AI evaluation, red teaming & governance

Know where
AI breaks

Decide what happens next.

We test AI systems, turn findings into practical controls, and help your team establish whether the response holds.

Work with Us Explore the evidence
From evidence to actionIllustrative study
Find the consequential path. An illustrative field of possible AI behaviors. A contrasting path exposes how an untrusted instruction can lead to privileged action. Evaluation isolates the path, governance places an approval boundary, and assurance replays the test. Untrusted instruction Privileged action Human approval Illustrative study. No client data.
Find the path that matters.Trace how an instruction in retrieved content reaches a privileged action. Keep the evidence that explains it.
Explore our experience
Selected experience & research

Frontier experience. Real-world impact.

Experience across leading AI organizations, public institutions, and international research venues.

Our team’s experience spans

OpenAI
Anthropic
Microsoft
Meta
Google
U.S. Navy
U.S. Department of War
Tripadvisor
TikTok
Scale AI
UC San Diego
UNESCO
MIT AI Alignment
Harvard Kennedy School
Stanford School of Engineering
Johns Hopkins School of Advanced International Studies (SAIS)
World Economic Forum
Central European University

Our work has been featured at

NeurIPS
UNIDIR
University of Cambridge
King's College London
University of Oxford
The American University of Paris
Wilfried Martens Centre
CERN

These references reflect the founder’s and collaborators’ professional experience, research, presentations, and affiliations. They do not imply endorsement, partnership, or client status.

The Evidence PackFrom a technical finding to a practical decision

A finding is
only the
beginning.

A red-team report can show you what went wrong. The harder work comes next: understanding why it matters, deciding who needs to act, and checking the response. We help your technical and governance teams work through that together.

The Syntony Evidence Pack keeps the question, evidence, uncertainty, ownership, and next test in one record.

What an engagement delivers
Syntony Evidence RecordExample 01
Illustrative example. No client data.

When a document becomes an instruction.

01

Question & intended use

Can an agent use retrieved content without treating it as an instruction to take privileged action?

02

Evidence & finding

A hostile instruction in a retrieved document leads to an attempted privileged tool action.

Evidence included

Trace, prompt, tool call, affected privilege, and replayable test case.

Retrieved content → agent instruction → tool attempt
Input

A document enters the retrieval context.

Boundary

External content is treated as an instruction.

Action

A privileged tool action is requested.

03

Uncertainty & limits

One failure path, under defined conditions. Other inputs, permissions, and system versions require further testing.

04

Owner & proposed control

System security owns remediation; the program owner sets release conditions. Apply content isolation, tool-call approval, and a retrieval allowlist.

05

Decision

Restrict the affected tool capability until the proposed controls are verified.

06

Re-test plan

Replay the scenario after remediation and material changes to the model, tools, or retrieval layer.

Evidence, ownership and follow-through.
How we work with youExplore our expertise ↗

Start with the
decision ahead.

A release. A difficult finding. A system that has changed. We agree on the question, then build the evidence you need to answer it.

01
Red teaming & evaluation

Before you release.

Find the failure paths ↗

Test your model, agent, or workflow against realistic and adversarial conditions.

What you take away

An evaluation protocol, documented findings, and tests your team can run again.

02
Governance & decision support

When the finding
needs a response.

Make the response practical ↗

Turn technical evidence into responsibilities, controls, and operating decisions.

What you take away

Clear ownership, review gates, and a decision record your organization can use.

03
Continuous evaluation & assurance

As your system evolves.

Check whether the fix holds ↗

Revisit the evidence as models, data, tools, and operating conditions change.

What you take away

Versioned regression tests, control reviews, and an updated evidence record.

BLACKLINE · THE CLOSE THE LOOP GAME

Test the boundary.
Follow the consequences.

Choose one of nine original AI security operators and take on two ten-level campaigns: business and government.

Find the failure, assign responsibility, and prove the response in replay. Practice the decisions behind Syntony’s human-led evaluation and governance work, then explore Resonance and the GovTune AI waitlist.

Play BLACKLINE Original operators · 20 missions · Decisions with consequences
From the research benchRead our field notes ↗

Built around
the hard questions.

Our tools explore how findings, decisions, and follow-up stay connected. All three are in development.

Observe In development

Governance Lag Monitor ↗

Tracks whether public frontier-AI findings lead to documented controls or decisions, and how long that response takes.

 Map In development · Public preview

Resonance ↗

Maps attack paths through an AI system, connecting each path to the test record, responsible team, response, and next test.

 Connect In development · Join the waitlist

GovTune AI ↗

Drafts governance records from red-team traces to help teams assign responsibility and plan mitigations while preserving human decision workflows.
Who we work withThe context changes the work

Different institutions.
Real decisions.

We work with AI labs, companies, public agencies, universities, and civil-society organizations. The questions, responsibilities, and operating constraints differ. We shape the work around yours.

See how the work adapts
01 AI labs & research teams 02 Companies & product teams 03 Public agencies 04 Universities 05 Civil society
The people behind the workMeet the team
ContactResearch Triangle Park, NC · Working globally

What do you
need to know?

Tell us what you’re building, what you’ve found, or what comes next.

Work with Us hello@syntonyresearch.ai ↗
Our first conversation
  1. 01 The question you need to answer.
  2. 02 The evidence that would help.
  3. 03 A useful scope of work.

We can begin under an NDA.

Phone828-418-7587 ↗LinkedInSyntony Research ↗
Global headquarters600 Park Offices Drive, Suite 374
Research Triangle Park, NC 27713 ↗

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.

Syntony provides evaluation, governance, research, engineering, and decision support. Clients retain responsibility for external relationships and decisions controlled by third parties.

Global headquarters Research Triangle Park, North Carolina

Our home in
the Triangle.

Rooted in Research Triangle Park.
Working with organizations around the world.

Preparing the map…

One headquarters. A global outlook.Explore the map, or get directions.