Multi-agent AI: Advanced Control Examination & Test
2026-01-29
This release introduces an advanced multi-agent AI Control Examination & Test evaluation engine designed to handle the full complexity of GRC control testing, going beyond isolated evidence checks to reason across datasets, documents, tables, and images as a unified whole.
The new Control Examine & Test orchestrates specialized AI agents in a succinct context to correlate evidence from multiple sources, generate intermediate merged datasets (cross-referenced correlations), and generate code to evaluate structured data against control test criteria (test tables). By combining the multi-agent AI evaluation engine and Trustero Trust Graph with newer foundation models, the new Control Examination & Test delivers substantially higher accuracy, approximately 97% in internal benchmarks*, and substantially higher consistency over multiple iterations while improving explainability and coverage.
The Trustero AI Control Examine & Test raises the bar on repeatable and accurate control tests at substantially lower cost compared to other control test solutions.
*Based on expected outcomes from 1,200 AI control test evaluations spanning 70 unique controls.
Enabling Advanced Examination and Test
A new toggle on the control management page allows you to enable Advanced Examination and Test for individual controls.
This mode is disabled by default to preserve backward compatibility with existing control evaluations. Advanced examination performs a more comprehensive analysis, and evaluations may take longer to complete. If a control is already producing the desired results using the standard evaluation, you may choose to keep Advanced Examination and Test disabled.
Use Advanced Examination and Test
Improved Explainability and Traceability
Trustero AI now provides significantly greater transparency into how control test conclusions are reached. Each evaluation begins with an overall summary that explains the control outcome, identified gaps, and key recommendations, then dives deeper into detailed explanations of the reasoning and supporting source citations.
The system applies stricter evaluation logic by design, ensuring that control results reflect a comprehensive review of all relevant evidence and how those artifacts relate to each other.
This layered approach enables teams, auditors, and reviewers to clearly see what Trustero AI examined and how it arrived at its conclusions, improving trust, audit readiness, and confidence in the results.
Gaps and Corrective Action Recommendations
As controls are evaluated, Trustero AI proactively identifies gaps and classifies issues such as missing, incomplete, or outdated evidence. For each identified gap, the system provides targeted corrective action recommendations to help teams address deficiencies efficiently.
A new summary section at the top of the Control Examine & Test results highlights these gaps and recommendations, allowing reviewers to quickly understand risk areas and required next steps.
Gaps and Recommendations
Detailed Test Results and Citations
Trustero AI presents detailed analysis results for each evidence artifact reviewed, showing what was evaluated, how it was interpreted, and how it contributed to the final control determination. Findings are directly linked to supporting documents, tabular data, images, and any intermediate artifacts generated during analysis, allowing reviewers to trace conclusions back to their source.
Detailed Test Results
Evidence Cross-reference and Test Tables
Trustero AI can now correlate large tabular evidence sources by joining data across common fields to produce a single, unified piece of evidence, removing the need to manipulate raw evidence prior to examination. This merged dataset is then evaluated as a test table, applying row-level validation as well as overall test criteria to determine whether the combined evidence satisfies the control’s test procedure.
Below is an example control test procedure that results in cross-referenced evidence evaluated using a test table:
Identify Terminated Employees with Active Access: Compare the current user lists (specifically for AWS, Google Workspace) against the active employee roster from the HR system (TriNet) to identify accounts that belong to terminated employees. Match usernames or employee identifiers between each user list and the HR TriNet FTE employees list. Flag any users that are listed but do not appear in the HR active list. For the control test to be successful, 100% of the users in the systems must appear on the HR active list.
Note the phrase in the control test:
“Match usernames or employee identifiers between each user list and the HR TriNet FTE employees list.”
This indicates that evidence should be cross-referenced using these data sources. You should see both the cross-referenced results and the final test table analysis cited as sources for the evaluation.
Citations of Test Result Utilizing Correlated Tables
When a correlation is performed, the resulting merged table is available as a cited reference in the control test results. A detailed explanation of how the tables were joined is included in the citation, along with references to the original sources of the correlation.
Cross-reference Table Between Data Sources
Also note the phrase in the control test:
“Flag any users that are listed but do not appear in the HR active list. For the control test to be successful, 100% of the users in the systems must appear on the HR active list.”
This indicates that Trustero AI should generate a “test table” to analyze the resulting data, including the conditions applied to each row of the table and the overall success criteria that determine the control test results.
Test Table Evaluation of Joined Data Set
Other Notable Improvements in Advanced Examination & Test
- Dynamic model selection across the multi-agent workflow allows Trustero AI to choose the most appropriate LLM for each evaluation step. This has significantly improved analysis of complex, image-based evidence.
- More holistic evidence selection and coverage enables Advanced Examination & Test to better determine which evidence is relevant to a control test procedure, improving both evidence selection and detection of missing evidence.
- More consistent date-based evaluations improve detection of outdated evidence and interpretation of review and approval dates, including dates discovered within documents and images.
Try it Today
The new Trustero Advanced Examination and Test is available today. Log into your Trustero account to explore the capabilities of multi-agent AI for GRC.