Services
Every engagement is scoped to your architecture and threat model. Pick a category below, or tell us what you're building and we'll recommend the right approach.
Adversarial testing for LLM apps, agents, and model pipelines.
LLM apps, agents & model safety
End-to-end adversarial testing of AI systems: prompt injection, tool and agent abuse, data exfiltration and insecure output handling, plus jailbreaks and guardrail bypasses, including review of what an autonomous agent is authorized to do and how far it can be pushed beyond it.
Retrieval-Augmented Generation pipelines
Security assessment of Retrieval-Augmented Generation systems: detecting index / knowledge-base poisoning, insecure retrieval, information leakage through context, and document-level manipulation.
Smart-contract and full-protocol security across EVM and non-EVM chains.
Economics, oracles & governance
Whole-protocol security review that goes beyond individual contracts, analyzing economic design, liquidation and incentive mechanisms, oracle dependencies, and governance to surface systemic and market-driven risk.
Solidity & Rust · EVM & non-EVM
Line-by-line security review of smart contracts across EVM chains (Solidity on Ethereum and L2s) and non-EVM ecosystems (Rust on Solana, NEAR, and beyond), covering economic, protocol, implementation, and chain-specific program-model risks.
Offensive security across your apps, APIs, cloud, and people.
Web, mobile, desktop & APIs
Manual, exploitation-driven penetration testing of web, mobile, and desktop applications and their REST, GraphQL, and gRPC APIs, chaining real vulnerabilities into demonstrable attack paths across the OWASP Web and API Top 10.
Adversary simulation & human-layer attacks
Objective-driven, stealth-oriented simulation of a real threat actor across the full kill chain, including phishing and social-engineering campaigns that test not just your tech but your people, detection, and response.
AWS · Azure · GCP config & secrets
Configuration and secrets review across AWS, Azure, and GCP: surfacing excessive permissions, exposed resources, weak IAM, insecure defaults, and leaked credentials across repos, images, buckets, and pipelines before an attacker finds them.
Threat modeling, secrets reviews, and AI security training.
Threat modeling & secrets management
Design-time security guidance combining collaborative threat modeling with a review of your secrets management (storage, rotation, Vault/KMS), to identify and prioritize risk before code ships.
Technical training on AI best practices
Hands-on technical training for engineering and security teams on building and defending AI systems: secure LLM, agent, and RAG design; prompt-injection and jailbreak defense; guardrails; and the OWASP LLM Top 10, mapped to your own stack.
How we work
We define targets, rules of engagement, and success criteria together.
Hands-on assessment with continuous communication on critical findings.
Severity-rated findings, PoCs, and prioritized remediation guidance.
We re-test your fixes and confirm each issue is genuinely resolved.