CCNA Virtualization, Cloud, and AI Essentials Questions

Prompt injection occurs because the model treats user-supplied text as instructions. Input validation and sanitization that detect and neutralize override patterns prevent malicious instructions from being processed as legitimate commands, directly addressing the root cause. Authentication, retraining, and temperature changes do not alter the instruction hierarchy and therefore leave the injection vector open.

By default, Kubernetes allows all pods to communicate with each other and with node services. NetworkPolicy provides a declarative way to restrict ingress and egress at the pod level, effectively segmenting workloads so a compromised container cannot reach unrelated services or the node. Admission controls and resource quotas address different concerns and do not constrain network paths.

Under the shared responsibility model for SaaS, the provider secures the application, runtime, and infrastructure, while the customer owns data governance and access control. Classifying data and deciding what may be stored are data-owner duties, and managing identities, authentication, and permissions controls who can reach that data. Infrastructure patching and physical security remain with the provider.

Live migration traffic carries complete guest memory contents, so it must be both isolated and protected. Placing vMotion on a dedicated VMkernel port group in an isolated VLAN removes it from the guest data path, and enabling vMotion encryption ensures that even a compromised guest or a tapped uplink cannot read or alter the migration stream. Together these controls directly remediate the auditor's concern.

Hypervisor introspection is specifically designed to monitor guest VM memory and state from the hypervisor level, enabling detection of escape attempts. Other controls like virtual firewalls or guest-based HIDS operate at different layers and cannot observe the hypervisor-guest boundary. For detecting direct hypervisor memory access by a guest, introspection is the appropriate virtualization-specific control.

Proper S3 configuration requires a defense-in-depth approach that prevents public access while explicitly restricting actions to necessary services. By blocking public access, enforcing TLS for transit, and using explicit IAM roles, the organization minimizes the attack surface. This is vital because S3 buckets are frequent targets for misconfiguration that leads to data exposure, making granular policy control an essential defensive requirement for protecting cloud-based sensitive information.

Hardware-based confidential computing features such as AMD SEV-ES and Intel TDX encrypt VM memory so that even the hypervisor and co-resident VMs cannot read it. This directly addresses the cross-VM memory disclosure, whereas the other options address migration encryption, network segmentation, or console clipboard controls, none of which isolate physical memory between guests on the same ESXi host.

Instance metadata service credentials are a frequent target because they grant the instance's role permissions. IMDSv2 adds a session-oriented token requirement that defeats simple SSRF and credential-harvesting scripts, while a hop limit of 1 blocks access from containers or proxies on the instance. These controls preserve legitimate access through the token flow while removing the easiest paths to steal temporary credentials.

The shared responsibility model is the framework that assigns security ownership between the cloud provider and the customer. For IaaS, the provider handles the physical datacenter, hardware, and hypervisor, while the customer is responsible for the guest OS, applications, and guest-level firewall configuration. The other concepts address access scope, layered controls, or automation, not the division of duties.

The two symptoms — reading another build's environment variables and reaching the host filesystem through a mount — both stem from workloads sharing a single kernel and Docker Engine instance. Container isolation is namespace- and cgroup-based, so a misconfigured or privileged container can see peer processes and host paths. Placing each team's agents in its own virtual machine restores a separate kernel and process table per tenant, which is the change that directly removes both exposures at once.

A VM escape is a critical vulnerability where an attacker gains access to the host machine from a guest VM. This allows the attacker to compromise other VMs on the same host or the physical hardware itself. Understanding this threat is essential for GSEC professionals, as it represents the highest level of breach in a virtualized infrastructure, requiring rigorous patching and hypervisor hardening.

Infrastructure as Code (IaC) allows for version-controlled, automated, and audited deployment of cloud resources. By treating infrastructure as software, security teams can scan templates for misconfigurations before deployment. This is vital in cloud environments because manual configuration is prone to human error, which is the leading cause of security breaches in modern cloud and virtualization deployments.

Micro-segmentation is critical in virtualized environments because traditional network perimeter defenses cannot see traffic moving between VMs on the same host. By applying host-based or hypervisor-level firewalls, security teams restrict traffic based on identity and function rather than IP address. This mitigates the risk of a compromised workload pivoting to sensitive internal assets within the shared virtual infrastructure, which is a key security requirement for modern cloud architectures.

Reducing sensitive data exposure in an LLM assistant requires controlling both what data the model can access and what it can emit. Least-privilege tool access limits the sensitive fields available to the model, while output filtering catches and redacts sensitive patterns before the user sees them. Together they provide defense in depth; the other options affect randomness, model artifact storage, or training, none of which govern runtime data flow.

Client-side encryption with customer-managed keys stored on-premises ensures that the cloud provider never has access to the encryption keys or plaintext data. Provider-managed keys leave the provider with decryption capability, failing the requirement. Other options like CASB or server-side encryption do not fully prevent provider access to plaintext.

Least privilege for a serverless function means its execution role should grant only the actions and resources required to do its job. Replacing an administrative role with one scoped to the specific bucket and database table removes unnecessary permissions and reduces the impact of compromise. Credential encryption, concurrency limits, and network restrictions do not shrink the role's effective permissions.

Admission control is the enforcement point in Kubernetes that can evaluate an image's signature or scan attestation before a pod is scheduled. By requiring a valid signature or attestation from the internal scanner, the firm guarantees that only images that passed its process can be deployed. Network policies, pull policies, and non-root settings affect runtime behavior but not image admissibility.