Neuraphic builds products that defend AI systems from adversarial attacks. The only way to know those defenses actually work is to have people whose full-time job is breaking them. This role exists to find the failures before anyone else does — and to turn those failures into better products.
What you'll work on
You'll red-team Prion and Claeth continuously. Not surface-level prompt injection tests — deep, creative adversarial campaigns that probe the boundaries of what our models can detect. You'll develop novel attack techniques, document them rigorously, and work with the ML team to close the gaps you find. Every successful attack you execute makes the next version of the product harder to break.
A significant part of this role is building attack datasets. The models that power our products are only as good as the adversarial examples they've been trained against. You'll construct and curate taxonomies of prompt injection, jailbreaking, and manipulation techniques — organized not just by method but by intent, severity, and evasion sophistication. These datasets become training data, evaluation benchmarks, and the foundation of our security research.
You'll also conduct vulnerability research against AI systems more broadly, contributing to the field's understanding of where these systems fail and why. Some of this work will be published. Some will stay internal. All of it will inform the products.
What we're looking for
Someone who breaks things methodically, not randomly. You approach a system the way a lockpick approaches a lock — with patience, a theory of how it works, and a willingness to try approaches that shouldn't work but might. You've done adversarial testing against real systems, whether that's AI models, web applications, or compiled binaries. The domain matters less than the mindset.
You write clearly. Security research that can't be communicated is security research that doesn't ship. You'll write internal reports, contribute to external publications, and explain complex failure modes to engineers who need to fix them. If you've published CVEs, written technical blog posts, competed in CTFs, or built tooling for offensive security — all of that is relevant.
How to apply
Email [email protected] with the subject line "Security Researcher." Include your resume, links to any published research or security work, and a short description of the most interesting vulnerability you've ever found — what made it interesting, and how you found it.