Delegating to AI Without Losing Professional Accountability: The OBO Chain
Delegating to AI Without Losing Professional Accountability: The OBO Chain
Category: Legal & Professional Services | Reading time: 6 min
The Accountability Problem That Keeps Partners Awake
A senior partner at a law firm authorises their AI assistant to draft and send a routine contract renewal for a long-standing client. The AI does so — but includes a clause that was appropriate for a different client's jurisdiction, not this one. The client signs. The clause becomes a problem six months later.
Who is accountable?
The partner will say: "I authorised the renewal, not the specific clause." The client will say: "Your firm sent it." The bar association will say: "You delegated to an AI — you are responsible for its output."
This scenario plays out across professional services — law firms, accounting firms, consultancies — wherever licensed professionals use AI to extend their capacity. The delegation of professional judgment to an AI system creates a accountability gap that existing frameworks weren't designed to handle.
The On-Behalf-Of (OBO) chain, implemented in the Taos governance kernel, is the technical infrastructure that closes this gap.
What an OBO Chain Is
An OBO chain is a cryptographically signed record of every delegation of authority that led to an AI action. It answers the question "on whose authority did this happen?" at every step of an AI workflow.
For a contract review workflow initiated by a partner:
{
"initiator": "m.washington",
"role": "senior_partner",
"execution_id": "exec-contract-7821",
"chain": [
{
"principal": "m.washington",
"role": "senior_partner",
"action": "initiated_workflow",
"scope": "contract_renewal",
"client_matter": "CLIENT-441",
"timestamp": "2024-03-15T10:00:00Z"
},
{
"principal": "ai_agent",
"role": "document_processing_agent",
"action": "extracted_contract_clauses",
"authorised_by": "m.washington",
"policy_rule": "allow_contract_extraction",
"timestamp": "2024-03-15T10:00:01Z"
},
{
"principal": "ai_agent",
"action": "applied_jurisdiction_template",
"template_id": "UK-commercial-standard-v2",
"authorised_by": "m.washington",
"policy_rule": "allow_template_application",
"timestamp": "2024-03-15T10:00:02Z"
}
],
"chain_signature": "sha256:a3f2b1..."
}
The chain signature prevents any record from being altered after the fact. Every action the AI took is attributed to the authorising partner, through a verifiable delegation record.
Why This Matters for Professional Liability
Professional accountability for AI delegation has three dimensions:
1. Who authorised it? The OBO chain identifies the licensed professional who initiated the workflow. There is no ambiguity — the chain starts with a human principal, and every AI action is attributed back to that chain.
2. What were they authorised to delegate? The Rego policy defines the scope of each delegation. A junior associate can initiate routine contract extraction workflows; only partners can initiate workflows involving client-facing documents above a certain value or risk level. The policy encodes the firm's delegation rules.
3. Was the policy followed? Every AI action that executes was permitted by a policy rule evaluation. The rule that permitted each action is recorded in the OBO chain. If the AI did something it shouldn't have, the policy log shows either that a rule permitted it (and should be tightened) or that a policy control failed (and needs investigation).
Conflict of Interest: Policy-Enforced at Workflow Initiation
Law firms have a specific accountability requirement that AI systems must respect: conflict of interest checks. Before an AI can access any matter-related documents, the firm's conflict database must be consulted.
With Taos, this is a pre-condition rule:
package taos.legalops.matter_access
# Block any matter access if a conflict is on record
deny_conflict_of_interest {
input.conflict_check_status != "cleared"
}
deny_conflict_of_interest {
input.conflict_check_status == null
}
action = "deny" { deny_conflict_of_interest }
allow { not deny_conflict_of_interest }
No AI workflow can access matter documents without a conflict clearance. The check happens at the kernel layer — not in the LLM prompt, not in application code, but in the policy engine that the AI cannot bypass.
The Bar Association Scenario
Imagine your firm is audited by a bar association following a client complaint. The association asks:
- "Which AI actions were taken in relation to CLIENT-441?"
- "Who authorised those actions?"
- "Was the conflict check performed before any matter documents were accessed?"
- "Which policy controlled the AI's access to this matter?"
With the Taos OBO chain and audit log, you can answer all four questions from a single export:
- Every AI action on CLIENT-441 matters, timestamped
- The partner who initiated each workflow
- The conflict check status at the time of each workflow initiation
- The exact Rego rule that governed each access decision
This is the difference between a defensible response and an investigation that expands into a formal disciplinary matter.
Setting Scope Limits: What the AI Is Allowed to Do
Professional accountability requires that the AI's scope is explicitly bounded — not just logged after the fact, but enforced before execution.
In Taos, this is the allowed_tools capability in the ART (Ambient Runtime Token) issued at workflow start:
{
"principal": "m.washington",
"workflow": "contract_renewal",
"allowed_tools": [
"extract_contract_clauses",
"apply_standard_template",
"flag_non_standard_clauses"
],
"prohibited_tools": [
"send_to_client",
"execute_signature_request",
"modify_billing_record"
]
}
The AI can extract, apply templates, and flag issues. It cannot send documents to the client or execute signature requests — those require explicit human action. The prohibition is enforced at the kernel layer, not by prompt instruction.
The Engagement Letter: Making It Explicit
Some firms are now including AI governance language in their engagement letters:
"[Firm] may use AI-assisted document processing in the delivery of services. All AI-assisted actions are governed by our internal policy framework, are attributable to the supervising licensed professional, and are recorded in a tamper-evident audit log. No client-facing document or communication is sent without licensed professional review and approval."
The Taos OBO chain makes this statement technically accurate and provable. The "attributable to the supervising licensed professional" clause is backed by a cryptographic record, not an assertion.
The Bottom Line
AI delegation in professional services is not a question of whether to use AI — it's a question of whether you can account for what it does. The OBO chain gives licensed professionals the accountability infrastructure they need: every AI action attributed, every delegation recorded, every scope boundary enforced.
You can scale your practice with AI. You keep your accountability. These are not trade-offs.
Tags: OBO chain, legal AI, professional accountability, law firm AI, conflict of interest, delegation governance