An agent that is already live can be talked into doing the thing it was told to refuse. The test is that conversation, on purpose, against one agent you own.
The fee is $8,000. Current rate. We confirm it on the invoice before any work starts. The figure can change until that invoice is sent. One public surface or one role. A second agent, or an agent with many tools and other agents behind it, uses Software & AI Automation terms: $10,000 to open, then $220 an hour.
Pricing
- Current rate
- $8,000
Current rate. We confirm it on the invoice before any work starts. The figure can change until that invoice is sent. A wider surface uses Software & AI Automation terms. API usage and any cost associated with the work is billed through. We do not mark it up as a surprise line — it is named in the proposal.
What the engagement covers
Prompt injection, jailbreaks, tool misuse, and the data the agent gives up that it should have refused. The packet names what we tried, what we found, and the step that closes it. One retest of those fixes is included.
Model usage during the test follows the same rule as the rest of the practice. API usage and any cost associated with the work is billed through. We do not mark it up as a surprise line — it is named in the proposal.
What stays where it already is
A website or a piece of software is not restated here. From the public internet, the external exposure report is the look. With a login or with the source, software assurance is the look. Agents this practice ships already pass adversarial testing before production. This engagement is for an agent that is already live.
Nothing probes until the target is confirmed yours, in writing. We do not promise the agent is clean.