One company thought it had around 100 AI agents operating internally. When it looked more closely, it discovered approximately 1,500.
That example, shared during Open Future Forum's inaugural General Counsel Executive Forum on October 30, captured one of the central problems facing corporate legal teams: AI adoption is moving faster than many organizations can govern it.
The private lunch in San Francisco brought together General Counsel, Chief Legal Officers, technology executives, legal AI specialists and insurance professionals to discuss what happens when AI systems begin taking actions on behalf of businesses.
The conversation covered enterprise AI governance, liability, insurance, cybersecurity and the changing relationship between in-house lawyers and outside counsel.
Here are six lessons from the discussion.
1. You cannot govern agents you don't know exist
AI agents are appearing across different parts of the enterprise.
Marketing teams can build agents using Microsoft Copilot Studio. Engineering teams can deploy them through AWS Bedrock or Azure. Employees may also connect AI tools to business applications without centralized oversight.
Unlike a conventional chatbot, an AI agent can use software tools to retrieve information, update records, send communications and carry out tasks.
That creates a practical problem for General Counsel.
Does the company know which agents are operating, who authorized them, what information they can access and which actions they can perform?
A traditional software inventory may not provide the answer.
The first step toward meaningful governance is establishing an agent inventory with clear ownership, permissions and accountability.
2. Treat enterprise agents more like employees than software licenses
One useful analogy from the discussion was to think of an AI agent as a new employee.
An employee has an identity, a manager, a job description and access permissions. Those permissions can be changed or revoked.
Enterprise agents need similar controls.
Participants discussed organizations assigning unique identities to agents rather than allowing them to operate through employee accounts.
This makes it easier to determine which systems an agent accessed, which actions it performed and who was responsible for its deployment.
The discussion also covered the importance of a kill switch: a mechanism to suspend an agent when its behavior becomes unsafe or unauthorized.
Instructions alone are not sufficient. An agent told not to access sensitive information may still make mistakes.
Where possible, permissions and restrictions should be enforced by the surrounding software infrastructure rather than relying solely on the AI model to follow instructions.
3. An AI policy is not evidence of what an agent did
Most large companies already have policies governing technology, cybersecurity and data access.
AI introduces an additional challenge.
An organization may have approved an agent, tested it before deployment and documented the intended controls. But can it reconstruct what that agent actually did three months later?
That distinction matters in litigation, regulatory investigations and insurance claims.
An effective audit trail may need to record the agent's identity, instructions, tool calls, permissions, accessed resources, actions and human approvals.
It is also important to distinguish an audit of observable activity from an audit of the model's internal reasoning. The latter may not be fully accessible.
The discussion highlighted the limitations of relying on annual reviews for systems that can change behavior after deployment.
Predeployment testing remains necessary, but organizations also need ongoing monitoring and defined procedures for responding to unexpected activity.
For General Counsel, the practical question is whether the company can demonstrate that its controls operated as intended.
4. Liability and insurance are becoming harder questions
Who is responsible when an AI agent causes a loss?
The answer may involve the company deploying the agent, the technology vendor, contractual allocations of responsibility and existing principles of negligence, agency or product liability.
The Air Canada chatbot dispute provides an early illustration of the issue. In 2024, a British Columbia tribunal held the airline responsible for incorrect information supplied through its chatbot.
More autonomous agents raise additional questions because they can take actions rather than simply provide information.
Participants discussed how vendors and enterprise customers are beginning to negotiate responsibility for agent behavior, including indemnification, audit rights and limitations of liability.
Insurance is another developing area.
Traditional cyber and technology liability policies may cover certain AI-related losses, but the wording matters. Dedicated AI liability products are also emerging.
The discussion emphasized the importance of understanding coverage before an incident occurs.
For General Counsel, that means working with risk management, security teams, brokers and insurers to identify potential gaps.
A company's ability to demonstrate effective governance may also become relevant to how insurers assess its risk.
5. AI is changing how in-house teams work with outside counsel
The conversation moved beyond governance to the economics of legal work.
AI tools such as Harvey are allowing legal departments to handle more initial contract review, due diligence, research and policy analysis internally.
A legal team can review a collection of contracts against its playbook, identify exceptions and prepare draft analysis before involving outside counsel.
The law firm can then concentrate on the issues requiring specialist expertise and judgment.
Shared workspaces are also changing collaboration between internal and external lawyers, allowing both sides to work with common documents and review outputs.
This raises questions about the traditional billable-hour model.
If AI reduces the time required for routine legal work, clients will increasingly want to understand how that affects fees.
But lower costs are not the only potential benefit.
AI may also allow legal departments to obtain deeper analysis, faster responses and broader coverage of legal issues within existing budgets.
Several participants emphasized that AI adoption has not necessarily translated into reduced legal headcount. Instead, it is changing how lawyers spend their time.
6. The next generation of lawyers will need to learn differently
One of the more interesting discussions concerned junior lawyers.
Traditionally, associates developed their judgment through research, document review, drafting and negotiations.
AI can now perform significant portions of that work.
If junior lawyers spend less time on foundational tasks, how will they acquire the experience needed to become senior advisers?
Participants considered a growing role for lawyers who can build, maintain and supervise legal AI workflows.
That includes developing legal playbooks, validating outputs, maintaining precedents and understanding how AI systems reach their conclusions.
There was also an optimistic view.
AI can explain why a contract provision matters, identify departures from a company's usual position and help junior lawyers understand the reasoning behind legal decisions.
Used well, it could improve training rather than undermine it.
The challenge for law firms and legal departments will be preserving the development of legal judgment while changing the work through which that judgment is acquired.
What should General Counsel do next?
The discussion left several practical questions for legal leaders.
Does your organization have a reliable inventory of AI agents? Can you identify who owns each agent and what it is authorized to do? Can you reconstruct its actions if something goes wrong?
Have you reviewed your vendor agreements and insurance coverage for AI-related risks? Are your legal teams monitoring agents after deployment rather than relying entirely on initial approval?
And are you rethinking how your legal department works with outside counsel as AI changes the time and resources required for routine tasks?
These are no longer exclusively technology questions.
They affect corporate governance, legal exposure, operational risk and the economics of the legal function.
About the General Counsel Executive Forum
Open Future Forum launched its General Counsel Executive Forum to bring senior in-house legal leaders together for candid discussions and professional relationships.
The inaugural gathering in San Francisco was hosted with Foley & Lardner, Lockton and Simmons & Simmons.
Thank you to the participating General Counsel, speakers, sponsors and everyone who contributed to the discussion.
The forum is chaired by Open Future Forum founder Murray Newlands, with Louis Lehot of Foley & Lardner serving as co-chair. Its board also includes Preet Gill of Lockton and Emily Jones of Simmons & Simmons.
The community will continue exploring AI governance, corporate risk, legal operations and the changing responsibilities of General Counsel.
The purpose is straightforward: bring together people facing similar decisions, give them room to exchange experience, and build relationships that last beyond an individual event.
Learn more about the General Counsel Executive Forum.
Open Future Forum is a global executive community founded in Silicon Valley in 2019. It brings together senior leaders through invitation-only forums, executive events and original research.
Last updated: October 31, 2026
Frequently Asked Questions
Join the Conversation With Your Peers
Open Future Forum brings General Counsel and Chief Legal Officers together with CEOs, CFOs, CISOs and AI leaders for candid, private discussions on AI governance, risk and the future of the legal function.