The Office of the Superintendent of Financial Institutions (OSFI) has set forth guidelines to ensure that Federally Regulated Financial Institutions (FIs) maintain robust data security, risk management, and operational resilience. Private AI’s advanced machine learning and natural language processing technologies provide helpful tools that can help FRFIs meet these requirements efficiently and effectively.
Since the OSFI guidelines are often high level, this article provides practical insights into how Private AI can support compliance with key OSFI guidelines, particularly B-13 Technology and Cyber Risk Management, the Integrity and Security Guideline, and the Draft Guideline E-23 Model Risk Management.
1. Data Identification, Classification, and Protection (B-13: 3.1.4 and Integrity and Security Guideline: Principle 8)
One aspect of OSFI’s B-13 Technology and Cyber Risk Management guideline mandates that FIs implement controls to identify, classify, and protect both structured and unstructured data. This process must include periodic discovery scans to detect any deviations from established standards. Classification in the cybersecurity context typically refers to “Public, Private, or Confidential” data.
Private AI’s technology excels in the identification and labelling of sensitive data across both structured and unstructured formats. Knowing which kinds of data points are and where can help with the classification of the data in accordance with internal standards.
By leveraging machine learning models trained to identify over 50 different types of personal data across 53 languages, Private AI can accurately and efficiently detect and classify sensitive information within an organization’s data repositories. Automating this basic yet critical step ensures that the very cornerstone of data management, i.e., knowing what you have and where, is done reliably and without sinking in a lot of time and effort.
For example, Private AI’s solution can detect Payment Card Industry information like bank account number, CVV, and credit card expiration even in unstructured data and in hidden places. Doing so is no easy feat in banking, where data can live within free text fields in databases, embedded files, storage buckets, and network drives with handwritten content on PDF scans, Word documents and images. The volumes, variety, and complexity of the data involved mean FIs are dealing with messy, disfluent data, OCR errors, and multilingual and multinational information.
Private AI’s flexible deployment allows developers to integrate it with continuous monitoring capabilities to allow for periodic discovery scans to automatically identify changes or deviations from the established data protection controls. This proactive approach strengthens data security but also ensures compliance with the ongoing identification and classification requirements set forth by OSFI.
2. Secure-by-Design Practices (B-13: 3.2.1)
In the same guideline, OSFI emphasizes the adoption of secure-by-design practices to safeguard technology assets, including applications, microservices, and APIs. Security controls should be preventive and applied end-to-end, starting at the design stage.
Private AI’s solutions are designed to integrate seamlessly into the development lifecycle of FIs, allowing developers to embed robust data protection measures directly into applications and services from the outset. By implementing Private AI’s API during the design and development stages, FIs can ensure that sensitive data is identified, classified, and protected even before it is ingested into the FI’s systems. This preventive approach aligns well with OSFI’s secure-by-design mandate, ensuring that data protection is an inherent part of the technology stack, rather than an afterthought.
For example, Private AI’s solution could be used to intercept incoming data streams to filter out specific personal information that the FI does not need to collect for its purposes, like the SIN number on a credit report. Unless the SIN is collected for purposes required under the law, the Office of the Privacy Commissioner advises not to collect it, and ties stringent information and consent requirements to its collection.
3. Data Protection and Loss Prevention (B-13: 3.2.5 and Integrity and Security Guideline: Principle 8)
The protection of data throughout its lifecycle—whether at rest, in transit, or in use—is a cornerstone of OSFI’s B-13 and Integrity and Security guidelines. These guidelines require FIs to design and implement risk-based controls, starting with clear information classification.
Private AI’s machine learning technology offers unparalleled accuracy in detecting and redacting sensitive information across all stages of the data lifecycle. Whether data is being stored, transferred, or actively used in processing, Private AI can automatically identify and mask sensitive information to prevent unauthorized access or leakage. This is particularly critical for data in transit, where the risk of interception is high. By integrating Private AI’s tools into existing DLP strategies, FIs can ensure that their data is continuously protected, meeting the stringent requirements of OSFI’s data protection guidelines.
4. Forensic Investigations and Root Cause Analysis (B-13: 3.4.5)
When incidents occur, OSFI requires FIs to conduct forensic investigations and root cause analyses to determine the impact and address the vulnerabilities that led to the breach.
Private AI can play a crucial role in these investigations by quickly identifying the types and extent of sensitive data involved in an incident. This capability enables FIs to assess the direct and indirect impacts more accurately and to implement targeted remediation actions. Additionally, by providing detailed reports on the data affected, Private AI assists in understanding the root causes related to data handling practices, ensuring that lessons learned are effectively integrated into future security measures.
5. Data Governance and Model Risk Management (Draft Guideline E-23: Principle 6)
OSFI’s draft Guideline E-23 highlights the interdependency between data and model risk, particularly in the context of AI/ML techniques. Organizations are required to establish policies and procedures that govern data in models, ensuring consistency with enterprise-level data governance frameworks.
Private AI’s technology supports this requirement by offering tools that enhance the privacy and fairness of data used in AI models. By detecting and redacting sensitive information before it is fed into models, Private AI helps organizations maintain the privacy of their data, reducing the risk of biased outcomes and privacy breaches. With regards to biases, the approach is this: filtering out personal information that constitutes one of the so-called “protected grounds” on the basis of which discrimination is prohibited, such as age, gender, and sexual orientation, improves the fairness of a decision-making system because it can no longer directly base its decision on such prohibited grounds. Beware, however, that AI models may be able to infer prohibited grounds from other data points that remain. Thus aligning the data composition with enterprise-level data governance ensures that AI/ML models are not only compliant with OSFI guidelines but also ethical and effective.
Conclusion
Private AI provides some of the advanced tools and technologies that FIs need to comply with OSFI’s rigorous guidelines. From data identification and classification to secure-by-design practices and robust data protection measures, Private AI helps financial institutions safeguard their data, mitigate risks, and ensure compliance across all stages of the data lifecycle. By integrating Private AI’s solutions into their technology infrastructure, FIs can meet aspects of OSFI’s standards with confidence, knowing that their data security and governance practices are both effective and future-proof.
