AI and Cybersecurity

Understanding the relationship between artificial intelligence (AI) and cybersecurity is vital for Dartmouth College employees to maintain strong digital defenses. This quick article covers the following:

• History of AI
• Benefits of AI for Cybersecurity
• How AI is being utilized in cyber attacks
• Generative AI and cybersecurity
• Determining if AI tools are trustworthy

History of AI

• The term “artificial intelligence” was conceptualized in the 1950s after Alan Turing, a British mathematician and logician, speculated about “thinking machines” that could reason like humans.

Benefits of AI for Cybersecurity

• According to a recent research report, the estimated global market for AI cybersecurity products will climb to about $135 billion by 2030
• AI can be used by cybersecurity organizations to supplement traditional tools like antivirus protection and data-loss prevention.
• AI can analyze large sets of data and find patterns, making it useful in cybersecurity tasks, including the following:
  • Detecting attacks (and reducing false positives)
  • Identifying and flagging email phishing attacks
  • Simulating social engineering attacks to test an organization’s security awareness
  • Penetration testing designed with AI to target an organization’s specific technology (identifying weaknesses before they can be exploited by hackers)
• Potential to stop security incidents before they occur, lowering IT costs

How AI is being utilized in cyber attacks

• Social engineering schemes - automated by AI to generate more attacks in a shorter period of time
• Password hacking - AI helps improve algorithms for password hacking, providing quicker and more accurate password-guessing
• Deepfakes - AI can make and distribute fake audio and video content to realistically impersonate an individual
• Data poisoning - altering the training data used by an AI algorithm, leading to bad output

Generative AI and cybersecurity

• What is generative AI? - a branch of AI focused on using existing data to generate new data
• Use cases
  • Data retrieval/analysis (e.g. assisting threat hunters with data retrieval for real-time insights into vulnerability management workflows)
  • Content generation and summarization (e.g. learning and summarizing security documentation)
  • Generate complex, unique passwords or encryption keys that are more difficult to hack, offering an additional layer of security
• Pros of generative AI in cybersecurity
  • Efficiency
  • In-depth analysis and summarization - ability to quickly and accurately conduct previously time-intensive data analysis tasks
  • Proactive threat detection - shifting from reactive to proactive cybersecurity to detect threats before they occur
• Cons of generative AI in cybersecurity
  • Requires high levels of computational resources (power, storage, etc.)
  • Open-sourced generative AI (e.g. GPT-based tools) is making it easier for cybercriminals to conduct attacks
  • Ethical questions related to privacy and control over data used in AI training datasets

Determining if AI tools are trustworthy

• NIST has developed an AI Risk Management Framework (AI RMF) to better manage risks to individuals, organizations, and society associated with AI
• Decision-making should involve a broad set of stakeholders considering the relative risks, impacts, costs, and benefits.
• Ongoing monitoring, testing, and adaptation are crucial for maintaining trustworthiness as AI systems evolve.
• Human judgment is essential in determining trustworthiness metrics and thresholds.
• Characteristics of Trustworthy AI Systems:
  • Valid and Reliable
    • Fulfillment of intended use
    • Consistent performance over time.
    • Requires system functionality and structure
  • Safe
    • AI systems should not endanger human life, health, property, or the environment
    • Achieved through responsible design, deployment, and decision-making practices
  • Secure and Resilient
    • Allows systems to withstand adverse events.
    • Prevents unauthorized access and use.
    • Maintains system functionality and structure
  • Accountable and Transparent
    • Ensures access to information about the system and its outputs.
    • Enables individuals to understand AI system operations and outcomes
  • Explainable and Interpretable
    • Reveals the mechanisms underlying AI systems.
    • Provides meaning to system outputs.
  • Privacy-Enhanced
    • Safeguards human autonomy, identity, and dignity.
    • Privacy norms and practices guide AI system design and development.
  • Fair with Harmful Bias Managed
    • Addresses harmful bias and discrimination.
    • Bias categories include systemic, computational, and human-cognitive biases.
• Risk Management and Context:
  • Organizations must balance characteristics based on context, risks, and impacts.
  • Human judgment is essential in determining trustworthiness metrics and thresholds.