Responsible AI — from research to runtime.
Praxis Labs is an independent AI research lab building the tools, systems, and infrastructure that make modern AI safe, accountable, and useful in practice.
We work across the AI stack — from foundational research on alignment and evaluation, to the runtime systems that govern how AI is deployed in the real world.
Frameworks, benchmarks, and tooling for evaluating AI systems against the qualities that matter most: safety, fairness, transparency, robustness, and accountability.
Infrastructure for the next generation of agentic software — composition, sandboxing, verification, and policy enforcement for systems that act on behalf of users.
Empirical research into how AI models behave under adversarial pressure, distribution shift, and real-world deployment conditions — and how to measure when they fail.
Usable products and developer-facing tools that put responsible-AI research into the hands of practitioners — not just papers.
Phylax — Decentralized Trust Layer for AI Agent Skills
A Bittensor subnet that turns untrusted AI agent skill bundles into Signed Skill Safety Attestations (SSSAs) — portable, cryptographically-signed artifacts with enforceable execution policies. Real sandbox detonation. Evidence-gated scoring.
Phylax (φύλαξ): Ancient Greek for guardian, sentinel, watchman.
More projects coming soon across alignment, evaluation, and applied AI.
- Evidence over assertion. Claims about AI safety should be backed by verifiable traces, not vibes.
- Open by default. Research that shapes how AI is deployed should be inspectable by the people it affects.
- Adversarial mindset. Systems that don't survive adversarial pressure aren't safe — they're untested.
- Useful, not ornamental. Responsible AI is a property of shipped systems, not a marketing checkbox.
- Browse our repositories
- Open an issue or discussion on any of our repos
- Reach out: contact details coming soon
Praxis Labs · Responsible AI · Agents · Alignment · Evaluation