Name what may happen.
The service, workflow, purpose, tools, data, people, jurisdictions, dates, and excluded activity.
AI assurance for general counsel
Turn proposed AI use into a decision counsel can review.
Petrichor assembles the boundary, current evidence, named human authority, conditions, exclusions, expiry, and change triggers for one proposed agency use. Counsel keeps the decision.
$25,000 fixed fee Four weeks One account Up to three workflows
The first message requires no privileged advice, client name, confidential material, or evidence file.
The decision object
Counsel needs a bounded use, not a company-wide adjective.
Responsible, safe, and compliant are not reviewable answers. A permission decision needs a named provider, proposed use, current evidence, explicit exclusions, and retained client authority.
Show what supports it today.
Evidence, gaps, exceptions, owners, review dates, and the operating period behind each material assertion.
Keep the client’s right to decide.
Permission, restrictions, escalation, expiry, refusal, and reopening remain with the accountable client owners.
Representative decision packet
Put the conditions of yes beside the reasons for no.
This fictional structure shows the work product. It is not a client record, certificate, legal opinion, or evidence that a decision has been approved.
AI use permission record / Demonstration only
Fictional structureNo client data
Decision under review
May the agency use an approved enterprise model to develop campaign concepts from cleared, non-confidential source material?
Conditions that could support permission
- Inputs
- Only cleared source material. No client strategy, audience data, unreleased assets, or personal data.
- Human authority
- Named creative lead selects concepts. Named reviewer substantiates claims. Client retains final approval.
- Output boundary
- No generated concept becomes a final asset without rights, factual, brand, and disclosure review.
Evidence still required
- Vendor treatment
- Current enterprise terms, retention state, training treatment, access, and material-change monitoring.
- Rights record
- Source authorization, human contribution, similarity review, and third-party rights escalation path.
- Exception path
- Owner, response time, stop-work condition, notification duty, and decision-reopening rule.
Evidence groups
Inspect the facts that change the answer.
The packet keeps missing evidence visible. It does not convert an absent record into a favorable assumption.
01Scope and partiesWho is asking, what work is covered, and where the answer applies.
- Provider legal entity, client or SOW boundary, service, workflow, dates, jurisdictions, and exclusions.
- Named agency executive, operating owner, client decision owner, and escalation path.
02Authorized inputsWhat may enter the workflow and under which authority.
- Client instructions, data classes, brand assets, source material, licenses, permissions, and contract limits.
- Restricted information, prohibited inputs, approved accounts, and access conditions.
03Vendors and data treatmentWhich tools and suppliers participate, and what happens to submitted material.
- Model, platform, partner, and subprocessor record with material terms and configuration state.
- Retention, training treatment, access, location, monitoring, and supplier-change triggers.
04Human decisions and substantiationWhich consequential choices remain with named people.
- Who creates, selects, rejects, substantiates, clears, approves, releases, and responds to exceptions.
- Current review records for factual claims, brand decisions, quality, and task-specific performance.
05Rights and disclosureWhat was reviewed before an output can move forward.
- Human contribution, provenance, similarity, trademark, likeness, music, talent, and license records.
- Disclosure, ownership, registration, reuse, release, and qualified-counsel escalation decisions.
Decision states
A defensible no is as valid as a bounded yes.
The engagement makes the current decision state explicit. It does not optimize for approval.
Permit
The evidence supports the proposed use inside the named boundary and review period.
Permit with conditions
The use may proceed only with recorded restrictions, owners, monitoring, and expiry.
Defer pending evidence
The decision remains open because a material fact, control, or review record is missing.
Prohibit
The proposed use does not fit the client instruction, risk boundary, or available evidence.
Every decision names its owner, evidence date, conditions, expiry, notification duties, and material-change triggers.
Account Diagnostic
One account. Up to three workflows. Four weeks.
Use the engagement when a live client decision is blocked by unclear scope, fragmented evidence, or unnamed authority.
- 01
Decision boundary
The provider, account, service, workflows, uses, tools, data, partners, dates, jurisdictions, and exclusions.
- 02
Evidence schedule
The current records available for each material assertion, plus gaps, owners, and review state.
- 03
Human authority record
Who makes the creative, factual, legal, claims, brand, exception, and final release decisions.
- 04
Permission conditions
Permitted, conditional, deferred, and prohibited uses with expiry and notification duties.
- 05
Decision meeting
A structured review with the accountable business and legal owners. The client retains the decision.
Review one proposed AI use
No client name, privileged advice, confidential material, credentials, or evidence files in the first message.
Fit and boundaries
Built for a live permission decision, not a generic AI opinion.
Strong fit
- An agency or provider has proposed a specific AI-enabled use.
- The use touches client work, data, assets, claims, rights, or release decisions.
- The business needs a reusable evidence record, not another broad questionnaire.
- Accountable business and legal owners can review the facts and retain authority.
Not this engagement
- A legal opinion, privilege determination, contract interpretation, or rights-clearance conclusion.
- A certification, endorsement, safe-harbor claim, or approval guarantee.
- A company-wide promise that every AI use, model, output, or employee is covered.
- A request to make missing evidence, unresolved exceptions, or client restrictions disappear.
01Is this legal advice or an outside-counsel opinion?
No. Petrichor structures the operating boundary, evidence, responsibilities, conditions, and decision record. Qualified counsel remains responsible for legal advice and legal conclusions.
02Does the packet approve the agency or certify its AI use?
No. The packet supports one client decision about one bounded proposed use. It is readiness and implementation work, not certification, endorsement, preapproval, or a compliance guarantee.
03Does counsel give up decision authority?
No. The client keeps final authority, evidence requests, restrictions, escalation rights, and the right to refuse, pause, or reopen the use.
04What causes the decision to reopen?
A material change in the proposed use, model, supplier, terms, data class, partner, jurisdiction, client instruction, evidence state, incident, or exception can reopen the decision.
05What should we send in the first message?
Only a non-confidential summary of the service or use, your role, organization, desired decision date, and main open question. Do not send privileged advice, client material, credentials, or evidence files.
Agency leadership track
Need the commercial and operating case for the agency? See AI Assurance for advertising agencies.
Start with one use
Bring one permission question. Leave with the evidence path.
We will identify the decision owner, define the boundary, expose the missing evidence, and determine whether a four-week diagnostic is warranted.