Rehan Jalil is CEO of cybersecurity and data protection infrastructure firm SECURITI and ex-head of Symantec’s cloud security division.
Generative AI, particularly in the form of sophisticated language models, has undoubtedly revolutionized many aspects of our lives. However, its rise has also brought pressing privacy and governance risks that demand our attention: What really happens when tools like Google’s Vertex or Open AI’s GPT 4 are misused?
With the exponential growth of generative AI tools for the enterprise, leaders are realizing that, unfortunately, there is a darker side to generative AI.
While the hype around AI language models is real, organizations need safeguards when it comes to data fed to those same models. The reality is that anything that goes into the learning process can never be taken back, which risks exposing sensitive and personal information forever. Mixing data in these models can also break transparency and regulatory controls.
Generative AI Concerns
Generative AI’s rapid rise exemplifies the ongoing challenge data leaders encounter in striking a balance between fostering data-driven innovation and fulfilling their organizational obligations. These technologies offer a wealth of opportunities to enhance operations in different industries. However, the use and deployment of large language models (LLMs) bring associated risks and concerns that need careful handling.
In fact, as enterprises leverage AI more broadly within their processes and infrastructure, they need to pay close attention to:
• Data Leakage: Large datasets containing sensitive information might be used for training models without adequate security measures. Data ranging from private messages to financial records to personal identifiable information (PII) can be shared when security, access controls and protocols are insufficient.
• Data Re-Identification: Generative AI models’ ability to recognize correlations, identifiers and patterns raises the risk of re-identification. Even when masking certain data fed to the algorithms, they can still link seemingly anonymous data back to individuals.
• One-Way Flow Of Information: Generative models’ unidirectional information flow can obscure output generation. After training, these models don’t reveal how they are producing responses to queries, creating a lack of transparency and making data accountability even more difficult, particularly when teams need to address regulatory compliance and maintain certain data standards within highly regulated fields.
• Liabilities Across Various Domains: From intellectual property to legal compliance to data ethics, the challenges stemming from complex architectures and transparency gaps make it even more difficult to rely fully on the outputs from generative AI, not to mention how much harder it becomes to adhere to a wide range of data regulations.
Security concerns arise in practical applications, highlighting the need for data security, regular audits and secure deployment. These difficulties highlight how crucial it is to prioritize ethical considerations, which include fairness, openness, responsibility and compliance. This all-encompassing strategy seeks to successfully reduce potential risks while encouraging moral and compliant conduct in the creation and application of generative AI technologies.
How To Enable The Safe Use Of Generative AI
Chief data officers (CDOs), chief information security officers (CISOs) and leaders in data management grapple with the task of providing benefits to the business while navigating the fine equilibrium between data-hungry teams and data responsibilities.
Striking a chord between swift, precise analytics and safeguarding comprehensive data integrity across divisions is their imperative. In light of data landscape obligations and technical advancement, organizational focus should be on methods that enable the secure application of generative AI.
• AI Model Safety: This entails constant risk assessments, careful model discovery and preventative steps to fend off adversarial attacks and data poisoning. Organizations can improve the security of their generative AI systems and their outputs by implementing these practices.
• Enterprise Data Usage: This involves a comprehensive understanding of the data types that are being used, enabling risk assessments and privacy considerations. Controlling access entitlements to this data is crucial as well, as it ensures that only authorized users can interact with and influence AI models. This multi-layered strategy ensures data protection and compliance while enabling safe use.
• Prompt Safety: This requires taking preventative steps to thwart malicious prompts that could cause an AI model to produce offensive or hazardous information. The proactive detection and mitigation of attempts to extract biased or sensitive information from the models is equally important. Organizations can ensure that the outputs adhere to ethical standards and avoid any abuse or unforeseen repercussions by developing strong mechanisms for fast formulation and vetting.
• AI Regulations: Organizations must proactively engage with a variety of regulations that govern the use of AI technologies as the regulatory landscape surrounding AI continues to change. This entails keeping up with laws governing data protection, algorithmic transparency and moral AI standards. Organizations can promote a safer and more responsible AI ecosystem by embracing these growing rules and making sure that their use of generative AI adheres to moral and legal standards.
Generative AI has ignited excitement across industries, offering to automate tasks and uncover insights from vast datasets like never before. However, with this excitement comes inevitable risks and responsibilities. The same qualities that make generative AI such an innovative tool also make it potentially dangerous if not governed carefully. The lack of transparency in how generative AI models work raises concerns about trust and ethical implications. To tackle this and build much-needed trust, it’s critical to ensure that people understand how these models make decisions and comply with regulations.
To ensure innovations don’t mean a lack of safety for enterprise data, comprehensive data governance, controls, unwavering transparency, consistent review, user education and active user involvement are necessary. By implementing these strategies, the secure deployment of generative AI can be enabled. This approach capitalizes on the transformative potential of generative AI while mitigating risks, safeguarding privacy and fueling ongoing research and discourse.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
Read the full article here
