Adapt beyond the scan
Drost maps the real surface, forms hypotheses, uses the target response to choose the next action, and changes course when the evidence changes.
Human-supervised autonomous offensive security
Find out how far an attacker can actually go.
Drost doesn't stop at finding vulnerabilities. It autonomously discovers, exploits, and chains weaknesses across your full technology stack—pursuing each foothold until it proves the deepest reachable impact.
Follow the attack pathEvidence at every boundary
- 01Public surface
- 02Identity
- 03Source & CI/CD
- 04Cloud control
- 05Operations
Drost does not assume every engagement reaches every layer. It advances only where scope, evidence, and the rules of engagement permit.
The real attack surface
Attack paths do not stop at the application. Neither does Drost.
Traditional testing often divides the environment into separate products and reports. Adversaries do not. Drost follows the connected path and records where access becomes authority, and where authority becomes material impact.
Cross system boundaries
The meaningful path may begin in a web application and continue through identity, automation, source control, cloud authority, or operational infrastructure.
Stop at proof
Observations remain observations. Confirmed findings require evidence from the real target, with uncertainty and untested paths kept visible.
Drost Community Edition
The open-source foundation of Drost.
Connect your coding agent to a container-native security toolkit with a single command. Drost Community Edition packages the original Drost-v1 MCP server and its executable toolchain for local, authorized security work.
Install Community EditionOne multi-platform container
159 MCP tools
Apache-2.0
Two ways to use Drost
Run Drost Edge. Or bring in the team.
Use Drost Edge for self-directed testing from your own machine, or bring in Drost Elite when the attack path demands a managed, human-supervised operation.
Drost EdgePublic beta
Autonomous security validation from your terminal.
Run authorized engagements from macOS or Linux, monitor usage and credit in the Edge Console, and receive findings immediately on your computer.
- Customer-controlled local runtime
- Pay-per-token, pay-as-you-go pricing
- Findings immediately available locally
Drost EliteManaged
Drost can be the elite team.
A human-supervised autonomous offensive-security operation for environments where the attack path may cross teams, systems, and high-consequence control planes.
- Written rules of engagement and stop conditions
- Human authorization at consequential gates
- Containment, remediation, retest, and closure support
Proof before promise
Cyber capability should leave receipts.
A model's confidence is not a security outcome. Drost is evaluated on whether it can produce evidence from the real target, preserve uncertainty, and remain honest about what was and was not proven.
- authorized DrostBench targets
- 100
- best evaluated end-to-end result
- 73/100
- sanitizer-confirmed native-code receipt
- ASAN
DrostBench v1
A 100-challenge offensive-security benchmark
Three frontier-model configurations, one fixed Drost attack stack, 30 minutes per target, and end-to-end objective capture as the win condition.
Read the researchKnown-answer validation
Drost rediscovered nginx-poolslip and proved it with ASAN
Source-to-sink reasoning, a working trigger, and a sanitizer-confirmed crash - clearly labeled as reproduction of a known issue, not a new CVE claim.
Read the researchDrostBench results measure objective capture on a benchmark, not universal real-world performance. The nginx result is a known-answer validation and is not presented as a novel vulnerability discovery.
Digital-asset infrastructure
Audit the organization behind the protocol.
Onchain code is only one part of the system. Drost validates whether an adversary can compromise the people, pipelines, cloud authority, signing systems, nodes, and operational controls capable of changing what reaches the network.
Explore digital-asset assuranceProtocol operations
Signing authority
Cloud & delivery
Drost Defender · live shadow mode
The evidence doctrine also runs on defense.
Drost Defender observes real attack traffic, preserves evidence independently of model judgment, and publishes a sanitized view of its investigation on an intentionally public testbed.
Evidence-grounded investigation without placing a model in the request path.
Shadow-only today: it observes and recommends, but does not block or remediate.
The trust boundary is the product
Offensive capability needs explicit control.
Drost is built to increase the depth of authorized testing without hiding scope, authority, evidence quality, or uncertainty behind an AI-generated answer.
Scope before action
Targets, exclusions, rules of engagement, and stop conditions are explicit before testing begins.
Human authority at the gates
Consequential actions remain subject to the agreed authorization model and human supervision.
Evidence before conclusion
Confirmed findings require receipts from the real target. Unsupported theories do not become results.
Uncertainty remains visible
Untested paths, partial proof, negative results, and unresolved coverage stay clearly distinguished.
Confidential scoping
Bring us the boundary you cannot afford to misunderstand.
Tell us what is at stake, what is in scope, and what proof would change the decision. Start directly with hello@drost.ai.