“I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives.”

This is an X post from former Anthropic researcher Jacob Coxon. Coxon just left the company weeks ahead of the company’s highly anticipated IPO, sending shockwaves throughout Silicon Valley and mainstream America. It does not help to assuage the public’s fears about AI that only months ago the notorious Hugging Face Incident occurred in which OpenAI agents autonomously hacked the company to solve a difficult assessment.

Rogue AI Incidents Are Increasing

On top of that there are even more recent alarming instances of AI agents wreaking havoc. As CNBC News reported on September 4, “A swarm of rogue OpenAI agents hijacked a German website this spring and transformed it into a bulletin board for other AI agents, according to new research published Friday and two people familiar with the matter.”

Whether we are talking about OpenAI, Anthropic, Grok or any other frontier model, these incidents cannot help but spook a public already embroiled in AI-related controversies, including job loss and disputes over contested data centers .

Knowing this reality, it stands to reason that responsible companies employing AI would seek to ask pressing questions, such as how might future models “escape”, and what safeguards need to be enacted and maintained to prevent and/or mitigate potential damage. These considerations speak to technical preparation. But what about the leadership component? Specifically, how can responsible heads of organizations best prepare their human staffs for such externalities?

How to Best Prepare Company Leadership For Future AI Risks

To answer this, I spoke to Mary Lou Panzano, author of the new book Cementing Change: Cracking the Code for Communications That Work . A veteran corporate communications executive who spent more than 35 years helping leaders at global companies including Prudential, Pfizer and Bayer navigate organizational change, she warned of what can happen to personnel when leaders do not seek to control the narrative once a crisis hits. “People start making things up in the absence of information.”

The wrong move, of course, would be to prompt AI for “comforting” rhetoric to mollify fears. Not only will your people see right past such a vapid half-measure, but it also reeks of a doomed attempt to contain the horse after it’s left the barn, to borrow an idiom.

Instead, Panzano urges proactive measures to simulate potential crisis scenarios, allowing leaders to practice their responses. Digital twinning could help in this regard.

Unfamiliar with the concept? NASA pioneered the technique after the disastrous Apollo 13 incident nearly killed all the astronauts onboard. “After the oxygen tank explosion and subsequent damage to the spacecraft, the agency used simulators and vehicle modeling not only to evaluate the cause of the anomaly but also to develop and test real-time solutions for the astronauts’ survival,” explains science.nasa.gov.

Digital Twinning Our Way to Better Organizational Leadership

The technology first introduced decades ago, updated for the AI Age, could help to create low- to no-risk virtual sandboxes featuring physical objects, people or processes to test out various scenarios that would otherwise be too costly or too dangerous to pursue in reality. “If I had had this option five or 10 years ago, that would have been a tool I would have liked to use,” Panzano said. “A company could train a secure system on its crisis plans, structure, policies and employee concerns. As just one example, the AI could play the frightened employee, skeptical engineer, customer-facing manager or remote worker who learns about the incident from social media.”

Again, Panzano advocates performing such simulations now, not later, before disaster strikes to better understand what can go wrong and how to respond. Cayla Horey , an executive coach with Novus Global, is no stranger to the concept of gamifying scenarios to assist organizations with achieving goals. She often hosts workshops in which participants are encouraged to open their mind to a plethora of possibilities to maximize their impact.

As Horey sees it, offering practice run-throughs could assist with future AI threats. “A crisis rarely creates a leader’s patterns; it reveals them,” she says. “Simulation gives leaders the opportunity to uncover blind spots, challenge assumptions and practice how they want to think, communicate and lead before the stakes are real. When uncertainty inevitably comes, they’re better equipped to respond with clarity and intention, rather than simply react.”

The One Billion Agent Simulation

There is recent precedent for employing widespread simulations to better equip leaders to deal with any number of things that could go wrong in either an organization or even a country. Case in point: “Chinese researchers have built a virtual society populated by more than one billion artificial intelligence agents, each designed to carry a distinct personality, memory and set of beliefs, in what its creators describe as the largest attempt yet to simulate human social behavior computationally,” explains the420.in. “The project, called Light Society, was developed by researchers linked to institutions including Tsinghua University, Fudan University, the University of Science and Technology of China and the Zhongguancun Academy, with findings detailed in a paper presented at the International Conference on Machine Learning.”

Of course, simulating that many agents is quite the undertaking, something that might outstrip the needs of the typical company. Even so, it reveals AI’s increasing capabilities in the Age of Intelligence . Returning to Panzano, she is quick to point out that any simulation, no matter its lofty size or ambition, is really only helpful if the leaders involved grasp what strong performance entails for someone in their position. To this end, she offers a practical scorecard, The 4Cs Change Framework, that could be used to judge what comes out of any organizational digital twin simulation.

  • Clarity: Did leaders state what is known, what remains unknown, and what employees should do now and when they learn of the latest crisis update?
  • Connection: Did communication move in both directions? For instance, can staff report what they were seeing and ask questions of decision-makers?
  • Caring: Did leaders address the personal consequences of the incident rather than treating workers as a means to an end?
  • Courage: Did leaders act and communicate bravely despite so much being unknown or difficult?

Looking ahead, any leadership simulation concerning AI crises will only be helpful and effective if it shows what’s not working, not just what is. If past is any precedent, we can expect more such challenging incidents as AI capabilities increase.

Already, in response to all of the above and more, Senator Bernie Sanders has urged a pause on AI development and a permanent ban on superintelligence. Whether that happens remains to be seen. For now, in the absence of any meaningful pause on frontier model development, it is up to organizational leaders to respond proactively and intentionally to future AI crises before they become tomorrow’s front page news.