Security
Operations

Six Security Questions to Ask Before Choosing an Application Platform

Evaluate a platform's isolation, credentials, secret access, verification, and recovery with six practical security questions.

Forte TeamPublished November 16, 20267 min read

An application platform runs code supplied by customers, including customers who may be trying to attack it. It also holds the authority to build releases, manage infrastructure, and supply secrets to applications. A security evaluation needs to establish what prevents that customer code from acquiring the platform's wider authority.

A smooth deployment demo won't answer that question. These six questions follow the workload and its credentials through the protections that should contain a compromise, then through recovery if containment fails. We believe providers should explain those protections concretely, and that includes Forte.

1. What separates my code from another customer's code?

Start with the fact that customers cannot be assumed to trust each other. They can install dependencies and run build scripts because executing their code is the service they bought. The provider needs an isolation model that treats malicious code as an expected input.

Builds and running applications both need that review. Ordinary containers share a host kernel, and giving them separate filesystem views doesn't create a hardware-virtualization boundary. At Forte, we require VM-level isolation for untrusted customer execution. Our architecture article explains the position and the additional controls it needs. AWS's Fargate isolation guidance provides a concrete example of hardware-virtualized tasks recommended for strong isolation.

Follow storage as well as processes. Source, temporary files, and caches may survive or cross boundaries differently from a running container. Ask where they reside, what another workload can reach, and what happens when a build finishes or an environment is reassigned. “Everything runs in containers” is incomplete without an explanation of which workloads share a kernel, storage, or privileged services and whether any execution path uses weaker protection.

2. What can a compromised workload credential do?

Isolation cannot stop an attacker from using a valid credential available inside the compromised workload. The next protection is the scope of that credential: which resources it can read or change, which services accept it, and whether it can obtain a more powerful identity.

Pick a build identity and examine effective permissions, including inherited groups, roles, resource policies, and identities it can assume. Its scope might be one artifact, build, or project. The provider should be able to explain why that scope is needed for the work, rather than presenting a short expiration time as the whole protection.

A credential lasting minutes can still authorize a large transfer. Expiration limits the opportunity to use it, while permission scope limits what that use can accomplish. The AWS IAM security practices recommend temporary credentials and least privilege together. Request evidence for both and for who is permitted to obtain fresh credentials.

3. Can customer execution reach privileged platform systems?

Scheduling workloads and managing infrastructure require authority customer code shouldn't inherit. The provider needs to explain which paths connect those systems to customer execution and what checks a caller faces along the way.

Private networking can restrict reachability. Once a connection succeeds, service authorization must still decide whether this identity may perform this operation on this resource. Include administrative endpoints and older components in the review; a stronger rule at one API doesn't help if another exposes the same data with weaker checks.

Follow a hypothetical stolen identity through those connections. Could it use another service, assume a different role, or retrieve a credential with broader permissions? A useful answer explains which control still denies access if another control fails. Simply placing a service on an internal network doesn't answer that question.

4. Who can retrieve and decrypt my secrets?

Encryption protects stored data only within its access model. If an identity can retrieve encrypted secrets and ask the encryption service to decrypt them, it can obtain the plaintext without defeating the encryption algorithm.

Ask which components can read encrypted records, which identities can request decryption, and how those permissions are limited to the customer and workload that need the value. AWS KMS key policies illustrate that key use has its own authorization rules. Protecting a root key doesn't remove the need to restrict requests to use it.

Then follow the plaintext. It might enter a build, running application, workflow history, or operator tool. Establish which copies persist and who can retrieve them. This is also a responsibility your team shares: supplying an account-wide administrator credential to an app that only needs to read one database grants authority infrastructure isolation cannot remove.

5. What evidence shows that these boundaries hold?

A diagram describes the intended architecture. Evidence needs to show how the deployed system is checked against it, including attempts at actions that should fail. Ask about tests in an authorized assessment environment that try to read another workload's artifact, use a workload token against a privileged service, or retrieve another customer's configuration.

Coverage and timing determine what those results establish. Which products and regions were checked, when was the relevant control last evaluated, and who owns exceptions? The provider should also explain how new features and operational shortcuts are reviewed so they don't quietly introduce weaker paths.

Independent assessments add evidence within their scope. A penetration test or compliance report covers particular systems over a particular period. Ask how its findings relate to the workload and execution mode you would use instead of treating the report's existence as proof about every service.

6. How do I recover safely while the platform is containing an incident?

Replacing an exposed secret is useful only if the replacement isn't immediately exposed through the same compromised path. The response plan needs to explain how customers revoke credentials, obtain replacements, and restore service while the provider contains the incident.

Consider losing access to builds, deployments, sign-in, or the normal support portal. Establish which actions your team can perform directly with the database, payment processor, or other credential issuer, which depend on the platform, and how the provider will communicate when it is safe to supply replacement secrets.

Scope assessment matters too. Ask which records help establish access to your resources, what the provider can determine from them, and what may remain uncertain. Your team needs a reliable place to receive that information when its normal workflow is unavailable.

Practice the customer-controlled steps in staging using a synthetic credential: revoke it, replace it, deploy, and confirm application recovery. Record the dependencies the exercise reveals. An incident is a difficult time to discover that your recovery account depends on the affected sign-in service.

Keep the evidence with the answer

QuestionEvidence to request
What separates customer execution?Build and runtime isolation model, workload placement, storage boundaries, and exceptions.
What does a stolen credential authorize?Effective permission scope, trust policies, lifetime, and paths to assume other identities.
What separates customer code from platform operations?Relevant network paths, service authorization, and restrictions on administrative access.
Who can retrieve and decrypt secrets?Access model, plaintext handling, retention, and audit coverage.
How are boundaries verified?Assessment scope, tests of denied access, review dates, and ownership of exceptions.
How does safe recovery work?Customer actions, alternative communication paths, and a tested rotation procedure.

Record each answer, its evidence, the review date, and unresolved exceptions. That record lets the team assess a changed requirement or new execution mode without reconstructing what the original evaluation covered.

Forte's position is that VM isolation belongs in the foundation of customer compute, accompanied by restricted identities and independently enforced permissions. Our security requirements and published controls and responsibilities are starting evidence to examine, rather than a reason to skip the questions. Bring the table to a technical evaluation through Forte Support and ask where each answer applies to your proposed workload.

Bring your security review to Forte

Use the evaluation table to discuss your application, required controls, and the evidence your team needs.

Why Forte Requires VM Isolation for Customer Code
Security
Architecture

Why Forte Requires VM Isolation for Customer Code

The security requirements behind how we build and run your applications: VM isolation, least privilege, and independent access controls.

6 min read
Railway's Breach Shouldn't Have Been Possible
Security
Architecture

Railway's Breach Shouldn't Have Been Possible

Customer builds shared a VM, and one stolen build credential led to other customers' secrets. Those are architectural failures, not zero-day exploits. Our founders spent 30+ combined years as security leaders and staff engineers at AWS and Meta.

6 min read
Choosing Which SaaS Vendors to Keep
Operations
Architecture

Choosing Which SaaS Vendors to Keep

Choose which SaaS tools to keep by testing their jobs, required features, exports, and the steps needed to leave.

5 min read