AWS re:Invent 2025 - From Cloud to AI Securing the New Cloud Attack Surface (SEC220)

Securing the New Cloud Attack Surface with AI

The Changing Attack Surface

• AI is being rapidly adopted across cloud environments, with over 80% of environments using at least one hosted AI model or AI-powered service.
• Threat actors are also leveraging AI to accelerate and automate their attacks:
  • AI can be used to generate phishing emails, weaponize vulnerabilities, and automate reconnaissance faster than ever before.
  • Emerging use of AI in malware, such as using language models to receive and execute commands.
  • New attack techniques like "living off the LLM" and compromising AI-related components.
  • Risks from low-code/no-code platforms like Vibe, which can introduce vulnerabilities by design.

A Framework for Understanding the Evolving Threat Landscape

• The presenters propose a quadrant model to categorize the new threats:
  1. Known techniques targeting expanded attack surface
  2. Known techniques targeting new attack surface components
  3. Novel techniques targeting existing attack surface
  4. Novel techniques targeting new attack surface components
• For each quadrant, the presenters outline the required security responses:
  • Improving existing security platforms and practices for known techniques
  • Extending security tooling to cover new attack surface components
  • Innovating to defend against novel techniques
  • Leveraging AI-powered tools to keep pace with attackers

Securing AI Workloads

1. Visibility:
   • Mapping all AI components, including infrastructure, models, data, and applications, is critical for securing AI.
   • The presenters demonstrate how AWS Security Hub's Service Catalog can identify AI services and resources across the environment.
2. Scanning and Monitoring:
   • Extending existing security scanners and posture checks to cover AI-specific risks, such as model vulnerabilities and malicious behavior.
   • Integrating with low-code/no-code platforms like Lavabel to apply security policies across all application types.
   • Leveraging the security graph to prioritize and remediate the most critical AI-related risks and exposures.
   • Implementing runtime monitoring to detect threats like model compromise and anomalous agent activity.
3. Accelerating Security Operations:
   • Using the AWS Security Graph and MCP (Managed Compute Platform) integrations to provide security context and investigation capabilities to security teams.
   • Deploying automated security agents that can query the security graph, perform investigations, and remediate issues without manual intervention.

Enabling Faster, More Proactive Security

• Attack Surface Management:
  • Automatically scanning and validating external exposures, correlating them with the security graph to understand impact and ownership.
  • Automating the response process, including ticketing, pull requests, and remediation, to address exposures before they can be exploited.
• Security Agent Capabilities:
  • Integrating security agents that can leverage the security graph to perform investigations, provide verdicts, and automate remediation.
  • Enabling security teams to scale their operations and focus on higher-impact activities rather than repetitive tasks.

Key Takeaways

• The attack surface is expanding due to the rapid adoption of AI, requiring a multi-faceted security approach.
• Leveraging the security graph and AI-powered tools can help security teams keep pace with the evolving threat landscape.
• Automating security processes, from detection to remediation, is crucial for addressing the speed and scale of modern attacks.
• Integrating security into the development lifecycle, including low-code/no-code platforms, is essential for securing the entire cloud ecosystem.