01 / DEFENSE FOR AI
Let agents do more.
Keep control of
what happens.
Help employees delegate safely. Give security teams company-wide policy, connected evidence, and focused investigations.
Explore Defense for AISafety for autonomy. Intelligence for defense.
Safeguard the work AI does. Give your defenses the power of AI.
Protect your business from mistakes, misuse, and unintended actions. Strengthen your defenses with autonomous discovery and verified remediation.
Research roots
Built on UC San Diegosystems research
Venture backed
Read why Two Small Fishinvested in Gensee
Open-source partners
Two directions. One mission.
Make autonomy
work for your business.
Gensee builds AI agent safety and security tools for enterprises: Defense for AI safeguards employee and agent work. AI for Defense brings autonomous discovery and verified remediation to cyber defense.
Let agents do more.
02 / AI FOR DEFENSE
Find the weakness.
Prove the fix.
Bring autonomous discovery, repair, and verification to existing software and agent-generated code.
Seeking design partners
Explore AI for DefenseBuilt for both sides of the business
Keep work moving.
Keep security in control.
For employees & engineering
Room to do
your best work.
Delegate the ambitious task. Explore another approach. Keep routine work moving inside clear company boundaries.
- Fewer unnecessary interruptions and approval requests
- Isolated workspaces for exploration and experimentation
- Clear explanations when an action needs attention
For security teams
Clarity to make
the right call.
Set the boundaries, connect the evidence, and focus review on consequential exceptions. Give your team context to act.
- Reusable company policies and visible coverage
- Related findings grouped into focused investigations
- Tiered screening designed to reduce repetitive triage
Inside Gensee
From an agent’s action
to an informed decision.
Follow one finance-file change through the actual product. Every screen has a job to do.
Scroll to follow the investigation
1
2
3
01 / Activity
See the work around the exception.
-
1
The reported action
See what the harness reported, including its stated outcome.
-
2
The affected resource
A finance-file change sits alongside ordinary project work.
-
3
The evidence basis
Source labels distinguish runtime reports from independent OS observations.
1
2
3
02 / Policy & alert
Understand why the action matters.
-
1
The matched policy
The finding names the company condition and policy revision.
-
2
The action and target
Inspect the reported write to forecast.csv and its source context.
-
3
The next decision
Carry the finding and its supporting evidence into an investigation.
1
2
3
03 / Investigation
Keep the evidence connected.
-
1
The original finding
The saved case links back to the original detection.
-
2
The supporting action
Review the resource, reported outcome, and matched condition together.
-
3
The case history
Preserve a record of changes and assessments as the investigation develops.
1
2
3
04 / Employee context
Give both sides a voice.
-
1
A focused question
Ask about the relevant action and task, with supporting context.
-
2
The employee explanation
Retain the response alongside the investigation evidence.
-
3
The analyst assessment
Document the decision. An explanation does not automatically change policy.
Actual product · synthetic demonstration. The console and evidence workflow are real; source reports and the employee explanation are simulated. These screens demonstrate investigation, not preventive blocking.
Watch the captioned walkthrough
Silent walkthrough with on-screen captions. The annotated screens above provide a text alternative for the workflow.
Where it matters
Real work.
Real consequences.
Human mistakes, agent errors, and malicious inputs can meet in the same workflow. Build protection around the work itself.
A shared foundation
Autonomy needs
an architecture.
The same discipline that governs an agent’s work can govern an agent’s defense: understand the action, bound its authority, verify the result.
Connect actions
to their evidence.
Bring together supported agent reports, system observations, policy context, and outcomes. Keep the source and its limits visible.
Give the operation
the authority it needs.
Scope resources and temporary permissions. Isolate exploration and control consequential changes at the relevant boundary.
Make the result
earn your trust.
Check the exact output, inspect the evidence, and promote accepted changes. Keep a record of what was authorized and what happened.
Built on systems research
Look beneath the interface.
Founded by UC San Diego professor Yiying Zhang and former Google engineering leader Shengqi Zhu, Gensee brings systems research to agent safety. Meet the team
TRANSACTIONAL WORKSPACES
Explore, merge, promote.
When an agent finds
Explore, merge, promote.
Know what rollback means.
How live workspace branches return useful results—and where their boundaries end.
Read the research ↗ LONG-HORIZON AGENT SAFETYWhen an agent finds
a way around the boundary.
What real incidents teach us about shared services, persistent state, and control over time.
Read the analysis ↗Before we get started
A little more clarity.
Answers for the people doing the work—and the people protecting it.
What does Gensee help protect?
Gensee helps businesses manage the safety and security of work delegated to AI agents: human mistakes, agent errors, malicious instructions, unintended changes, and company data crossing the wrong boundary. Defense for AI brings together company policy, connected evidence, and focused investigations. Explore the customer scenarios.
How are Defense for AI and AI for Defense different?
Defense for AI helps employees use agents safely and gives security teams control over that work. AI for Defense is our enterprise design-partner initiative for discovering vulnerabilities, validating findings, and preparing and verifying fixes. Both serve enterprise teams; Gensee Crate Personal serves individual developers.
Will employees need approval for every action?
No. The approach is to let routine work proceed within reusable company policies and reserve review for consequential exceptions. Policy enforcement, security alerts, and employee notifications are separate choices. An evaluation should measure both unnecessary interruptions and missed issues against your own workflows.
How does tiered screening help security teams?
First, screen incoming findings for likely noise using policy and task context. Next, enrich the remaining findings with related actions, evidence, and potential impact. Finally, send consequential exceptions to a human reviewer with the context needed to decide. High-risk evidence should not be dismissed on model confidence alone. See the screening approach.
Which agents and environments can we evaluate?
Start with the tools your employees already use. The current demonstration includes the company console and managed Mac evidence workflows, with supported Claude Code and Cowork sources. Coverage and enforcement depend on the integration, platform, and configured release. We scope an evaluation around your actual environment and acceptance criteria.
Can Gensee undo an agent’s mistakes?
Supported isolated workspaces let developers explore alternatives, review changes, and discard unwanted state. Rollback cannot unsend an email, recall leaked information, or reverse arbitrary external transactions. External effects need their own controls. Read how workspace rollback works.
How do we get started, and how does pricing work?
Tell us which product line interests you and what you want to evaluate. Defense for AI uses custom pricing based on deployment and integration needs. AI for Defense is seeking design partners, and Personal is free. Compare the options or contact us to discuss a demo, partnership, or another question.
Still have a question?
Tell us what your team is working on.
Let’s find the right place to start.
Let’s put it to work
Put AI to work.
Keep your business in control.
See Gensee in action, or help shape what comes next. Start with the work your people and agents actually do.