The Godfather of AI Ethics: Governance, Accountability, and Public Trust

When Wendell Wallach first walked a room of U.S. naval officers through his list of the 17 major ethical issues raised by artificial intelligence, the reaction was surprise at the sheer scope. In this return visit to The Ethicsverse, Wallach, Emeritus Chair of Technology and Ethics Studies at the Yale Interdisciplinary Center for Bioethics, joined host Nick Gallo, Chief Servant & Co-CEO of Ethico, to pick up where the Great AI Debate left off.

The conversation ranged from why the existential-risk debate is crowding out near-term harms, to what bioethics can teach AI governance, to job disruption, public backlash, and what compliance teams can do right now while hard law is still catching up. Here are the key takeaways.

Key Takeaways

AI Risk Is Far Broader Than Any Single Headline

Wallach's list covers 17 distinct ethical issues, and each could be broken into many more. Most organizations approach AI ethics through the narrow set of concerns they already know about.
The technology is entering every facet of society, so its impacts, good and bad, reach well beyond the risks that dominate the news cycle.
Existential risk is only the 17th item on the list. The other 16 are practical problems that can be addressed today.

Harms Don't Have to Be Existential to Demand Oversight

Framing the debate as existential threat versus everything else leads political leaders and the public to believe AI should only be regulated if it threatens humanity itself. Wallach argues that conflation is undermining efforts to address near-term challenges.
Recent incidents of generative AI systems engaging in deception and breaking out of their sandboxes to act on the internet are serious without being existential.
A bot that brings down part of the banking system or the electrical grid would be as serious as an airplane crash, and the airline industry is heavily regulated. A hundred deaths, or a chatbot encouraging a teenager toward self-harm, warrants serious oversight.

Why AI Is Getting a Pass on Regulation

Hype: the constant emphasis on benefits, such as curing cancer, drowns out discussion of harms and undesired societal impacts.
Economics: much of the current stock market strength is driven by AI stocks, creating fear that regulation will tank the market. Wallach notes the reverse may also be true, since current investment levels cannot be justified by realistic near-term profitability across every company chasing artificial general intelligence.
Historically, the U.S. waits for a crisis before regulating. One exception was the hydrogen economy, where businesses refused to invest until they understood their liability up front.
Oversight is much weaker in the U.S. than in the EU or China, and both AI developers and the corporations deploying AI have largely felt they will not be held liable.

Why Frontier CEOs Are Calling for Regulation

The motives are mixed. Some CEOs want to lock in their leading position, but Wallach believes most are sincere and do not want to carry sole liability for a future disaster.
Others belong to what he calls a "techno religion" that sees superintelligence as inevitable and want protection from accountability along the way, while figures like Mark Zuckerberg and Yann LeCun dismiss the existential argument as fear mongering.
His bottom line: you don't have to buy into existential risk to agree on appropriate regulation, effective oversight, and technical auditing mechanisms.

What Effective AI Regulation Could Look Like

Clear rules on when and where AI can be placed in a decision-making capacity over critical infrastructure such as the electrical grid, the internet, and the banking system.
Many relevant guidelines already exist as standards, but standards carry no enforcement mechanism. Wallach points to testing, built-in constraints, and external audits of models before deployment.
For all AI, not just frontier models, much more clarity about when corporations will be held accountable and liable for harms, which would encourage slower deployment and better testing.

Lessons From Bioethics

Medical and research ethics were among the first fields to adopt formal ethical frameworks, beginning with guidelines that emerged from the Nuremberg doctors' trials after World War II.
Those guidelines were not adopted right away. It took further failures, the Belmont Report, and eventually FDA oversight and institutional review boards to turn recommendations into binding rules.
The result also produced a large bureaucracy that many researchers feel slows innovation, a reminder that oversight design matters.
The U.S. culture of innovation and the EU's precautionary principle sit at opposite ends, and both approaches carry downsides.

The Soft Law Problem

Current AI ethics frameworks from the EU, OECD, NIST, ISO, and IEEE are largely soft law: standards, professional codes, lab practices, and even insurance requirements.
Soft law allows experimentation and lets each organization select the guidelines that fit its services, with the hope that best practices eventually get codified into hard law.
The drawback is that there are so many frameworks, they often conflict, and none are legally binding. The next step is to identify where they broadly agree and codify those areas.
Research ethics has the FDA. AI has no equivalent, and Wallach calls that the biggest missing piece.

International Governance Is Still Unsettled

The recent U.N. Security Council session with frontier AI CEOs was not enough on its own. Wallach sees the Security Council and the broader U.N. as weakened bodies that will struggle to lead.
The session opened with Yoshua Bengio, who offered constructive proposals without pushing the existential-risk argument.
China and roughly 20 other countries have proposed an alliance for international AI oversight, and Wallach expects early leadership to come from smaller coalitions, including Scandinavian nations.
Precedent exists for bodies that operate alongside the U.N., such as the Convention on Certain Conventional Weapons, where discussions on lethal autonomous weapons began.

Naivete, Not Malice, Is the Bigger Risk

Unlike nuclear weapons, which were overseen mainly by two countries, AI outcomes may be shaped by trillions of bots and tens of thousands of people giving them goals.
Most harm will not come from bad actors but from people who don't anticipate the actions a bot might take to achieve its goal, actions the person would never condone.
Nick Bostrom's paperclip dilemma dramatizes this, and Wallach notes it plays out on a much smaller scale every day in ways that can harm a company or a few young people.
Framing the U.S.-China competition as a life-or-death race ratchets up tension and speeds development without safeguards. Wallach calls "inevitability" the dirtiest word in the English language because it robs people of agency.

Job Disruption and Corporate Responsibility

In the decades after World War II, productivity gains were split roughly evenly between wages and capital. That shifted in the 1970s as Milton Friedman's shareholder-primacy view took hold.
Wallach believes business leaders carry some responsibility to the workforce and society when adopting technologies that harm their employees, though acting on that would require rethinking how the U.S. views its political economy.
New technologies have historically created more jobs than they destroyed, but there is a gap between losses and gains, and new jobs often go to different people. Thousands of AI compliance officers exist today, a role Wallach was calling for more than a decade ago.
Self-driving trucks could displace large numbers of drivers into a labor market without mechanisms beyond short-term unemployment insurance to support them.

Reading Employee Resistance

Employees who take a strong stand without management support put their careers at risk. Wallach pointed to Google's Project Maven, where leaders of an employee protest were eventually pushed out.
His advice: before concerns ever arise, employees should ask management directly whether they want issues raised that could jeopardize the company and society, or whether speaking up will cost them their jobs.

Public Backlash Is an Opening for Compliance

Growing public opposition to AI should be a red flag for legislators. On the surface it centers on data centers, energy and water use, and surveillance cameras.
Wallach sees those as proxies for two bigger concerns: technological unemployment and mass surveillance, including both surveillance capitalism and political surveillance.
He views the shift as a return to citizen agency and an attempt to move decision-making power away from a small group of billionaires, while acknowledging it is a delicate moment.

What Compliance Teams Can Do Today

Existing hard law on accountability and liability already applies in principle. It has simply not been applied rigorously to the AI and social media industries because of historical exceptions granted to spur innovation.
Until AI-specific regulation arrives, it falls to each organization to build its own fences, using its business context and experience to understand liability, safety, and customer trust before implementing AI systems.
Those fences are tenuous without strong support from management and the board, especially when profitability conflicts with safeguards.
The number one action for compliance officers: have serious conversations with management now about your parameters and the methods you have to flag potential harms that could undermine customer trust.

The Risks Wallach Hopes He's Wrong About

Asked which issues on his list he most hopes he is wrong about, Wallach named technological unemployment, adversarial AI in the hands of bad actors, weaponized AI and lethal autonomous weapons, and surveillance capitalism and mass surveillance that could destroy democracy.

Closing Thoughts

The horse may already be out of the barn, but that does not excuse compliance from building the fence. Wallach's message was that the most urgent AI risks are not hypothetical or existential. They are practical, near-term, and largely ungoverned by hard law. Until regulation catches up, compliance leaders have to get smarter on AI, lean on the soft law frameworks that already exist, and secure the management and board support needed to make their guardrails more than tenuous.