Platform · Audit
Every AI action, on the record.
Who asked, from which AI, what ran, with which inputs, and what happened, for every action.
Record
“What did the AI do?” deserves an answer.
Sooner or later someone asks why a record changed, or who sent that email. Every action that goes through Bazinga leaves the same line, allowed, refused, or failed, by a person’s assistant or by an agent, so the answer is one search away.
Audit trail
| Time | Who | Assistant | Action | Result |
|---|---|---|---|---|
| 10:42 | Emma Clark | Claude | Update invoiceInvoice 1043 · Open → Paid | Done |
| 10:39 | Sales ops agent | Copilot Studio | Onboard new customerLakeside Logistics · starts Monday | Done |
| 10:31 | James Chen | ChatGPT | Look up dealsClosed this month · over 50k | 14 records |
| 10:27 | Ana Moreno | Claude | Delete recordInvoice 0988 | Denied |
| 10:21 | Purchasing agent | n8n | Create purchase orderHarbor Paper · ERP didn’t respond | Failed |
Who.
The person, or the agent and the account its key runs as.
From where.
Which assistant made the request: Claude, ChatGPT, Copilot Studio, or one of your agents.
What.
The action, and the inputs it ran with, visible to administrators only.
The result.
What happened: done, refused by permissions, or failed.
Investigate
Find any action in seconds.
Filter to the question.
By person, agent, assistant, action, or time, to answer “what happened here?” in seconds.
Refusals are recorded too.
A pattern of denied requests shows an agent reaching for what it shouldn’t, before anything goes wrong.
Back to the person.
Records kept in Bazinga stamp who created and last changed them. A connected system sees Bazinga’s account, and the record here says who asked.
Change
Know what AI did. And who changed what it can do.
A tool can quietly do more today than it did yesterday. Say someone changes the workflow behind “Update invoice” so it also changes the bank account: you see who changed it, when, and the reason they gave, and can bring back yesterday’s version of the workflow.
Execution history
What AI did.
Every request: who asked, from which assistant, what ran, with which inputs, and what happened.
Change history
Who changed what AI can do.
Changes to the workflows behind your tools are tracked as builders make them, and committed to a Git repository you own, on GitHub for example, with an author and a reason. An earlier version of the workflow can be restored in one step.
Keep
Kept, and yours to take.
Kept by plan.
How long the record stays searchable in Bazinga depends on your plan.
Exported when you need it.
Download it for an auditor or an investigation, or to keep it longer in the security tools your team already runs.
Read by people, not just machines.
Plain names and plain actions, so the record answers questions without an engineer to translate it.
Contain
Unusual activity, contained.
A record tells you what happened. Limits keep a runaway agent from doing it a thousand times.
Rate limits.
For each person and each assistant, so a runaway agent can only do so much.
Changes flagged.
Actions that change data are flagged, so assistants can ask the person before running them.
Disconnected in one step.
Revoke an assistant or a key, and its next request is refused.
Prove
What to show your auditor.
Three answers to start the review with.
What AI can do.
Every exposed data set, operation, and workflow, and the roles that may use them.
What AI did.
The record of every action through Bazinga, filtered to the period and the systems under review.
The platform underneath.
Plant an App’s security practices are audited under SOC 2 Type II and certified to ISO 27001.
Related capabilities
Ask what the AI did.Get an answer in seconds.
See the record Bazinga keeps for every action, and what it looks like when an agent reaches for something it shouldn’t.