At D.C. Bar Summit, Experts Examine Risks and Governance Challenges of Autonomous AI
July 06, 2026
On June 30, the D.C. Bar CLE Program presented its first AI, Innovation, and Intellectual Property Law Summit, bringing together more than 60 practitioners, policymakers, and technology professionals. The all-day event featured discussions on how the latest advancements in artificial intelligence, such as agentic AI, large language models, and generative AI, will shape the future of patent, trademark, and copyright activity.
Keynote speaker Aleksandar Bijelic, chief technology officer at SymbioFoundry, where he leads the design and implementation of AI systems, focused on agentic AI. Bijelic says these systems can independently set goals, create plans, and execute multistep tasks. Because agentic AI is more advanced than traditional chatbots, operating with even less human oversight, Bijelic stressed the importance of governance.
"The AI autonomy and delegated decision-making are already in the enterprise," he said. "No matter if you want it or not, it's there. The unmanaged use of AI is already happening. The choice is not AI or not AI. The only choice is to govern AI and the shadow AI. It's very simple."
He said high-level employees in companies throughout Europe and the United States are using shadow agentic AI chatbots to create presentations using proprietary materials or unverified data. "They knew they shouldn't be doing it," Bijelic said, "but the pressure to deliver was so high that at the end of the day, they thought it was worthwhile."
"And while this is happening, we are lagging [behind in] the current governance, the IP protection, and the accountability structures — all of them are significantly behind the speed of the technology deployment," Bijelic added.
Time for a Kill Switch
Emerging AI systems are doubling their capacities every three and a half months, according to Bijelic, mentioning that "there is a huge pressure on any technology person to develop and deliver as fast as possible operational gains and operational efficiency."
Bijelic characterized AI governance not as a bureaucracy, but rather "how you preserve speed without losing the responsibility." Emphasizing the risk of not governing agentic AI, he said people are allowing technology to make decisions on their behalf. "Agentic AI can decide to hire or fire someone," he warned. "[It] can decide to invest in certain stock options. A majority of the investments on the stock markets are not happening by humans anymore. They are happening by the trade bots and agentic AI … so the question is, who is responsible if that investment is wrong?"
Bijelic recommended constant monitoring of agentic AI and developing a "kill switch" as a way to stop it before it creates damage. "We're not talking about creating a 'kill switch' because one day AI will overrule the world and kill all the humans," he clarified. "We are talking about things like agentic AI being able to deploy the mortgage or the loans or being able to invest. We want to make sure that we're able to safely disconnect agentic AI from work if we find it's drifting or [engaging] in [implicit] biases."
Anytime a company deploys an agentic AI system, an in-house principal could be responsible for the actions of that technology. The challenge is determining whether that principal will have the knowledge, skills, and technical capacity to understand everything that is happening with the agentic AI system.
"So, if your offices are using agentic AI to help do some legal work, you are being responsible to verify the facts … and all the work," said Bijelic. "You cannot say, 'Well, we bought the agentic AI from this company, and they are the ones responsible. They are not. You're responsible."
Effective Governance Structure
Bijelic shared SymbioFoundry's five-layer governance system, starting with developing a policy that reflects "risk awareness and accountability structure" for the organization's board. "[The policy] enables the board to keep control of the whole agentic AI," he said.
The second layer is executive control, where companies establish a new board, such as an executive AI governance council led by a risk and compliance officer who holds responsibility for AI, rather than the IT staff, chief technology officer, or software developers.
The third layer, mandatory classification, involves two sublayers. The first is the integration policy, containing board-approved language concerning the delegation policy for the agentic AI technology and a risk tolerance statement. The second sublayer is control, which involves the executive AI governance council that will deploy the operational constraints and add the social technical impact analysis.
"The key output for the third layer is that every agentic AI decision has to be explicitly traceable to board-approved policies, to a named principal with no ambiguity about the authority for that agentic AI, regarding who has deployed it or who is shutting it down," Bijelic said.
In the fourth layer, technical enforcement, each agentic AI has its own identity during runtime. "We need to know exactly each agentic AI [system or chatbot] — what is its identity, which roles it has, which data it can access, and at what time we can shut it down," Bijelic explained.
The fifth layer is assurance, which contains the audit proofs of when the agentic AI was deployed each time, yielding quarterly reports to ensure that an organization is maintaining the system properly.
Bijelic warned against operating on the extreme opposites of the spectrum, such as avoiding agentic AI altogether or going gung ho with the technology with zero governance. "The companies positioned in the middle, which are carefully using the agentic AI and carefully implementing the policy, are the best ones moving forward," Bijelic said.
Is AI Coming for Lawyers?
The importance of governance over AI was reiterated during the session "Open the Pod Bay Doors Hal: Will AI Replace Lawyers or Create New Opportunities?" moderated by Laura Possessky, former chair of the ABA Science & Technology Law Section.
Among the panelists was Gabrielle Kohlmeier, founder and CEO of GK Strategic Advisory, who mentioned that a lot of people think the current ethical rules for practicing attorneys work with regard to AI. However, agentic AI calls for new rules, Kohlmeier argued.
"It's because of how much forward-thinking is required and how much we need to get ahead of things as opposed to just having a record of what happened before," she said.
When asked if AI will replace first- and second-year associates, the panelists agreed that AI will augment associates' work productivity instead of replacing them. "There are things that AI is great at that humans are not, and there are things that humans are great at that AI is not," said Nick Reese, founder and CEO of Triantha. "Where you really find power is when you find a way to put those two things together within a governance framework and you increase productivity per hour per associate."
Courtney Thomas, legal ethics counsel at the D.C. Bar, weighed in on the ethical considerations for law firms when using AI, citing common pitfalls such as AI's tendency to hallucinate and produce false information. "Some of the more day-to-day issues that I've seen [here] and in other jurisdictions through ethics opinions and articles are competency, supervision, and confidentiality," she said. "Those are things that come up multiple times."
Kohlmeier said technical competency is knowing how to use AI while retaining your critical thinking skills. "I am an AI-fluent [person], not AI-first kind of person," added Kohlmeier. "We should need to know when to use AI and when not to use it. We really can't be competent with AI tools if we don't really know how the technology works."
