Implementing AI, though a promising venture, is hardly a simple initiative. From incorrect responses and hallucinations to a lack of transparency, security vulnerabilities, and data privacy concerns, enterprises seeking to adopt AI must contend with a litany of obstacles.
Experts joined DBTA’s latest webinar, Mitigating AI Risks: Governance, Security and Compliance Strategies, to examine how the right mix of governance, security, and compliance technologies and best practices can ensure AI success.
Suzanne Weller, director, research and development, privacy and security, Informatica, asserted that “effective governance is vital to ensure that AI projects deliver valuable outcomes.” This is increasingly critical as “the adoption rate of generative AI is accelerating, and organizations are making substantial investments in AI projects in order to secure a competitive advantage,” Weller said.
However, these enterprises are facing significant challenges as they transition AI pilot projects into full-scale systems that deliver tangible value. AI governance is key toward breaking down these obstacles, yet it is also a complex process, Weller explained.
Informatica makes AI governance simple, enabling organizations to build confidence in their AI initiatives with trust to accelerate ROI. This is achieved through four pillars:
- Inventory: Catalog structured, unstructured data and AI assets with AI-assisted classification that drives the discovery of AI systems from major ecosystems.
- Control: Implement workflows for compliance and risk management paired with access management for unstructured data.
- Deliver: Ensure content quality for unstructured data and provide access management in AI pipelines.
- Observe: Deliver observability over AI performance across major ecosystems and establish an audit trail of risks and controls throughout the AI lifecycle.
“AI in the enterprise hits on a lot of the same challenges that we’ve dealt with for quite a while,” said Steve Karam, principal product manager, AI, SaaS, and growth, Delphix by Perforce . “AIOps is DevOps in a new coat.”
Karam focused on the challenge of data privacy as it relates to AI, where a changing regulatory landscape—paired with evolving threats—makes it difficult for enterprises to keep up.
Delphix by Perforce acts as the “compliant data layer,” ensuring that data—in every place it lives—is secured and compiled without losing the data’s statistical and analytical value. Data is masked consistently across systems without sacrificing its utility, persisting security against data theft and leakage. Delphix by Perforce helps deliver future-proofed AI success, which is achieved through:
- Enhanced model training with realistic, masked data
- Improved data compliance with irreversible, integrated masking
- Multi-cloud integrity with a consistent, multi-cloud platform
- DevOps for data-enabled AI apps
Susan Laine, chief technologist, data thought leader, erwin by Quest, echoed Karam’s sentiment, explaining that, as far as AI goes, “we’re still struggling with the basics.” Failing at perfecting these basics—data availability and data quality—prevents enterprises’ ability to trust their AI projects—“if we don't trust the data, how will we trust AI?” Laine pointed out.
A key challenge, according to Laine, is that organizations are taking on too much in regards to data and AI governance. Implementing a “govern as you go” approach is a more feasible, more effective method of governance, ensuring that use cases are met at each step of the governance process to drive alignment with business goals.
To help determine the efficacy of governance strategies, erwin by Quest offers a certification process that ensures that AI models and its data are fit for use. This allows enterprises to move their models to production with confidence, assured that their projects are deemed ready for scale. erwin by Quest then observes the model’s outputs, alerting the organization to model drift so that they may react, tune, and remediate any new challenges that arise.
Dr. Kjell Carlsson, head of AI strategy, Domino Data Lab, argued that effective AI governance has been solved. The challenge, however, is making it work efficiently, flexibly, and at scale.
This is further compounded by the fact that with more AI, there’s more risk, and its constant evolution makes it even more complex to govern. According to a survey from BARC, 95% of enterprises must replace, re-write, or update their AI governance frameworks and processes for today’s GenAI-enabled model landscape.
The solution, then, is evolving existing AI governance capabilities to adapt to the increasingly irregular, dynamic landscape of AI, according to Carlsson. From people to processes and technology, enterprises must re-evaluate every aspect of their governance strategies according to the following building blocks:
- Visibility
- Auditability
- Reproducibility
- Control
- Automation
“If you don’t have these building blocks, you might as well go home,” joked Carlsson.
Even with these building blocks, enterprises may still fail to keep up. That’s why, with AI platforms such as Domino Governance, organizations can transform governance from complexity to momentum. Domino Governance automates and orchestrates the collection, review, and tracing of materials required to ensure compliance with internal and external policies, Carlsson explained.
This is only a snippet of the Mitigating AI Risks: Governance, Security and Compliance Strategies webinar. To view the full webinar, featuring detailed explanations, a roundtable discussion, and more, you can view an archived version here.