, information and authority to do it.
09: Auditability, traceability and defensibility (Regulatory Focus)
Whether you can reconstruct exactly what the AI did and who approved it, months later.
10: Transparency and Disclosure (Regulatory Focus)
Whether the people affected are told AI was involved, applied automatically rather than remembered.
Each domain is scored on the same five-level scale:
Most organisations assess themselves at level 2 and turn out to be at level 1. The distinction is whether anything actually stops the rule being broken.
Level 3 is the reason independent assessment exists. You cannot credibly verify your own controls.This is an assurance scale. Each level asks whether a control blocks what it claims to block, and whether anyone independent has checked.
Worked example
An anonymised profile from a financial services organisation, assessed across both instruments. A domain at 3 or 4 is one where you can let people move quickly, because something is genuinely catching mistakes. A domain at 0 or 1 is where the risk is real and the brake is missing.
Knowing which is which is what lets you stop applying the same brake everywhere.
The self-assessment version of the scorecard is free to use, and most organisations find it worth the hour. It takes you as far as level 2. Level 3 needs a pair of eyes from outside the team that built the control, which is what an independent assessment is for.
