Practice planning and running an AI red team exercise. Learn to turn an observed failure into a clear finding, a proposed control, and a retest plan.
Three hours of guided practice in Syntony’s red teaming portal, with discussion led by practitioners in AI security and governance.
December 1, 2026 · 1–4 p.m. Eastern
Live online · $795 USD per person
U.S. Department ofOrganizations reflect professional experience and affiliations, as described in the speaker bios. Organizations do not sponsor or endorse this workshop.
Work through a guided AI red teaming exercise and build a record of your own findings. The session connects attack techniques, systems thinking, and practical governance.
Define the workflow, the threat, the people who could be affected, and what would count as a failure.
Practice multi-turn and multilingual attacks, authority claims, and indirect prompt injection. Adjust your approach to the system’s responses.
Look beyond the model response to tool permissions, documents, human approvals, and downstream effects.
Record prompts, conditions, observed behavior, consequences, and uncertainty so someone else can understand and repeat the test.
Propose a control, identify who should own it, and define what evidence would show that the response works.
Finding a failure is the start of the work. Syntony teaches you to connect evidence to human impact, accountable decisions, and a check that can be repeated.
Work in Syntony’s proprietary red teaming portal, compare approaches, and discuss the findings with the workshop team.
Build a conversation over multiple turns. Change your next prompt in response to the agent. Combine language switching, false claims of authority, and instructions in retrieved documents.
Compare the source, summary, approval, and action. Test what the reviewer could see and stop. Use interconnected risk mapping to trace how one failure affects the next team or system.
Use the finding to decide what must change before deployment. Consider tool permissions, document templates, and approval rules. Choose an owner for the fix and a way to test it again.
Red teaming and governance belong in the same exercise. We connect what failed, who could be affected, and who has the authority to change it. That connection is how testing becomes useful to the organization.
You’ll build three connected work products during the exercises: a risk map, a reproducible test record, and a decision and retest plan. Take the format back to your next evaluation or deployment review.
The people, systems, and decisions involved—and where a failure could spread.
Your prompts, test conditions, observed behavior, and the evidence behind the finding.
A proposed fix, a responsible owner, the remaining uncertainty, and what to check next.
Supplier assessment → AI summary → deployment review
Nathan Heath leads the workshop, joined by Susan Nesbitt and Chad Ratashak. Together, they bring experience in frontier-model red teaming, enterprise governance, intelligence, and financial crime.
Founder, Syntony
Nathan connects adversarial testing with systems analysis and governance. He brings more than 14 years across AI security, decision science, and national security to the way Syntony tests systems and processes.
Nathan founded Syntony to help organizations understand how AI systems fail and use that evidence to act. His experience includes the U.S. Department of War, U.S. Department of State, Harvard Law School, the Hudson Institute, and technology startups.
He is a Truman National Security Fellow and an Expert Advisor to the Cloud Security Alliance on catastrophic AI risk. He holds an M.A. in Law and Diplomacy from The Fletcher School.
Full bio
Managing Director, Cantellus Group
Susan advises governments, technology companies, and international organizations on agentic AI strategy, governance, and responsible deployment. She helps leaders move from experimentation to enterprise adoption.
Susan serves on the IASEAI Policy Committee, advises AI Safety Connect, and contributes to international initiatives with UNESCO on AI governance and the ethical use of AI.
At the World Economic Forum, she built global partnerships at the intersection of technology, policy, and societal impact.
Owner, Midwest Frontier AI Consulting
Chad brings counterterrorism and anti-money-laundering analysis to AI risk. His work spans model red teaming, responsible AI use in legal practice, and AI x Bio threats.
In 2022, Chad was the Private Sector Leader for Ransomware Attacks on Critical Infrastructure in the U.S. Department of Homeland Security’s Public-Private Analytic Exchange Program.
He has spoken to professional organizations across law, financial crime, and federal contracting, including ACI, NBI, ACAMS, ACFCS, and Govology.
Follow the same workflow from the first risk map to a deployment recommendation. The live session combines practice, comparison of results, and discussion with the speakers.
December 1, 2026
1–4 p.m. Eastern (EST) · Live online
Identify the people, agents, data, and tools involved. Set the test boundaries and define what counts as a failure.
You’ll produce: a risk map and test plan.
Adapt your prompts to the agent’s responses. Combine language changes, authority claims, and instructions in retrieved content. Check meaning with fluent human review and track actual tool behavior.
You’ll produce: a record of your prompts, results, and test conditions.
Trace missing details and unsupported assumptions. Check what reaches the reviewer, who can stop an action, and whether approval matches the agent’s permissions.
You’ll produce: a finding that locates the failure in the workflow.
Explain the consequences, compare fixes, and state what remains uncertain. Decide who needs to act and what evidence would justify deployment.
You’ll produce: a recommendation, proposed owner, and retest plan.
A supplier’s agent was tested with read-only access. By the time it goes live, it can change records. How did that happen?
“Controls tested with read-only access.”
A clear testing limitThe evidence covers reading records. It says nothing about changing them.
“The supplier’s controls have been tested.”
The limit disappearsA qualified result becomes a broad assurance. The reviewer never sees the missing detail.
Approved for
read + write access.
No one compares the requested permissions with the conditions used in testing.
Can now
change records.
A document error has become a change in what the system is allowed to do.
This can happen through an ordinary mistake—or an attacker can try to cause it with a misleading document and an adaptive conversation.
You’ll test both. Trace what changed, check whether anyone could have caught it, and decide what evidence is needed before the agent gets those permissions.
This is a guided practitioner workshop. Come familiar with LLM applications, prompt injection, agent tools, and basic security concepts. For a foundations course or a deeper technical programme for your team, explore custom training.
A risk map, a record of your adversarial tests, and a finding linked to a proposed control, owner, and retest. These are working outputs from the guided exercise that you can develop further for your own evaluations.
SOC 2 provides assurance about the controls covered by the examination, which can include AI. This workshop gives you practice testing a specific workflow: whether warnings survive, approvals cover the next action, and a reviewer can catch the gap. You’ll connect that evidence to your risk reviews.
AICPA guidance on AI and SOC examinations ↗You’ll practice adapting an attack as a conversation develops, then investigate how the outcome depends on documents, permissions, and human review. The exercise connects the observed behavior to a proposed fix, owner, and retest plan.
The three-hour live online workshop, practice in Syntony’s proprietary portal during the session, discussion with the workshop team, and the Evidence Pack you build. Setup and access details will be sent before the workshop.
Tuesday, December 1, 2026, 1–4 p.m. Eastern (EST, UTC−5). Standard registration is $795 USD per person. Use ATTUNE through November 15, 2026 for 15% off ($675.75).
We use research on human-led multi-turn attacks, multilingual testing, and agent security. We’ll discuss what the researchers tested, what they found, and where those results may not apply to your system.
Tuesday, December 1, 2026 · 1–4 p.m. Eastern (EST). Live online. Register securely through Stripe. We’ll email your workshop access and preparation details.
$795USD / person
Includes the live workshop, practice in our portal, group discussion, and the Syntony Evidence Pack you build during the workshop.
Register and pay$675.75USD / person · Save $119.25
Use code ATTUNE through .
Register with ATTUNEUse ATTUNE at checkout for 15% off through 11:59 p.m. Eastern Time on November 15, 2026. Your final total is shown before payment.
Optional: share your background and interests to help us prepare for the workshop. Use the same email address as your registration.
Customize the audience, use cases, technical depth, format, and duration. Combine red teaming with security, governance, or incident response.