Freshfields partner says AI agents challenge board reliance differently from LLMs
Freshfields partner Ethan Klingsberg says AI agents pose a different challenge than LLMs to the line between board reliance and delegation. His test asks for a supporting record and directors informed enough to rely on it, but the text available ends before it says how that applies to agents.
The Board Room · Leadership desk
Klingsberg's reliance test: a record supporting the decision, and a board engaged enough to rely on it in good faith. Lawyers can check most LLM output against sources before the board meets. AI agents act somewhat autonomously; he says they pose a different challenge.
- constraint Klingsberg's reliance test A record that supports the decision, and a board informed and engaged enough to rely on it in good faith, claim 2
- capability LLMs Most output can be fact-checked; lawyers can check inputs against sources before the board meets, claim 6
- capability AI agents Act somewhat autonomously, connect to software, databases and memory; some manage other agents, claim 8
- constraint Boardroom use of AI Starting to move toward AI agents, a different challenge for reliance versus delegation, claim 7
| Who | How | Kind | Claim |
|---|---|---|---|
| Klingsberg's reliance test | A record that supports the decision, and a board informed and engaged enough to rely on it in good faith | constraint | 2 |
| LLMs | Most output can be fact-checked; lawyers can check inputs against sources before the board meets | capability | 6 |
| AI agents | Act somewhat autonomously, connect to software, databases and memory; some manage other agents | capability | 8 |
| Boardroom use of AI | Starting to move toward AI agents, a different challenge for reliance versus delegation | constraint | 7 |
What happened
- For LLMs, he says most output can be fact-checked, and lawyers can test the underlying analyses and inputs against sources before the board meets.
- Agents differ, he writes: they act somewhat autonomously, connect to software, databases and possibly internet sites, and some advanced agents manage other agents.
- He also writes that directors understandably push for shortcuts, and that shorter presentations and faster technology have helped.
Why it matters
- constraint On the logic of his bank example, an agent's bottom-line answer delivered without its analyses and inputs would leave a board in the position he warns against.
- cost The pre-meeting check lawyers run on LLM output would have to cover what an agent did, which we'd expect to add counsel time and preparation before each board meeting.
- exposure Directors who cannot show the inputs behind an agent-assisted decision put at risk the business judgment rule presumption that the usual reporting protocols are meant to earn.
The test Klingsberg describes concerns what the board knew and did. Writing in a Freshfields memorandum posted at corpgov.law.harvard.edu, he said that where no conflicts exist his answer typically derives not from whether he thinks a decision will beat the alternatives, but from whether there is a foundation in the record to support it and whether the board has received enough information and been engaged enough to find its reliance reasonable [4][2]. Reasonable reliance, he wrote, "does not mean delegation of judgment and decision-making" [1].
His own example shows how little a conclusion carries alone. A reputable bank tells a target's board that an all-cash price is the highest available after a market check, and that the price is fair from a financial point of view [3]. If the board took no part in the market check, does not understand its scope, and never received the financial analyses and the inputs and assumptions behind the opinion, he wrote, it is in serious danger of veering from reasonable reliance into delegation [3].
That standard has been workable for LLMs because their output can be tested. Klingsberg wrote that most of what LLMs generate can be fact-checked, and that lawyers can check the underlying analyses and inputs against sources before the board meets [6]. He describes agents differently. They act somewhat autonomously, use an LLM among other tools, can connect to software, databases, memory and potentially sites across the internet, and some advanced agents manage other agents [8]. He says that presents a different level of challenge for the balance between reliance and delegation [7].
In our view the difference follows from that description. A pre-meeting check works on a finished body of analysis. An agent that picks its own sources and directs other agents produces a process as well as a result, so the board's foundation would have to include what the agent did. That step is ours. The text available ends mid-sentence, before any recommendation for agents and before any discussion of the METR report named in the post's title [9]. It speaks of a "foundation in the record", and does not set a documentation standard for agent work.
The tradeoff is speed against record. Klingsberg wrote that directors understandably push for shortcuts, and that shorter presentations and faster technology have helped [10]. Agents extend that gain. The protocols he describes combine the bottom-line conclusion with the substantive analyses, inputs and candid qualifications, and he says they position directors to earn the business judgment rule presumption in nearly all non-conflict decisions [5]. In our view, a board that accepts agent-produced work on conclusions alone this quarter will have no foundation to point to at its next contested decision.
What to watch
- Whether the rest of the Freshfields memo sets out a protocol for agent-produced board materials, and what it asks directors to receive.
- What the post says the METR report means and does not mean for directors, since the available text stops before reaching it.
- Any court ruling on the business judgment rule for a board that relied on AI agent output.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence35
- Adoption
- Insufficient
- Hype gap
- Insufficient
- Incentives30
- Confidence45
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Klingsberg wrote: "Importantly, 'reasonable reliance' does not mean delegation of judgment and decision-making."
- [2]
In the absence of director conflicts or other loyalty issues, Klingsberg says his answer typically derives not from whether he thinks the decision will be better than the alternatives, but from whether (a) there is a foundation in the record to support the determination and (b) the board has received sufficient information and been sufficiently engaged to understand this foundation and determine in good faith that its reliance would be reasonable.
- [3]
In Klingsberg's example, a reputable investment bank reports to a target's board that an all-cash price is the highest available after a market check and is fair from a financial point of view; if the board was neither engaged in nor understands the scope of the market check and sale dynamics, and has not received the financial analyses and related inputs and assumptions behind the opinion, it will be in serious danger of veering away from reasonable reliance and into delegation territory.
- [4]
Ethan Klingsberg is a partner at Freshfields Bruckhaus Deringer LLP, and the post is based on his Freshfields memorandum.
- [5]
Boardroom reports regularly combine the bottom-line recommendation or conclusion with the substantive analyses and inputs behind it and candid qualifications, and Klingsberg says this approach positions directors to earn the business judgment rule presumption in nearly all non-conflict decision-making scenarios.
- [6]
Klingsberg writes that most of what LLMs generate can be fact-checked, and that even if advice to a board is facilitated by LLM research, the underlying analyses and factual inputs can be spelled out and checked by lawyers against sources in advance of the board meeting.
- [7]
Klingsberg writes that the role of AI in the boardroom is starting to evolve from reliance on LLMs toward assigning tasks to AI agents, and that agents present a different level of challenge for maintaining the balance between reasonable reliance and delegation.
- [8]
Klingsberg describes AI agents as able to act somewhat autonomously, using an LLM among other resources and tools, often configured to connect to software, databases and memory, including potentially sites across the internet, and notes that some advanced agents can manage other agents to plan and execute tasks.
- [9]
The post is titled 'Agentic AI and the Duty of Care: What the METR Report Means and Does Not Mean for the Boardroom'; the text supplied ends mid-sentence while describing how agents can work as a team.
- [10]
Klingsberg writes that, given the breadth and volume of risks, directors understandably push for shortcuts, that the shift from long memoranda to PowerPoints and concise executive summaries has helped, and that technology has further increased experts' ability to respond quickly to boardroom demands.
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
1 independent publisher whose own reporting we read for this story.
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.