Case Study Download (PDF)
The Snapshot
1
A full LLM security assessment of a customer-facing financial literacy chatbot, completed before go-live.
2
A combined AI + Cloud + Web engagement uncovered a cloud token disclosure risking full environment compromise.
3
All critical findings remediated pre-production, zero exposure to live customers or regulatory consequences.
The Client
A global financial institution serving 3.6 million customers across 500+ domestic branches and 23 countries. The institution deployed an AI-powered financial literacy chatbot, embedded into a customer-facing website, and commissioned a full LLM security assessment before go-live.
The Challenge
-
A new class of risk in regulated environments. Deploying an LLM-powered chatbot in a financial institution creates attack surfaces that existing security frameworks were not built to evaluate and that regulators are increasingly scrutinising.
-
Public-facing deployment amplifies exposure. The chatbot is designed to be embeddable across any website, including sites outside the institution's control. It can be loaded and exploited from arbitrary third-party contexts making abuse prevention, origin validation and output integrity critical requirements.
-
AI risk sits on top of existing infrastructure risk. LLMs do not replace legacy vulnerabilities, they add to them. An institution serving 3.6 million customers across 500+ branches and 23 countries cannot afford blind spots at any layer of its stack.
The Solution
AI Pentesting methodology – A robust, structured approach grounded in MITRE ATLAS, OWASP LLM Top 10 and OWASP Agentic Top 10 ensuring systematic coverage of every known AI attack category.
AI + Cloud + Web Application, a combined engagement – The deployment integrated the corporate website and cloud backend. Cloud and Web Application Pentesting, applied alongside the AI methodology, was crucial: cloud analysis revealed a token disclosure that exposed credentials, risking full cloud environment compromise.
Enabling trusted digital innovation – By securing the AI layer, the cloud infrastructure and the Web Application before go-live, the engagement enables the institution to extend financial literacy services to millions protecting end-users, cloud assets and regulatory standing simultaneously.
MITRE ATLAS
OWASP LLM Top 10
Cloud Pentesting
Key Findings
Critical issues surfaced across every layer of the stack, all identified and remediated before public launch.
LLM
Unbounded Consumption
No rate limiting on token usage or input length, an attacker could degrade availability across all embedded instances.
Impact: Critical risk in a public-facing, embeddable deployment.
Web App
Chatbot Hijacking
The embeddable chatbot could be loaded and exploited from arbitrary third-party websites outside the institution's control.
Impact: An attacker could embed the chatbot in other websites with financial impact to the client.
Cloud
Cloud Token Disclosure
Cloud credentials exposed through the website–chatbot integration layer, with potential full cloud environment compromise.
Discovered via: Devoteam Cyber Trust Cloud Pentesting applied to the integration layer.
The Impact
Through this engagement, Devoteam Cyber Trust enabled the client to:
- Discover and remediate all critical findings pre-production, before public launch.
- Close a cloud token disclosure that risked full cloud environment compromise.
- Reach go-live with zero exposure to live customers or regulatory consequences.
- Extend financial literacy services to millions while protecting users, cloud assets and regulatory standing.
Back to Case Studies
Your Security
Starts Here
Partner with Devoteam Cyber Trust to gain expert-led cybersecurity assessments and a culture of continuous risk improvement.
→ Get in touch
