Companies face their own prisoner's dilemma on AI deployment, one workplace expert says, and no one is coming to save them
After Anthropic's CEO, Dario Amodei, published a 3,800-word essay calling on the AI industry to slow its pace of development, industry insiders, government officials, business leaders and regular Americans found themselves debating whether the companies building AI's most powerful systems should be the ones deciding how fast it advances.
Titled "We Must Pace the Frontier," the essay argues that AI companies are building systems faster than they can align, test or fully understand, and it lays out a three-step plan for reining in that growth, starting with Anthropic inviting outside evaluators to work inside the company with employee-level access. Sam Altman, CEO of OpenAI, Elon Musk, CEO of SpaceX, and Demis Hassabis, chair of Google DeepMind, all voiced support within hours, and the essay came just days after a former Anthropic researcher resigned, warning that AI could pose an existential threat to humanity within the decade.
But the industry's consensus didn't extend to Washington. President Trump dismissed the safety warnings as a "hoax" on social media, writing that the only guardrail AI needs is a strong, smart president.
That fight over whether frontier labs should slow down has dominated headlines, but it misses an equally urgent question for Anna Tavis, clinical professor and chair of the Human Capital Management Department at New York University's School of Professional Studies. What does pacing mean for the much larger number of companies that are buying and deploying these tools, rather than building them?
"Frontier firms are talking about pacing the capability of these tools," Tavis said. "For us, the big question is what are we doing. How are we pacing deployment, and what needs to happen?"
Two levels of the same debate
Employers face the same dilemma without the luxury of coordinating an industry wide pause on their own, Tavis said.
"It's pretty much a prisoner's dilemma," she said. "No one wants to be in a position to freeze AI deployment on their own. The government is the only body that can regulate everyone at the same time, so doing it alone is basically surrender."
No single employer can simply decide it's too worried about the consequences and halt all AI implementation, tearing up vendor contracts in the process, Tavis said. Each company needs its own separate set of considerations instead.
Without a regulatory framework forcing anyone's hand, she argued, the burden falls on individual leadership teams, starting with HR.
"In the absence, in a complete crisis as far as I'm concerned, of the government that we currently have in this country, the responsibility rests with individual leadership," she said.
"Every company, every CEO and every board, and the HR department in particular, need to be particularly vigilant around the deployment part of [AI]," she said.
Who's accountable when an AI vendor gets it wrong
For Tavis, that vigilance starts with governance, meaning who inside a company is actually responsible for vetting AI vendors and reviewing the contracts coming in.
"Who are the technical leads? What is the governance structure?" she said. "It starts with the vetting of the vendors. Who are you working with?"
In the absence of a framework handed down from above, she said, every company needs to build those structures internally rather than wait for one to arrive.
"To say the vendor did it is not going to work," Tavis said. "The responsibility rests with the company."
That accountability needs a home, she added, a dedicated body with the expertise to press vendors on their own governance and transparency practices.
The case for investing in management
Amodei's essay leans heavily on the idea that AI's growing capability demands more human oversight, not less, a principle Tavis thinks applies just as directly inside the companies deploying these tools. She argues against the popular assumption that management layers are on their way out.
"Management is not going away, contrary to what everyone is saying, that managers are going to be replaced with AI and agents," she said, pointing to Meta's recent decision to begin rehiring people into management roles after earlier layoffs.
Agents can streamline operations, tasks and logistics, Tavis said, but judgment calls can't be automated away.
"The judgment decisions, the ethics decisions, the coordination decisions and the people impact decisions are not going to be done by the agents," she said. That preference for human judgment isn't limited to management alone. A recent industry report found strong support for keeping people, not AI, in charge of hiring decisions.
Companies need to retrain and invest in their managers rather than cut the layer entirely, she added, because the risks are much higher now and could scale quickly if left unaddressed.
Managing an anxious workforce
Amodei's essay arrives amid a steady stream of public warnings about AI, and Tavis said employers can't pretend their workforces aren't being influenced by it. A Pew Research Center survey of more than five thousand employed U.S. adults found 52 percent are worried about AI's future impact on their jobs.
"Management can't just bury their head in the sand when we're being bombarded every day with these apocalyptic messages," Tavis said, comparing the moment to wartime decision making, where a broader strategy exists but individual units still have to exercise their own judgment in real time.
"It's not just a pursuit of profit and growth that our CEOs have been trained for and selected for," she said. "Now there is this additional layer of tremendous responsibility."
Tavis said companies shouldn't wait for regulation before building their own AI governance frameworks.
"There's no one to wait for at this point," she said. "It has to be done in coordination with the leadership of the company or department."
That increasingly means rethinking org design decisions many companies made only a couple of years ago, before generative AI reshaped assumptions about how lean organizations could run.
"There may need to be a different kind of leadership, at all levels, a more situational leadership style," Tavis said. "And that's not what the majority of firms seek and hire for."
Some of that rethinking already shows up in how companies talk to employees about automation, a theme HRD America has traced in its reporting on managers doing less of the work and more of the talking.
For now, Tavis said, none of it can wait for a broader consensus to emerge.
"Companies have a lot of responsibility right now, and most importantly, to protect themselves, because no one else will," she said. "The responsibility rests on them."
