Reliable AI at scale requires trust engineering, not legacy QE
AI is in production faster than enterprises can prove it’s safe, fair or accountable. HFS Research and QualiZeal asked 101 enterprise leaders where the gaps are.
HFS Take 5 Report
Five findings at a glance
AI that cannot be audited, explained, or proven fair is an open liability, and it widens with every sprint.
101 Leaders Across Us Enterprises
Each square is one respondent. Switch the view to regroup them.
AI is in production. The controls are still in pilots.
57% run GenAI in production, but only 31% have a formal validation framework there.
n = 101 enterprise quality and technology leaders. Source: HFS Research in partnership with QualiZeal.
What the data shows
- 93% run GenAI in some form; 20% already at scale.
- 28% haven’t started a validation framework; 42% are piloting.
- 31% haven’t started a dedicated AI quality team.
37% run AI with no one holding the line
Ownership is spread across teams or missing entirely. QE owns AI quality in only 15%.
n = 101 enterprise quality and technology leaders. Source: HFS Research in partnership with QualiZeal.
What the data shows
- 22% let the team that built the model validate it.
- 12% hand it to risk and compliance, who can’t run the tests.
- 15% give it to QE, the function built to test.
Most see AI testing as different. Only 15% have built for it.
Probabilistic AI is still tested with tools made for deterministic software.
Current approach to testing AI systems
Testing methods in use today
n = 101 enterprise quality and technology leaders. Source: HFS Research in partnership with QualiZeal.
What the data shows
- 52% use frameworks built for deterministic software.
- 73% still rely on manual review of samples.
- 34% see the problem but haven’t acted on it.
Get all 16 pages of analysisIncluding HFS Research’s three recommendations
Download the report88% measure accuracy. 35% can evidence compliance.
Enterprises can show their AI is accurate. Far fewer can show it’s fair, explainable or compliant.
n = 101 enterprise quality and technology leaders. Source: HFS Research in partnership with QualiZeal.
What the data shows
- An 88%-accurate model can still fail a protected group.
- 63% can’t explain an AI decision to a regulator.
- Only 35% can produce compliance evidence today.
QE got the mandate. Only 24% got the budget.
QE took on AI validation, data quality and security. Budgets mostly didn’t follow.
QE’s added responsibilities, last 18–24 months
Did headcount and resources keep pace?
n = 101 enterprise quality and technology leaders. Source: HFS Research in partnership with QualiZeal.
What the data shows
- 65% of QE teams now own AI model validation.
- 48% got some extra resource, but not enough.
- For 6%, budget or headcount shrank as scope grew.
Enterprise AI has already crossed the production threshold. With GenAI systems live in more than half of organizations, and agentic systems with real autonomy deployed in over a third of surveyed companies, quality and governance remain largely stuck in pilot mode.
Get the full report
Every chart, the full demographics and HFS Research’s recommendations in one 12-page PDF.
Download the report
The people behind the data
HFS Research
Global research and advisory firm guiding Fortune 500 companies through IT and business transformation.
QualiZeal and ValidAIte™
AI-native quality engineering and AI assurance partner. ValidAIte™ evaluates AI for trust, safety and compliance readiness.
- 70+ enterprise clients
- 1,000+ people
- NIST AI RMF · EU AI Act
About the author
Leads HFS Research’s banking, financial services and IT services practice, with 15+ years in enterprise technology.
