European Union AI Act (August 2026): Sequence Corporate Pilots by 2 Filters

This guide sequences European Union AI Act corporate pilots for August 2026 through two mandatory filters. It produces a documented pass, hold, or stop decision for every candidate.

Apply two gates in order

Before any candidate enters a live test, apply the two gates in order: verify its legal pathway, then check whether its experiment is ready. Record a pass, hold, or stop decision for each candidate, and sequence only those that pass both gates.

For Filter 1, document the system’s intended purpose, your organization’s role, data flows, users, and planned test setting. Ask a responsible legal reviewer to verify the applicable pathway for those facts and the planned date. If it remains unclear, hold the pilot until it is verified.

Filter 2 demands a complete experiment design. Require a named hypothesis, a defined baseline, a measurable outcome, a bounded test population, a named owner, and a stop condition. If any field is blank, the pilot is not ready to run. This rigor ensures the test yields learning rather than merely exposing users to risk.

Order only pilots that pass both filters. Hold a pilot that clears legal verification but lacks a measurable learning objective, and also hold one whose legal pathway remains unverified. Record a pass, hold, or stop decision for every candidate.

Use the legal-verification gate before the experiment-readiness gate so that technical readiness does not substitute for a verified legal pathway. Treat reduced rework as a possible benefit to assess, not as a guaranteed outcome.

Build an evidence file before ranking

The supplied grounding contains no EU AI Act source or legal evidence, so it cannot substantiate a classification, obligation, deadline, or the August 2026 change for any pilot. Do not treat a search-result summary as evidence for a pilot decision; treat it as a starting pointer only, and replace it with the official text before ranking candidates.

For each pilot, have counsel retrieve the official EU AI Act text on EUR-Lex and match it against the documented intended purpose and role. Save the exact EUR-Lex text location (title, article, and paragraph) and the reviewer’s conclusion in the evidence file. This creates an auditable trail that survives staff turnover and regulator scrutiny.

Cross-check the EUR-Lex interpretation against the current European Commission AI Act guidance and any relevant implementing measures. Where the Commission guidance diverges from a literal reading of the Act, record both positions and flag the conflict for legal sign-off rather than resolving it in-house.

Apply the legal-verification gate first and the experiment-readiness gate second. Sequence only pilots that pass both, and hold any pilot whose applicable legal pathway has not been verified for its intended use. A pilot that clears legal verification but lacks a measurable learning objective stays on hold until the experiment design is complete.

Before ranking, confirm that every candidate carries a documented pass, hold, or stop decision tied to these two gates. Pilots missing either the legal verification record or the learning-objective threshold do not advance to sequencing, regardless of business priority or resource availability.

Compare sequencing choices

After both gates are passed, the portfolio may rank eligible pilots by learning value and reversibility. Keep any pilot with an unverified legal pathway on hold until its pathway is confirmed.

The parallel-run approach looks efficient but fails the two-filter rule. A pilot with an unverified legal pathway consumes test capacity, staff time, and documentation effort before any learning occurs. If that pathway later proves invalid, the entire test must stop, and the portfolio loses both the sunk capacity and the delayed learning. Speed without verification is a false economy under the EU AI Act timeline.

The winning approach applies both filters before ranking. First, confirm each pilot's legal pathway under the applicable EU AI Act classification, obligation, and deadline. Second, confirm the experiment has a measurable learning objective tied to that classification. Only pilots passing both filters enter the ranking step, where learning value and reversibility determine order. This keeps unverified pilots on hold without consuming live-test capacity.

Reversibility acts as a tiebreaker when learning value is similar. A pilot whose live test can be rolled back or paused without regulatory penalty ranks ahead of one whose deployment creates irreversible exposure. This check protects the portfolio from a single high-risk pilot blocking the entire sequence. Apply it after both filters pass, not before.

The hold rule is absolute: any pilot whose applicable legal pathway has not been verified stays on hold. Do not sequence it, do not allocate live-test resources, and do not count it toward portfolio learning until verification is complete. This rule alone prevents the portfolio from drifting into non-compliance before August 2026.

Price the decision with an example

Assess each pilot against two gates: documented legal verification first, then experiment readiness, including a measurable learning objective. If the team uses a learning-value score, define its rubric and any threshold in advance; a score cannot override an unverified legal pathway. Sequence only pilots that pass both gates.

This section alone uses a fictional three-pilot portfolio and its illustrative euro assumptions to show how the two-filter sequence produces a documented pass, hold, or stop decision. Pilot A, internal document search, has a counsel-recorded legal pass and a test-ready score of 8/10. Pilot B, invoice exception triage, has a legal hold and a score of 9/10. Pilot C, candidate ranking, has a legal pass and a score of 6/10. The score reflects the team’s own learning-value rubric, not an AI Act rating.

The sequencing rule is mechanical: A passes both filters, so it can be scheduled first; B’s 9/10 score cannot override its legal hold, so verify the legal pathway before any live test; C passes both filters but with a lower score, so it is scheduled after A. This portfolio sequencing approach wins under the two-filter rule because it prevents a high-scoring pilot from advancing while its legal status remains unresolved.

Use this hold-or-sequence worksheet for every candidate before ranking: (1) confirm counsel has recorded a legal pass for the applicable pathway; (2) confirm the experiment score is at least 6/10; (3) if both are met, assign a sequence slot; (4) if legal is on hold, move the pilot to a verification queue and block live-test entry; (5) if the score is below 6/10, hold for redesign. The final decision rules are pass, hold, or stop, and each must be documented in the evidence file before any pilot enters a live test environment.

Use this hold-or-sequence worksheet

Use this copy-ready portfolio worksheet, one row per candidate pilot, to record the evidence and produce a documented pass, hold, or stop decision. Treat an incomplete field as an unresolved issue rather than as an implied approval. The worksheet should contain: Name; Intended purpose and users; Organizational role; Planned test setting/date; Legal source and reviewer/date; Legal gate—pass or hold; Hypothesis; Baseline; Outcome metric; Stop condition; Experiment gate—pass or hold; Learning-value score; Reversibility evidence; and Decision and owner.

Make the status rules operational. A pass requires a documented legal verification, a measurable experiment objective, a baseline, an outcome metric, and a defined stop condition. A hold applies when required verification or experiment evidence is missing, inconsistent, or awaiting confirmation. A stop applies when the evidence shows that the proposed test should not proceed or when its stop condition is triggered. Record the reason and owner in the same row so the portfolio does not depend on informal recollection.

Reopen the worksheet whenever the intended purpose, organizational role, or test setting changes: legal verification must be renewed before the pilot continues. If the test population, data, or model changes materially, recheck both the legal pathway and the experiment design, then update the relevant evidence, status, decision date, and owner. This prevents a prior approval from silently carrying forward into a materially different test.

What to do next

Frequently Asked Questions

Who must verify a pilot’s applicable legal pathway before it can enter a live test?

A responsible legal reviewer must verify the applicable pathway.

What should happen if a pilot’s applicable legal pathway remains unclear?

Hold the pilot until the pathway is verified.

Which pilot facts should be documented for legal verification?

Document the system’s intended purpose, the organization’s role, data flows, users, and planned test setting.

When must a pilot’s planned date be considered during legal verification?

The responsible legal reviewer must verify the applicable pathway for the documented facts and planned date.

What must a pilot have before the experiment-readiness gate is applied?

It must have a measurable learning objective.

What decision must be recorded for every pilot candidate?

Record a documented pass, hold, or stop decision, and sequence the pilot only if it passes both gates.

Quick answers

Also worth reading: Run 15 Corporate Pilots a Quarter: 3 Gates, 70% Fail: Run 15 Corporate Pilots a · Kill, Extend, or Scale: 2026 Cost-per-Learn Benchmarks for Pilots: Kill, Extend, or Scale: 2026 · Guardrail Tiers: 12 Questions, 36 Points, 15 Pilots a Year: Guardrail Tiers: 12 Questions, 36