Three tech visionaries warn about AI trust failures

trust and – In a DisrupTV conversation, Dr. Vint Cerf, Dr. David Bray, and Cheryl Strauss Einhorn push CEOs, boards, and policymakers toward clearer AI governance—so teams can trust outputs when it’s earned, and demand recourse when it’s not.

The first warning arrived fast: don’t let AI agents talk to each other like they’re fluent, when they may only be quick.

Dr. Vint Cerf—one of the Internet’s co-creators—said he worries about “agents talking to each other using natural language.” His fear is simple and chilling: “We don’t need agents to misunderstand each other and execute at the speed of light compared to human speed.”

Cerf’s reference point wasn’t sci-fi. He connected the risk to something every business already knows: instructions can be wrong. even when the system follows them precisely. “Programs, at least the deterministic ones, do what you tell them to do. The trouble is sometimes what you tell them to do isn’t what you wanted them to do. It’s called a bug.”.

The conversation—hosted on DisrupTV by R “Ray” Wang of Constellation Research. with Vint. David. and Cheryl joining—kept snapping back to one theme: trust and accountability in the agentic future won’t be a matter of vibes. It will be engineered into how organizations give AI instructions. how they judge outputs. and who carries the consequences when something goes wrong.

Dr. David Bray put that into a governance frame that sounds almost old-fashioned: “I define governance as how we avoid anarchy. We’ve got to have anarchy protection, not just for humans, but for agents.”

To show what he means, Bray compared today’s patchwork AI environments to the chaos of early streets. “It’s sort of a repeat of 1910. We hadn’t invented stoplights yet. We hadn’t even figured out stop signs or right of way or sidewalks.” In the 1910s. he said. New York and Chicago streets carried trolleys alongside personal automobiles. human pedestrians. and horses—an image meant to mirror modern enterprises where different cloud-based AI models. local AI models. human users. and other analytic software tools all share the same road.

Cheryl Strauss Einhorn—an award-winning investigative journalist and founder and CEO of Decisive. a decision sciences company—made the stake personal. “When the hammer falls, it falls on us. AI doesn’t care. We are going to all be the ones who have to explain, and we’ve got to bear the consequences.”.

Taken together, their warnings land on the same decision point for leaders: whether an organization treats AI as something that merely executes—or as something that must be governed like a partner that can fail.

Cerf described AI as more than a tool when it comes to how people relate to it. “We’re trying to figure out how it thinks. ” he said. calling the work of interpreting AI outputs a matter of understanding the machine as a “new set of workers that we can relate to.” He added. “It looks a lot to me like a very smart research agent.”.

Bray focused on the behavioral side of the trust problem: “A healthy response for societies in these times is increasingly don’t trust the first thing you see unless you triangulate it… That’s what the CIA does.”

He brought the lesson to life through a memory from his own work. Bray once had to endure a public commenting system being flooded by bot-generated comments—and he was required to record all 23 million comments submitted. regardless of who. human or digital bot. was posting. His point wasn’t just that systems can be overwhelmed. It was that AI agents can amplify and also obscure human voices.

The trust question, then, isn’t only about whether AI is “correct.” It’s about whether people can tell what they’re looking at, and whether organizations have processes to verify it.

Einhorn returned to something leaders often skip: self-knowledge before you command a system. She said, “Each of us has a special sauce. It is the way we make decisions. And most of us don’t really have awareness of what that is.”

Her advice for prompting AI was practical and blunt: “If you’re going to lead the machine, what you actually need to do is spend more time to investigate your special sauce… so that instead of giving you somebody else’s answers… it can actually work specifically for you.”

She also framed the cultural shift that comes with leaning on AI for problem-solving. “This is not just some new software. This is actually a cultural change… about problem solving.”

That matters for accountability, too. If AI does something incorrectly, leaders can’t treat the outcome as an accident they don’t own.

Einhorn offered a sharp metaphor for how different uses of AI demand different levels of responsibility. “There are really two different ways that people use AI — the surgeon and the Lamborghini driver.” When the goal is one specific answer. she said. “we want one specific answer… we’re using it like the surgeon.”.

But even the surgeon analogy comes with risk. She acknowledged that “There are risks that a surgeon might make an error, and there also are risks that a surgeon could do everything right and still have adverse outcomes for their patient.”

The second model is for high-stakes decisions that require a process. “There are other times… you’ve got a really high-stakes decision, and you want to undergo a process. At that point. you’re the Lamborghini driver.” In that situation. she emphasized. “skill at navigating tight curves and knowing the capabilities as well as limits of the AI engine are essential to avoid the equivalent of AI car wrecks.”.

Bray brought the accountability lens back to institutional responsibility using a maritime warning system. “Whose flag is this AI agent flying when it is doing something?. Whose organization is it flying the flag up?” His answer was direct: “In this example. the organization. if it employs an AI agent. also takes responsibility for the agent. assuming good instructions are provided.”.

He also flagged a near-term problem for leaders: distinguishing what’s real. “By 2030, more than 40% of the information on the planet will have been synthetically produced by an AI… that’s going to create massive questions for CEOs and boards.”

Cerf circled back to the need for recourse when something goes wrong. “Establishing a mode for recourse in a variety of circumstances might be a very high benefit and maybe even a necessity.” He said he has been encouraging such approaches to AI. and both Cerf and Bray acknowledged that companies like Salesforce. Google. and others were taking similar benevolent approaches to AI in workplace and customer settings.

The leaders also shared the concern that AI is changing what people can access in the first place. Cerf pointed to “losing access to digital information” as a “serious issue,” including the risk of losing both software used to interpret data and the origin of data.

As the episode moved toward takeaways for CEOs and boards, the through-line became unmistakable: clarity of instructions, judgment over blind acceptance, and accountability structures that don’t vanish when harm happens.

The three experts highlighted the needs for organizations to ensure instructions given to generative AI are “clear. precise. and don’t have unintended consequences.” They connected the 1910s streets picture—New York and Chicago with trolleys alongside personal automobiles. human pedestrians. and horses—to the present era of incompatible systems sharing the same environment without modern safeguards like stoplights. stop signs. right of way. or sidewalks.

They also stressed that “good human discernment” is increasingly essential for deciding whether to trust AI outputs—and that it’s getting harder to distinguish between human-generated and AI-generated content. In their view, “good chemistry between humans and AI” helps separate productive relationships from those that are unhelpful.

Finally, they argued that organizations must work toward “clarity, as well as individual and organizational recourse,” if an AI does something incorrectly.

One more detail threaded through their conversation: the future they’re urging leaders to prepare for is collaborative. not purely transactional. All three experts emphasized employing “AI in the group. ” recognizing networked interplay between humans and AI based on intent and accountability. Cerf and Bray also emphasized moving beyond the Turning Test. focusing on AI that amplifies “individual and collective human abilities” and helps improve strengths.

Even the way the discussion was framed carried its own urgency: Cerf’s warning about agents misunderstanding each other and executing “at the speed of light. ” Bray’s governance warning about avoiding anarchy. and Einhorn’s insistence that accountability “falls on us” instead of disappearing into code.

This article was co-authored by Dr. David Bray, principal and CEO at LeadDoAdapt (LDA) Ventures, chair of the Accelerator, and distinguished fellow at the Stimson Center.

AI governance agentic AI accountability trust in AI cybersecurity misinformation Vint Cerf David Bray Cheryl Strauss Einhorn