Securing the AI Attack Surface with Tenable One

Securing the AI attack surface with Tenable One

As artificial intelligence rapidly integrates into corporate environments, organizations face new cyber exposure challenges, from unauthorized shadow AI tools to over-privileged rogue agents. In this webinar, we explore how Tenable One helps you discover, assess, and govern your AI attack surface to proactively manage cyber risk.

[00:01:46] Securing the AI attack surface

We introduce the core challenges of securing modern AI deployments, focusing on the hidden cyber risks introduced by employees and automated tools.

• Shadow AI expansion: Understand how shadow AI goes beyond employees using personal accounts to include unauthorized software and libraries running on corporate machines.
• Rogue agents: Learn how automated AI agents operate within your environment and why discovering and governing them is essential.
• Exposure management: See how Tenable One helps you discover, assess, and act on AI-related cyber risk across your organization.

[00:11:23] Understanding hidden AI risks and attack paths

AI adoption frequently outpaces security, widening your attack surface with every new tool, Copilot, or agent introduced into the environment.

• Data leakage in prompts: Prevent sensitive PII or PCI data from leaking through hardcoded skills and employee chat prompts.
• Unknown AI footprints: Identify unapproved AI libraries or backend servers that chatbots download to complete tasks.
• Hidden attack paths: Discover how over-scoped IAM roles connected to public-facing chatbots can inadvertently expose internal data.

[00:17:15] Discovering shadow AI software and assessing permissions

Gaining visibility into the AI software running on your servers is the first step in addressing unresolved vulnerabilities in your cloud workloads.

• Software inventory: Automatically detect installed AI frameworks, SDKs, and agent runtimes that bypass standard approval processes.
• IAM policy review: Identify over-privileged IAM policies associated with AI workloads to prevent unauthorized access to sensitive secrets.
• Container visibility: Monitor sandboxed environments and containers used by AI coding assistants to uncover hidden vulnerabilities.

[00:20:00] Governing conversations and monitoring prompt injections

Monitoring user interactions with AI tools is critical to identifying intentional abuse and preventing sensitive data exposure.

• Conversation tracking: Gain visibility into prompts sent to platforms like Copilot, ChatGPT, Claude, and homegrown AI applications.
• Jailbreak detection: Detect deliberate prompt injection attacks and model refusal bypass attempts from internal users.
• Behavioral analysis: Prioritize recurring high-risk activities, such as a single user repeatedly attempting to bypass AI guardrails.

[00:25:22] Assessing AI skills and generative AI posture

Beyond conversations, you must assess the structural configurations and automated skills that power your generative AI workflows.

• Hardcoded secrets: Detect AI skills that inject hardcoded authentication tokens or API keys into reusable code artifacts.
• Skill sharing controls: Ensure that overly permissive AI skills are not shared globally across your entire organization.
• Posture management: Enforce security best practices, such as requiring SSO, disabling public data sharing, and mandating trusted devices for tools like Claude Code.

[00:31:28] Inventorying AI agents and reviewing instructions

Business users frequently create custom AI agents that act autonomously, making continuous inventory and instruction review vital.

• Agent discovery: Automatically catalog Copilot agents, custom GPTs, and Claude projects created by users across your environment.
• Instruction analysis: Identify agents with empty instructions, which are prone to unpredictable behavior and unauthorized internet access.
• Access level auditing: Spot internal agents with access to sensitive internal documents that are accidentally exposed to the public.

[00:38:20] Analyzing agentic identity and cross-tenant blueprints

AI agents often rely on shared blueprints that dictate their permissions, creating scalable cyber risks if those blueprints originate from untrusted sources.

• Agentic identity tracking: Map how individual agents are created from underlying blueprints and trace their associated permissions.
• Cross-tenant risks: Identify external blueprints pulled into your organization that could maliciously alter the behavior of internal bots.
• Permission inheritance: Ensure that agents do not automatically inherit over-privileged Graph API permissions from misconfigured templates.

[00:41:52] Connecting attack paths and automating response

By connecting isolated findings, you can visualize complete attack paths and take automated actions to mitigate critical AI risks.

• Attack path visualization: Link AI chatbot vulnerabilities to sensitive data stored in cloud buckets to understand real-world cyber exposure.
• Riskiest user identification: Rank users and shared service accounts based on the volume and severity of their high-risk AI activities.
• Automated remediation: Trigger webhooks, send alerts, or pause software licenses automatically when severe AI policy violations occur.

[00:48:03] Final takeaways and Q&A

We conclude the session by answering audience questions and summarizing how to operationalize AI security within your environment.

• Expanded coverage: Protect emerging tools like Claude Code alongside established enterprise AI platforms.
• OWASP mapping: Align your AI security efforts with the OWASP Top 10 for LLMs, covering prompt injections, excessive agency, and supply chain attacks.
• Next steps: Activate AI discovery in Tenable One to close easy gaps, monitor risky conversations, and build automated response workflows.

Watch the Full Webinar

Ready to secure your AI attack surface? Watch the full webinar on-demand to see how Tenable One provides the visibility and governance you need to adopt artificial intelligence safely. If you have questions about enabling these features, reach out to your Tenable representative today.