Company Overview
Acme Bank is a mid-size European financial institution headquartered in Frankfurt, serving 2.3 million retail and commercial customers across the EU. Like most banks its size, Acme had deployed AI aggressively over the prior 18 months — automating credit scoring, KYC document verification, transaction fraud detection, and customer service triage.
By early 2026, Acme was running 47 LangChain agents in production, collectively processing over 200,000 actions per day. The agents had passed internal model risk management review. What they hadn’t done was comply with the EU AI Act.
The Compliance Challenge
With the EU AI Act’s high-risk AI provisions taking full effect in August 2026, Acme’s compliance team faced a stark reality: they had no infrastructure to meet the Act’s requirements for their agentic AI systems.
Specifically, they lacked:
- Automated logging — Agent actions were not systematically logged at the granularity required by Article 13
- Policy-level evaluation — No system existed to map agent actions to specific EU AI Act provisions
- Audit evidence — Producing a conformity assessment meant manually reconstructing agent behavior from scattered system logs — a 120-hour quarterly effort
- Drift monitoring — No mechanism existed to detect if agents changed behavior after deployment
The compliance team had estimated a traditional approach — manual instrumentation, custom logging pipelines, and quarterly audit engagements — would take 8–12 months and cost over €2 million in engineering and consulting fees.
The August 2026 deadline made that timeline impossible.
The AgentGovern Solution
Acme’s CTO connected with the Zirahn team in January 2026. Following a two-week proof-of-concept, they chose AgentGovern’s Professional plan and began full deployment.
SDK Integration
A single Acme engineer integrated the AgentGovern SDK across all 47 LangChain agents in four days:
from agentgovern import AgentGovernSDK
sdk = AgentGovernSDK(api_key=os.environ["AGENTGOVERN_KEY"])
# Applied to each of the 47 agents
governed_executor = sdk.wrap_langchain(
executor,
policy="eu-ai-act",
agent_id=f"credit-scoring-{region}",
environment="production"
)No changes to business logic. No new infrastructure. The same agents, now fully instrumented.
Policy Pack Deployment
Acme enabled the EU AI Act policy pack and configured it for their specific high-risk AI classifications. Within 24 hours of activation, the compliance dashboard was showing:
- Real-time compliance scores per agent
- Article-level policy mapping (Articles 9, 10, 13, 14, 17)
- Automated flagging of actions requiring human oversight
Dashboard and Drift Monitoring
Acme’s compliance officers got access to the dashboard in week two. They configured drift alerts with thresholds appropriate for their credit scoring agents, where even small behavioral shifts can indicate model degradation or data distribution changes.
Results
Quantitative
- Audit preparation time: Reduced from 120 hours to 8 hours per quarter (93% reduction)
- EU AI Act conformity evidence: 100% automated — every required artifact generated in real time
- Drift incidents detected: 3 incidents identified in the first 30 days, all before production impact
- Time to first regulator-ready report: 18 days from integration
- Engineering cost: 4 days of one engineer’s time vs. 8-12 months estimated for custom build
Qualitative
Acme’s compliance team describes the change as “fundamental.” For the first time, they have a continuous view into how their AI agents are behaving — not a quarterly snapshot. The drift detection capability, in particular, caught an issue with one of their KYC agents that would have been a significant regulatory incident if left undetected.
The August 2026 deadline, which had looked impossible in January, now looks achievable with time to spare.
What’s Next
Acme is expanding AgentGovern coverage to their NIST AI RMF compliance obligations and has engaged Zirahn’s team to build custom policy rules for their OCC model risk management requirements.
They’re also evaluating how to use the audit evidence AgentGovern generates to streamline their annual external model audit — reducing what has historically been a multi-week engagement with external auditors.