요약
Any law should be thoughtfully written, avoid unintended consequences and be flexible enough to adapt as technology changes. That’s a tall order.
본문
By Larry Magid | [email protected]PUBLISHED: October 1, 2026 at 8:00 AM PDT Getting your Trinity Audio player ready...Putting politics aside, I have mixed feelings about the outcome of Tuesday’s White House meeting between President Trump and some of the nation’s most powerful technology executives.
Trump and leaders from companies including Google, Meta, OpenAI, Nvidia, Anthropic and xAI signed what the White House called an Accord on Super Intelligence. It calls on companies developing advanced AI to establish internal safety controls, use independent outside auditors and have board-level oversight. But the agreement is voluntary. Trump described it as “morally binding” and said the companies are going to “police themselves.”
I understand the appeal of that approach. AI is developing at breathtaking speed, which makes it extremely difficult for government to keep up. That has long been true of technology, but it is especially true of AI. By the time Congress understands a technology well enough to regulate it, the technology will already have moved on.
Similar to social networking debate
I’ve seen this problem before. Nearly two decades ago, when social networking was the hot new technology, there was a similar debate over how much government should step in and how much we could rely on the industry to make its own products safer.
Beginning in 2006, Connecticut Attorney General Richard Blumenthal, now a U.S. senator, and North Carolina Attorney General Roy Cooper, who later became governor, led a coalition of state attorneys general pressing MySpace and other social networks to do more to protect young people.
In January 2008, MySpace reached an agreement with 49 states and the District of Columbia that included more than 60 safety measures and the creation of an industrywide Internet Safety Technical Task Force to study technologies including age and identity verification.
I was appointed to that task force, which was directed by Harvard’s Berkman Center for Internet & Society and included representatives from Facebook, Google, Microsoft, MySpace, AOL and Yahoo, as well as child-safety organizations and academic researchers. After nearly a year of research and deliberation, including evaluating numerous proposed technologies, we concluded that the available age-verification technologies had significant limitations and that there was no single technological solution to the problem. We issued a report, but nothing else happened.
Seventeen years later, we finally have reasonably good systems that can estimate a young person’s age without necessarily requiring everyone to produce a government ID. Had we recommended that government mandate one of the technologies available in 2008, we might have locked ourselves into a system that was ineffective, impractical, unconstitutional or quickly made obsolete by better technology.
Tech moves faster than law
There’s another lesson from the history of social media. Sometimes the problem is that laws written for one generation of technology remain in place long after the technology changes.
One example is Section 230 of the Communications Decency Act, passed in 1996, which generally protects online services from being treated as the publisher or speaker of content posted by their users. If someone defames another person online, the person who posted the defamatory material can be held responsible, but the platform generally cannot be treated as the publisher of that material. It’s a bit like saying that a local government isn’t responsible for something defamatory that one citizen says about another in the town square.
That framework made a lot of sense in the days of AOL and CompuServe, when online services were largely places where people posted and exchanged information. But today’s social media platforms do much more than simply host what users post. It’s as if that local government not only provided a town square but also handed a megaphone to some of the people speaking in it because what they were saying was likely to attract more attention and engagement. Algorithms recommend, rank and amplify what we see, functions that lawmakers in 1996 could hardly have anticipated. Yet, 30 years later, Section 230 remains on the books even though the internet has changed enormously.
Self-policing alone isn’t sufficient
When Blumenthal and Cooper were pushing for stronger protections nearly two decades ago, many in the tech industry argued that they could police themselves. But attitudes changed. In 2019, Facebook CEO Mark Zuckerberg publicly called for a greater government role, writing that companies shouldn’t be making so many important decisions on their own. He called for regulation of harmful content, elections, privacy and data portability. I’m not suggesting that his ideas for regulation would have provided sufficient protections, but at least he acknowledged that government had a role to play. Today, the conversation is less about whether to regulate than how. But one of the most ambitious federal efforts, the Kids Online Safety Act, or KOSA, has been debated for years without becoming law.
My point is that government does have a role to play in providing a broad framework for AI, similar to how federal law allows the FTC to hold companies accountable for deceptive practices without telling them exactly what to promise their customers. To some extent, the White House accord establishes such a framework, but unlike the FTC Act, it doesn’t have the force of law.
As the saying goes, “There ought to be a law.” But any law should be thoughtfully written, avoid unintended consequences and be flexible enough to adapt as technology changes. That’s a tall order.
Disclosure: Larry Magid is CEO of ConnectSafely, a nonprofit internet safety organization that receives financial support from OpenAI, Google and Meta, which are mentioned in this article.