Priced per engagement, agreed before we start
Three ways to work with us. Each is a fixed fee for a scope we agree in writing, and most teams start with one agent and widen from there.
Fixed fee, 3-4 weeks
Assessment
Find out what your agents can do
An adversarial assessment of one agent or a small fleet, across permissions, tool use, prompt-injection exposure and data leakage.
- Inventory workshop and agent threat model
- Hands-on adversarial testing across all four domains
- A prompt-injection corpus built from your own untrusted inputs
- Ranked findings with reproductions and evidence
- An executive briefing plus a technical appendix
- A sequenced remediation plan your engineers can act on
Fixed fee, 5-8 weeks
Assess + remediate
Close the findings, then prove it
The assessment, plus hands-on remediation with your engineers and a verification re-test that shows each attack failing.
- Everything in the Assessment
- Fixes implemented as pull requests and configuration in your repos
- Least-privilege identities and just-in-time credentials for agents
- Tool guardrails, approval gates and egress controls
- A verification re-test showing the before and after for each finding
- The regression attack suite handed over, plus a team enablement session
Annual, built on your baseline
Continuous assurance
Keep it true as the agents change
Ongoing monitoring and testing so a tool added on Monday does not quietly undo the work.
- Everything in Assess + remediate
- Live agent, tool and data-source inventory with drift alerts
- Your attack suite running as a gate in CI
- Runtime anomaly signal on tool chains and egress
- A quarterly red-team refresh as new attack classes appear
- Evidence exports for customer security reviews and auditors
Not sure which one you need?
Start with a 30 minute scoping call. No charge and no obligation. We walk through your agent architecture and tell you where the exposure probably sits.
What each engagement covers
| Capability | Assessment | Assess + remediate | Continuous assurance |
|---|---|---|---|
| Agent, tool and data inventory | |||
| Adversarial testing across four domains | |||
| Ranked findings with reproductions | |||
| Remediation plan | |||
| Fixes implemented with your engineers | |||
| Verification re-test | |||
| Regression suite handed over | |||
| Drift alerts on new tools and scopes | |||
| Attack suite running in CI | |||
| Runtime anomaly signal | |||
| Quarterly red-team refresh | |||
| Evidence exports for audits | On request |
How an engagement gets priced
So the number on your proposal is not a surprise.
How many agents, and how coupled
One customer-facing agent is a different job from twelve that call each other. We scope by agent and by the tool surface behind it.
The size of the tool surface
Ten tools with tight schemas take less time than sixty tools, three MCP servers and a code interpreter.
Which environment we test
A faithful staging replica is the fastest path. Testing in production takes longer because it runs under tighter rules of engagement.
How far remediation goes
Advisory support costs less than our engineers writing, reviewing and shipping the fixes alongside yours.
Evidence requirements
If findings feed a customer security review, an audit or a regulatory filing, expect extra time on documentation.
Common questions
Get a scope and a number
Tell us what your agents do and what they can reach. You get a written scope and a fixed fee, usually within a few days of the call.