Harnessing AI for Justice, Security, and Opportunity for All

Artificial Intelligence is revolutionizing industries across the globe, and the federal government is no exception. The potential for AI to improve efficiency, decision-making, and public service delivery is immense.

Whether using algorithmic AI to analyze large datasets for insights, predict patterns, identify threats and prevent fraudulent activities, or deploying GenAI to improve employee productivity and improve constituents’ experiences, the benefits of AI are clear-- AI can enhance national security and improve government operations.

However, integrating AI into federal operations comes with significant challenges, particularly around protecting the data, algorithms, and AI models. AI systems require access to large datasets, that can include a lot of sensitive information.

Data security is as important as ever—data stays in the models and can be exploited through prompt engineering attacks. Equally important is safeguarding the AI algorithms and models to ensure they remain untampered with, allowing for trusted outputs.

On October 30, 2023, The White House issued an Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence. It is urging agencies to harness AI for good, and while realizing its myriad benefits, they must mitigate its substantial risks.

Steps to Ensure Secure Deployment of Algorithmic or Generative AI

1. Develop a Comprehensive AI Strategy

• Align with Mission Objectives: Ensure that AI initiatives are aligned with the agency's mission and goals.
• Stakeholder Involvement: Engage all relevant stakeholders, including policymakers, technologists, domain specialists, and end-users in the planning process to ensure holistic development and deployment of AI systems.
• Establish AI Governance: Put together a set of policies, procedures, and best practices to guide ethical and responsible use of AI.
• Drive Awareness: Educate employees and partners on risks that AI poses.

2. Prioritize Data Security

• Implement Robust Encryption: Use encryption to protect sensitive data across its full life cycle: at-rest, in-transit, and in-use.
• Enforce Zero Trust: Allow use of decrypted data by verified and authorized personnel only.
• Carry Out Regular Security Audits: Conduct regular security audits to identify and mitigate vulnerabilities.

3. Ensure AI Model Safety

• Prevent Data Poison Attacks: Have guardrails on what data can be fed into AI models to prevent manipulation or influencing the model.
• Verify Model Integrity: Perform regular testing and verify model integrity to check and maintain the trustworthiness of the system.

4. Monitor and Evaluate

• Continuous Monitoring: Implement continuous monitoring of AI systems to ensure they are functioning as intended and to detect any potential issues early.
• Feedback Loops: Establish feedback loops to gather input from users and stakeholders, enabling ongoing improvements.

Fortanix for AI and Data Security

At Fortanix, we live and breathe data security. We have built the Fortanix platform on Confidential Computing, so operations can be performed in a trusted execution environment for maximum protection from prying eyes.

Confidential Computing keeps your sensitive data and applications in memory safe, even if the underlying server infrastructure has been compromised. Confidential Computing enables our customers to elevate their data and AI security practice. For Federal government, Fortanix has helped solve several challenges, including the following:

• A Government Agency focused on healthcare improves and accelerates drug development: The agency needed insights from broad healthcare data-- one that resides across doctors’ offices, hospitals, and pharmaceutical industries. The sensitive data to be run in analytics and AI was tokenized to keep it private. Data decryption, when necessary for authorized and privileged personnel, happened in a confidential compute enclave.
• A Government Agency safeguards critical infrastructure data: The agency collected filed data and ran it in their AI models with Fortanix Confidential AI, installed on the edge. The data from the several edge infrastructures was consolidated in Fortanix Confidential Data Collaboration for further analysis, insights, and validation of the data integrity, all the while preventing any data exposure.
• Intelligence Community Agency uses computer vision to enable remote object detection: Using a large synthetic data set, the Agency securely trained an AI model to identify the differences between different types of airborne objects: jet airliner, fighter jet, missile. All data training and inferencing were done in a secure Trusted Execution Environment, with the result being encrypted before being delivered back to the requesting analyst.

Conclusion

Federal Agencies should consider the implementation of AI solutions and continually address potential risks and challenges. By adhering to data security best practices and prioritizing data and AI security, the federal government can harness the power of AI to better serve the public. At Fortanix we are committed to helping the public sector safely adopt AI for justice, security, and opportunity for all. Contact us today to learn more about our data security and Confidential AI solutions.