An online meeting’s audio is used to automatically generate a transcription of the discussion, thanks to an included AI-powered feature. That transcription also becomes the basis for a summary of the meeting and is used to generate action items, which are sent to each attendee. As with any meeting, tasks are assigned, and decisions are documented. The difference is that AI is now behind much of those task flows.
Who owns those meeting outcomes from the AI-generated output? Who is accountable if those action items or transcripts are wrong, incomplete, acted upon incorrectly, or not acted on at all?
Thanks to the increasingly common enterprise use of AI-enabled tools for unified communications, those questions are more complicated to answer than they used to be. Today, AI in UC isn’t just changing how work gets done. It’s also creating work that doesn’t clearly belong to anyone, and there are a host of downstream effects to consider.
Research from LRN, released this year, suggests this problem is emerging more quickly than many organizations can manage. Thirty-nine percent of organizations report using AI in at least one aspect of their programs, said Matt Plass, chief market strategy officer at LRN.
As organizations focus on adopting AI tools, some experts argue they’re asking the wrong questions. “The industry is hyper-fixated on whether AI models are ‘safe,’ while completely ignoring whether their business environments are actually equipped to handle them,” said Allen Martinez, chief AI architect at Noble Digital Agency and creator of BXAI-OS, an AI governance architecture. To start, it’s important to understand what unowned work really is and how it shows up in modern UC environments.
Across UC tools and platforms, different kinds of AI-enabled outputs can be generated: meeting summaries, chat recaps, transcripts, and action items. When these outputs were generated entirely by a human worker, authorship and, therefore the chain of accountability, was much more straightforward. For example, if the meeting transcription typed out by an administrative assistant had an error in it, the problem lay either with the admin themselves or a glitch in the audio recording.
But the addition of AI-enabled tools to this work complicates both authorship and accountability. Many day-to-day enterprise outputs now have hybrid authorship that includes human input, AI synthesis, and platform automation.
Traditional ownership models defined by author, manager, and system owner roles no longer map cleanly in many work environments. If this shift isn’t addressed, the results can include diffuse authorship, orphaned outputs, and fragmented accountability.
These issues aren’t unique to UC platforms, but they can be amplified there.
One key advantage of UC platforms is their cross-platform workflows, which support an organization and its users across multiple roles, responsibilities, and outputs – for example, from meetings to internal chats to project tools to a CRM. These platforms also enable multiple participants, both internal and external, to coordinate and interact across these functions. Modern UC platforms support a blend of real-time and asynchronous work, which is especially important for organizations working across multiple locations. And increasingly, these platforms have AI-enabled features, such as in-meeting assistants and chat copilots, embedded across their functions.
But the features that make UC platforms so useful in enterprise environments also amplify the risks of unowned work. For example, when action items are auto-generated and don’t have a clear chain of accountability, they can be acted on without validation. If summaries generated by an AI-enabled feature are incomplete or incorrect, they could support poor decision-making. The knowledge generated by a variety of AI-enabled tools can be scattered across systems, without a clear plan to track it or use it effectively. And as these AI outputs are reused and compounded across different tools, these problems can grow or become entrenched in an organization’s digital store of knowledge in ways that are difficult to find and fix.
In UC environments, this can happen in subtle but consequential ways. AI systems are increasingly acting on behalf of employees, Martinez notes – summarizing meetings, scheduling follow-ups, even communicating externally. The risk in this comes from the accountability gap between an AI taking action and a human explicitly approving it, he said. “It looks like a frictionless environment where AI speaks on behalf of the business, but no human ever gave it permission to do so,” he says.
In practice, that gap is already showing up in collaboration workflows. In many organizations, “there is no review of AI-generated summaries, which are then treated as reliable records,” LRN’s Plass said. That can lead to action items being assigned without context, and to sensitive discussions being stored without clear standards for retention. This is further complicated by shadow AI, as different employees use different tools related to the same meetings, documents, or discussions, producing inconsistent outputs.
A breakdown of ownership related to AI-powered UC features can result in several different failure points. Planning for these issues, and addressing them when they come up, have important implications for mitigating legal and compliance risks, among others.
One result can be authorship ambiguity, where there is a lack of clarity about who ‘wrote’ a meeting summary, transcript, etc. When AI is used to generate these outputs, it can be difficult to determine who truly produced them and, therefore, to know who is accountable for them. Was it the AI tool? The person who organized a meeting? The meeting participants? The company itself?
Authorship ambiguity is a challenge in several ways, including in determining who is responsible for the accuracy of outputs. Beyond determining the author of an given document, it’s also important to be clear on who is responsible to verify those outputs for accuracy, clarity, and comprehensiveness. Part of this process is deciding when that verification step is necessary, and how it is triggered. Ownership can also erode at the point where AI-generated outputs are distributed. “If an AI generates a meeting summary and auto-sends it to 15 stakeholders, nobody reads it critically because everyone assumes someone else did,” Martinez said.
One potentially useful feature of AI-enable UC tools is the ability to auto-generate and assign action items from meetings, chats, and other business communications. This can be straightforward and timesaving, but it can also get complicated if the chain of accountability for these tasks is unclear. For example, do all the task recipients accept these assignments as is? Where do they go if they have a question, or if they disagree with the automated assignment?
Finally, the tension between system vs. human accountability must be considered. Whether the cause of a problem is a user error, a platform issue, or a governance failure might determine the response, as well as the next steps.
Even where organizations are thinking about accountability, their approaches aren’t consistent. Some companies place the responsibility on the person using the tool while others assign it to procurement or IT teams, Plass says. In other cases, accountability can be treated as shared across functions, which sounds collaborative in theory but often muddies ownership. That lack of clarity is especially risky when AI-generated outputs are used as records. “If an AI-generated meeting summary is shared internally, used to assign next steps, entered into a system of record, or relied on for decision-making, responsibility lies with the team or leader who accepted and acted on it,” he said.
As stated above, the stakes of this accountability gap go beyond operations. The potential legal and compliance concerns are wide ranging and can include e-discovery implications (what counts as a document?), audit trails (can organizations trace decisions made using AI?), and record integrity (are AI-generated summaries official records?). There’s also the potential of regulatory exposure, especially in sectors like healthcare and finance.
But these risks are not always obvious or immediate. Martinez describes what he calls a “shadow ledger,” a hidden accumulation of ungoverned decisions and commitments created by AI systems. Over time, those outputs can become systems of record – one that organizations can’t explain or defend.
“The biggest risk is not some dramatic failure. It is invisible, compounded errors and a total loss of legal defensibility,” he said.
Some enterprises are starting to tackle this emerging problem. One approach involves implementing clear human-in-the-loop expectations where they might have previously been implied – for example, by assigning explicit review responsibility for AI outputs. New data governance considerations that require treating AI outputs as managed knowledge assets may also be warranted.
Ownership models may also need review to account for the new hiccup of AI ‘owners,’ or a lack of ownership altogether. For example, organizations could put in place guidelines that specify that a meeting organizer owns the outputs and the team lead owns their validations, while functional ownership is tied to stages of workflow. In addition to guidelines like these, the development of new policy guidelines is likely also needed. These can include guidelines for editing AI-generated summaries, accepting and following up on AI-generated action items, and using AI-outputs downstream.
Finally, there are platform-level controls that can be implemented, or that warrant review. UC vendors themselves are adding features like attribution markers, approval workflows, and edit histories, which can support endeavors like those above.
More mature organizations are starting to formalize these practices, Plass says, signaling a stronger alignment between innovation and oversight. Those that are deploying AI more effectively are more likely to integrate it into their training and governance frameworks, he says, defining ownership before tools are scaled across the organization. And high-performing programs are also more likely to benchmark performance, use analytics, and explicitly address risk, he says.
Beyond that, addressing these risks requires more than policy updates. “You have to design the human accountability structure first,” Martinez said. He argues that organizations need to define their decision architecture, a clear structure for who is authorized to make which decisions, and under what conditions. A modern decision architecture should include built-in human approval checkpoints, explicit decision rights for AI systems, and audit trails that record what the AI did, who approved it, and why, he says.
However, even with these carefully considered measures in place, a lot remains unresolved in this quickly evolving area of UC. In these early stages, there is still a lack of standard ownership frameworks that address these new ownership models. Practices are inconsistent across tools and visibility is limited across platforms. Vendors are moving more quickly than governance models, which becomes a problem when enterprises incorrectly assume users know or understand their responsibilities in the accountability chain.
As with other use cases, AI in UC isn’t just about automating work. It’s also reshaping responsibility in important ways. Without clear ownership, the efficiency gains realized through AI-generated outputs can also introduce risk.
But if challenged today to explain an AI-driven decision, many organizations would struggle to answer questions like who authorized the AI use case, who reviewed the generated output, and which policy governed the AI’s actions. Companies often default to pointing at the vendor, Martinez says, but regulators are focused on internal accountability.
During this quickly changing time, enterprises need to deliberately define ownership for these outputs, and align their people, platforms, and processes with those definitions. The organizations that stand to benefit the most from AI in UC won’t simply automate workflows. They’ll also make sure someone is still accountable for them.