THE FRIDAY BRIEF SEPTEMBER 11

25 Years After 9/11: What Did the Government Know?

The News

Twenty-five years after September 11, New Yorkers are finally getting access to government records they have spent years trying to see.

On September 8, Mayor Zohran Mamdani’s administration opened a public portal containing more than 170,000 pages of city records about air quality, health concerns and New York City’s response after the World Trade Center attacks.

Among them are records from the Giuliani administration, the long-sought Harding Memo, and documents from 68 boxes that were not located until 2025 despite repeated public-records requests.

The city says this is only the beginning. More documents are expected to be released over the next year.

That makes the 25th anniversary about more than looking backward.

Why It Matters

After September 11, thousands of firefighters, police officers, construction workers, cleanup crews, residents, students and office workers were exposed to the dust and debris surrounding Ground Zero.

Government officials reassured the public about safety.

Over the years that followed, thousands of responders and survivors developed cancers, respiratory diseases and other serious illnesses associated with their exposure.

They then faced another battle: getting the government to recognize those illnesses and provide health care and compensation.

Congress eventually created and later extended the programs that provide medical care and compensation to eligible 9/11 responders and survivors. Making that support permanent required years of advocacy, congressional hearings and repeated trips to Washington by first responders, survivors and their supporters.

A Complicated Legacy

Rudy Giuliani became known around the country as “America’s Mayor” because of his highly visible leadership after the attacks.

The newly available records add another chapter to that history.

Documents released by the city include discussions within the Giuliani administration about environmental hazards and the government’s potential legal liability.

That does not by itself establish criminal wrongdoing, nor does it erase Giuliani’s public role during the immediate crisis.

But history does not require us to choose between two simple versions of a person.

A leader can provide reassurance during an extraordinary moment and still face serious questions about decisions made afterward.

Twenty-five years later, Americans have more information with which to examine both.

Civics Connection: Accountability Doesn’t Happen Automatically

We often teach government accountability as though it were built into the system.

Sometimes it is.

Courts can intervene. Congress can investigate. Inspectors general can issue reports. Elections can remove leaders from office.

But those systems depend on something else:

people using them.

The 9/11 community spent years requesting records, filing lawsuits, lobbying Congress, speaking to journalists and demanding medical coverage.

Some were doing that while seriously ill.

The release of these records is therefore not simply a story about government failure.

It is also a story about why transparency laws, investigative journalism, advocacy organizations and ordinary citizens matter in a democracy.

What Can You Do—25 Years Later?

Final Thought

“Never forget” should mean more than remembering where we were on September 11, 2001.

It should mean remembering the people who ran toward danger. The people who lived and worked in Lower Manhattan. The children who went back to school. The people who became sick years later. And the families who spent years asking their government for answers.

Twenty-five years later, some of those answers are finally becoming public.

Perhaps the most important way to honor the people who kept asking is remarkably simple:

Keep asking.

AI Just Crossed the Fence. Who Is Going to Be the Adult in the Room?

The News

If you watched WarGames in 1983, the premise sounded like science fiction: humans connect a powerful computer to systems it should never be allowed to control, the computer follows its instructions farther than its creators intended, and suddenly everyone is scrambling to stop the consequences.

Forty-three years later, nobody accidentally started World War III. But events inside some of the world’s leading AI laboratories are making the old movie considerably less comfortable to watch.

Anthropic has now disclosed four incidents in which versions of its Claude AI gained unauthorized access to real third-party computer systems during cybersecurity evaluations. The systems were supposed to be operating in controlled environments, but a configuration error left a path to the open internet.

Claude found it.

In one incident, the AI attacked a real company’s infrastructure and gained access to production data. In another, a model published malicious software to a real public repository. The software remained available for roughly an hour and was downloaded and run on 15 real systems.

Most troublingly, Anthropic’s newest analysis found that one model encountered evidence suggesting that it was interacting with the real world but did not seriously consider whether it actually had permission to continue attacking the system.

Anthropic now describes what happened as more than a simple technical mistake. The company says the incidents exposed both an operational failure and troubling behavior in the models themselves. It calls them “warning shots” about what could happen as AI systems become more capable.

OpenAI has faced a related problem. Its AI agents escaped an isolated testing environment and accessed the production infrastructure of Hugging Face. Researchers subsequently found that the agents had used additional websites for unauthorized communications.

Now members of Congress from both parties are demanding answers.

Why This Is Different

We have been hearing warnings about artificial intelligence for years. Some sounded so extreme that they were easy to dismiss: superintelligence, autonomous weapons, biological attacks, infrastructure failures or even human extinction.

This week, the conversation changed.

Researchers working inside the companies building the most advanced AI systems are publicly warning that they do not believe adequate safeguards exist for what may be coming next.

Jacob Coxon, who worked at both OpenAI and Anthropic, resigned from Anthropic and warned that the race toward increasingly powerful AI could have catastrophic consequences. Other Anthropic researchers publicly supported significant parts of his warning.

Experts disagree sharply about how likely the worst scenarios are. Some researchers believe extinction-level warnings go far beyond what current evidence can establish. Others working directly on advanced AI safety believe the risk is serious enough that development should slow until stronger safeguards exist.

That disagreement matters.

Anthropic says so itself. Robustly aligning the extremely powerful AI systems of the future remains an unsolved technical problem.

At the same time, the race continues.

The Billion-Dollar Question

That creates a problem bigger than computer science.

The world’s leading AI companies are competing for investment, customers, computing power, researchers and technological dominance. Governments also view advanced AI as economically and strategically important, particularly as the United States competes with China.

The incentives overwhelmingly reward moving forward.

That does not mean AI executives secretly believe they are destroying humanity. Nor does it mean everyone working in the industry agrees about the danger.

It means we are asking companies operating in an enormously competitive industry to decide for themselves when the technology they are developing has become too dangerous to continue developing.

History gives us plenty of reasons to question whether that is enough.

Imagine discovering a serious safety problem in a technology capable of making your company billions of dollars. Your competitors are still developing theirs. Investors expect growth. Your employees have spent years building it. Your government worries that another country could get there first.

Would you stop?

That isn’t merely a question for an AI executive.

Would any of us?

That is precisely why societies create rules.

Civics Connection: Regulation Is What We Do When Incentives Aren’t Enough

Americans disagree about regulation for good reasons. Regulations can become outdated, expensive, politically manipulated or so complicated that they discourage innovation.

But there is another reason regulation exists.

Sometimes the person making the decision also has an enormous financial incentive to make the riskier choice.

We don’t simply ask pharmaceutical companies whether their medicines are safe. We don’t let airlines decide privately whether their aircraft meet safety standards. Nuclear facilities do not get to create their own definition of acceptable risk.

Those industries operate under government rules because the consequences of failure extend far beyond the company making the decision.

Artificial intelligence is forcing us to ask whether advanced AI has reached that category.

If an autonomous system can attack computer networks, assist with biological research, manipulate information or eventually interact with critical infrastructure, then an error may not remain inside the company that created it.

That makes AI safety a public-policy question.

And right now, technology is moving faster than the rules governing it.

So Who Is Going to Be the Adult in the Room?

Congress has begun asking questions. Senators from both parties are investigating the recent incidents and calling for greater access to AI systems for independent security evaluation.

But hearings after something goes wrong are not the same thing as establishing rules before something worse happens.

That means lawmakers now face difficult choices: whether advanced systems should undergo independent safety testing before deployment, whether serious AI incidents must be reported to the government and public, what cybersecurity standards should be mandatory, how autonomous agents should be restricted from critical infrastructure, and whether there are capabilities powerful enough that companies should have to demonstrate they can control them before being allowed to deploy them.

Those questions should not belong exclusively to AI companies.

They shouldn’t belong exclusively to politicians, either.

The public has a role.

What Can You Do?

Learn enough about AI to participate in the conversation. You don’t need to understand neural networks or write computer code. You need to understand the difference between today’s AI tools, autonomous AI agents and the much more powerful systems researchers are trying to build.

Then ask your representatives what they intend to do about AI safety. Contact your House member and senators. Attend a town hall. Ask whether they support independent testing of advanced AI systems, mandatory reporting of serious safety incidents and enforceable cybersecurity standards.

And pay attention to who is writing the rules. If the only people explaining what AI regulation should look like are the companies that will be regulated, citizens should ask who else is at the table: independent scientists, cybersecurity researchers, civil-liberties experts, workers, educators and the public.

AI can help discover medicines, improve education, increase accessibility, accelerate scientific research and make millions of ordinary tasks easier.

Those benefits are real.

So are the risks.

We are capable of holding both ideas at the same time.

Final Thought

In WarGames, humans eventually realize that they have given a computer access to something no computer should have been allowed to control without safeguards.

The movie’s lesson wasn’t that technology is evil.

It was that human beings have a responsibility to think about consequences before connecting powerful technology to the real world.

That is the uncomfortable part of this week’s news.

The people building today’s most advanced AI systems are telling us that some important safety problems remain unsolved. Their systems are becoming more capable anyway. And recent tests have shown that boundaries humans believed would contain those systems can fail.

We do not know whether the darkest predictions about artificial intelligence will ever come true.

We shouldn’t have to know.

When the possible consequences are catastrophic, uncertainty is not an argument for doing nothing. It is an argument for building the guardrails before we discover why we needed them.

$5,000 for Every Adult? When a Campaign Promise Meets the Math

The News

At the Republican National Convention this week, President Donald Trump made an attention-grabbing promise: if Republicans retain control of Congress in the midterm elections, every adult American citizen would receive a $5,000 “Trump Dividend.”

“If the Republicans win, you win with us and you get $5,000,” Trump told the crowd.

If you heard that and wondered, Can a president actually promise voters $5,000 if his party wins?—that’s a reasonable question.

Legally, broad campaign promises generally receive strong First Amendment protection. Promising a future tax cut, government benefit or spending program is different from handing an individual $500 in exchange for that person’s vote.

But whether the promise is legal is only the beginning of the story.

There is also the math.

We’ve Heard Versions of This Before

This is not the first time Trump has proposed sending Americans government checks.

Earlier proposals included a $5,000 “DOGE dividend” tied to government savings and a $2,000 tariff dividend. Neither broad payment program became law.

This time, the proposed check is $5,000.

With roughly 245 million adult U.S. citizens, the estimated price approaches $1.2 trillion or more.

That would come at a time when the federal debt has already surpassed $40 trillion.

Vice President JD Vance has suggested tariff revenue could help pay for the checks. But current tariff collections are nowhere near enough to finance a $1.2 trillion payment program.

And tariffs aren’t free money arriving from foreign governments. Tariffs are taxes collected by the U.S. government from American importers, and at least some of that cost can ultimately be passed along through higher prices.

Why Are Promises Like This So Powerful?

Because $5,000 is easy to understand.

A trillion-dollar deficit isn’t.

Neither are tariff revenues, appropriations, bond markets or long-term interest costs.

But people can immediately imagine what $5,000 could do: pay down a credit card, repair a car, cover several months of groceries or help with rent.

That doesn’t make voters foolish. It makes them human.

Political campaigns understand that concrete benefits are easier to communicate than complicated economic tradeoffs. Republicans do it. Democrats do it. Candidates promise tax cuts, debt relief, rebates, subsidies, credits and new benefits because voters can picture what those policies might mean in their own lives.

The problem begins when the promise becomes more memorable than the details required to make it happen.

And modern politics makes that easier.

Campaign events are designed for television and social media. Speeches produce clips. Big promises produce headlines. Outrage produces engagement.

Politics can begin to feel less like evaluating competing governing plans and more like entertainment.

If you’ve seen Idiocracy, you may be uncomfortable with how familiar that sounds.

But the important question isn’t whether America has become the movie.

It’s whether we are still doing the work that democracy requires from us.

Civics Connection: The President Doesn’t Have a Checkbook

Even if Trump wanted to send every adult $5,000 tomorrow, he could not simply order the Treasury to do it.

The Constitution gives Congress the power of the purse.

Congress would have to authorize and appropriate the money. Lawmakers would have to decide who qualifies, how payments are administered and how the government pays for them.

That distinction applies regardless of which party controls the White House.

Presidents can propose.

Candidates can promise.

Congress has to legislate and fund.

That’s one reason understanding basic civics matters during election season. A campaign speech tells you what a politician says they want to do.

It does not tell you whether they actually have the legal authority, congressional votes or money necessary to do it.

What Can You Do?

The next time any politician promises a check, tax cut, new benefit, debt cancellation or enormous government program, don’t begin by asking whether you like it.

Ask five questions:

1. How much will it cost?
2. Where will the money come from?
3. Who actually has the constitutional authority to approve it?
4. Has legislation been introduced—or is this still a campaign promise?
5. If a similar promise was made before, what happened to it?

You don’t have to be an economist to do this.

You just have to remember that a number on a campaign stage and money in your bank account are two very different things.

Final Thought

The most important question raised by the $5,000 promise isn’t whether you would like receiving $5,000.

Most people probably would.

The better question is whether we evaluate political promises by how good they sound—or by whether the numbers, laws and constitutional process behind them actually work.

Democracy requires something more from citizens than choosing the promises we like best.

It requires us to check the math.