A model follows hostile instructions embedded in retrieved content.
AI evaluation · governance · assurance
We test AI systems.
We test AI systems.
Then we make the evidence useful.
Syntony red-teams and evaluates AI systems, turns findings into practical governance, and re-tests systems and controls as models, integrations, and risks change.
EvaluateRed teaming, model and agent evaluation, benchmarks
GovernOwners, controls, escalation, and decisions
AssureRegression testing, change review, continuous evaluation
syn·to·ny /ˈsɪntəniː/ noun
The work is between evidence and action.
A red-team finding matters when someone can act on it. Syntony keeps the evidence intact as it moves from a technical test to an owner, control, and decision, then back into re-testing. Research, engineering, systems analysis, and strategic foresight support that loop when the problem calls for them.
The assurance loop
A finding should change something.
A red-team result is useful when the evidence can be reproduced, a responsible team can act on it, and the resulting control can be tested again.
Trace, prompt, tool call, affected privilege, and replayable test case.
System security owns remediation. The program owner owns release conditions.
Content isolation, tool-call approval, and a retrieval allowlist.
Hold privileged tool use from untrusted retrieved content until the control passes.
Replay the attack after remediation and after material model, retrieval, or tool changes.
Services
Three parts of one assurance loop
Start with evaluation, governance, or ongoing assurance. Research and engineering support each stage when they make the evidence stronger or easier to use.
Evaluate
Red-team and test models, agents, benchmarks, and AI-enabled workflows under realistic and adversarial conditions. Keep the evidence replayable.
Explore evaluation →Govern
Translate technical findings into named owners, operating controls, review gates, escalation paths, and decisions that client teams can use.
Explore governance →Assure
Re-run versioned tests and review controls as models, data, integrations, and operating conditions change. Keep decisions tied to current evidence.
Explore assurance →
Products
Products that extend the assurance loop
Syntony is developing three products that help teams carry evaluation evidence into governance and ongoing assurance. They are practical extensions of the same Evaluate → Govern → Assure work.
Governance Lag Monitor
Tracks whether frontier AI evaluation evidence leads to controls, release decisions, escalation, and follow-through.
Explore the Monitor →Resonance
Maps a customer’s AI system as testable attack paths, then connects evidence to controls, owners, decisions, and re-testing.
Open the risk map →GovTune AI
Carries red-team traces into findings, owners, controls, escalation logic, and decision records.
Explore GovTune AI →
Who we work with
Different institutions. Same need for evidence that holds up.
Syntony works with AI developers, companies deploying AI, government teams, research institutions, and civil-society organizations where evaluation evidence needs to withstand scrutiny and lead to action.
Founder-led
Nathan Heath
Nathan founded Syntony after repeatedly seeing strong evaluation work fail to reach the people making governance and release decisions. He is a decision scientist and AI safety researcher with fourteen years at the intersection of geopolitics and emerging technology.
He red-teams frontier models for OpenAI and Anthropic and co-founded The AIHL Project. Before Syntony, he spent six years as a decision scientist at National Security Innovations supporting U.S. Department of Defense clients on emerging-technology risk, AI integration, and security cooperation.
Nathan is a Truman National Security Fellow, advises the Cloud Security Alliance on catastrophic AI risk, and contributes to the Oxford Martin AI Governance Initiative. His work has been presented at IASEAI at UNESCO, UNIDIR, the Cambridge Centre for Geopolitics, and the UK MoD Deterrence and Assurance Academic Alliance, and has appeared in War on the Rocks, RAND Europe, World Politics Review, PRISM, and The Washington Post.
OpenAI & Anthropic Red TeamerTruman Security FellowCSA Expert AdvisorOxford Martin AI Governance
Contact
Bring us the system, finding, or decision.
If you need to test an AI system, turn existing findings into operational governance, or build a repeatable assurance loop, start there. We can bring in benchmark design, research, engineering, or workshops when the work calls for them. NDA discussions are standard.
Engagement boundary: We provide evaluation, governance, research, engineering, and decision support. We do not sell relationship access or guarantee decisions controlled by third parties.
Do not include classified information, CUI, export-controlled data, credentials, client evidence, or other sensitive system details in public email or scheduling fields. We will establish an appropriate channel before evidence transfer.