Ranges rather than a benchmark
Public cyber benchmarks are fixed task sets, and once a set is public it tends to end up in training data, so a high score can reflect a solution the model has already seen as much as what it can do inside a real environment. Pwned Labs generates a new PwnRange instance every time, with randomized names, addresses, credentials, flag placement and decoy resources, so an agent cannot replay a run it has already done and has to find the path in the environment it is actually given.
What we cover
Pwned Labs runs its ranges on live control planes rather than emulations. Scenario families cover AWS IAM privilege escalation and role chaining, Entra ID and hybrid identity paths from a web foothold to on-premises Active Directory, GCP service account and federation abuse, Kubernetes RBAC, secrets and escape chains, and CI/CD and supply chain pivots through pipelines, runners and artifact registries. AI-system surfaces such as MCP servers, agent tool credentials and indirect prompt injection can be composed into any family.
AWS Azure GCP Kubernetes CI/CD
How it works
1 Deploy
Each run gets an isolated environment provisioned from Pwned Labs infrastructure as code into a dedicated account, subscription or project, with no shared state between runs.
2 Attack
Each run starts from an entry point set in the lab: a public web application, leaked cloud credentials, exposed Kubernetes credentials, or a vulnerable service reachable on the network. The model connects through an Inspect-compatible agent bridge or an SSH entry point on an attack host, with the objective and the scope stated in the prompt, and scenarios can also be delivered as OpenEnv environments or into accounts you supply.
3 Score
Progress is scored against milestones and flags that exist only in that instance, each checked as reachable when the environment is built. A run returns a scorecard broken down by objective and stage, and a harness can stream its own telemetry to the run so a trajectory sits alongside the score.
Built for controlled evaluation
Pwned Labs runs each range with the model API as its only permitted egress. Network policy is deny by default and verified before each run, real-time monitoring can stop a run on out-of-scope traffic, and no real credentials or real organization names exist inside a range. More detail on isolation and data handling is on our Trust and Security page.
Evaluation and training from one generator
For evaluation, a scenario family keeps a fixed attack graph and a difficulty label while every instance differs, so results compare across models and over time. For training, families vary in topology and can be produced at volume with verifiable rewards. Results can be reported through PwnBench, the Pwned Labs scenario suite with difficulty labels and human expert baselines.
Who it is for
Pwned Labs builds PwnRange for model developers evaluating cyber capability under their safety frameworks, government and third-party evaluators who need realistic enterprise and cloud ranges, and agent builders, red team agent startups and security vendors building on frontier models who need somewhere real to test them.
Working with us
Early access to PwnRange starts with a scoped pilot of scenario families that Pwned Labs builds to your requirements. We provide the range, either hosted by us or deployed into accounts you supply. You run your own agent, on your own model and your own keys, and we do not host models.
Frequently asked questions
What is PwnRange?
PwnRange is a Pwned Labs service that generates isolated attack ranges across AWS, Azure, GCP, Kubernetes and CI/CD so AI labs can evaluate and train offensive security models. Every run gets a fresh environment, the model is given an objective, and the run is scored against flags that exist only in that instance.
How is a generated range different from a public cyber benchmark?
Public cyber benchmarks are fixed task sets, and once a set is public it tends to end up in training data, so a high score can reflect a solution the model has already seen. PwnRange generates a new instance every time, with randomized names, addresses, credentials, flag placement and decoy resources, so an agent cannot replay a run it has already done and has to find the path in the environment it is actually given.
Which environments do PwnRange scenarios cover?
Scenario families cover AWS IAM privilege escalation and role chaining, Entra ID and hybrid identity paths from a web foothold to on-premises Active Directory, GCP service account and federation abuse, Kubernetes RBAC, secrets and escape chains, and CI/CD and supply chain pivots. AI-system surfaces such as MCP servers, agent tool credentials and indirect prompt injection can be composed into any family.
How does a model connect to a PwnRange range?
Through an Inspect-compatible agent bridge or an SSH entry point on an attack host, with the objective and the scope stated in the prompt. Scenarios can also be delivered as OpenEnv environments or into accounts you supply.
How is a PwnRange run scored?
Progress is scored against milestones and flags that exist only in that instance, each checked as reachable when the environment is built. A run returns a scorecard broken down by objective and stage, and a harness can stream its own telemetry to the run so a trajectory sits alongside the score.
How do we get access to PwnRange?
PwnRange is in early access, which starts with a scoped pilot of scenario families that Pwned Labs builds to your requirements, with the range either hosted or deployed into accounts you supply. You run your own agent, on your own model and your own keys.
About Pwned Labs
Pwned Labs is a hands-on cloud and AI offensive security training platform. It runs real-world attack labs and bootcamps for AWS, Azure, GCP, Kubernetes and AI systems, including the ACRTP, MCRTP, GCRTP, KRTP and AISRTP certifications. Free labs are at pwnedlabs.io/explore.