All News (Page 403)

Lawyers for the Department of War and Anthropic sparred in a California federal court on Tuesday over Anthropic’s challenge to the Pentagon labeling it a “supply-chain risk” to national security and banning all government contractors from using the company’s sweeping AI tools. Anthropic is seeking an injunction barring enforcement of that order.

The case—which involves a historic first in that the Department of Defense, informally renamed the Department of War (DOW) by the Trump administration, labeled a U.S.-led  business as a supply-chain risk to national security—is rooted in a contract negotiation that escalated quickly. The DOW wanted to add a blanket “all lawful use” clause to its contracts with the AI firm so the military could use Anthropic’s Claude tool for any legal purpose.

The presiding judge in the case expressed doubts about the sweeping authority the Pentagon had wielded in the case. Federal District Judge Rita Lin said she would issue a ruling on Anthropic’s legal challenge “in the next few days,” and spent Tuesday’s hearing asking the parties questions about their disagreement.

A heated dispute over how to use AI

During contract negotiations with the Pentagon in February, Anthropic balked at the possibility of the military using Claude for lethal autonomous warfare and mass surveillance of Americans, and attempted to insist on provisions expressly forbidding such use. Anthropic, led by founder Dario Amodei, said it hasn’t thoroughly tested those uses and doesn’t believe they work safely. The DOW claimed those guardrails were unacceptable and that military commanders need latitude to make determinations on missions. 

On Feb. 27, President Trump posted on Truth Social directing “EVERY” federal agency to “IMMEDIATELY CEASE” all use of Anthropic’s tools. That same day in a post on X, Secretary of War Pete Hegseth labeled Anthropic a “supply-chain risk” and said “no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity with Anthropic.” The risk label is usually reserved for nation states, foreign adversaries, and other threats. 

Anthropic followed by filing a lawsuit on March 9, alleging the government “retaliated against it” for expressing its views on safety guardrails and had violated the First Amendment in doing so. It also claimed the government violated the process laid out in the Administrative Procedure Act and the Fifth Amendment’s right to due process.

In briefs in the case and in court on Tuesday, the government said the administration’s actions were in response to Anthropic’s refusal to implement certain terms in its contract, and argued free speech wasn’t at issue in the case. Deputy Assistant Attorney General Eric Hamilton said the government has unrestricted power to determine which companies it will contract with. Hamilton said Anthropic’s conduct had raised concerns that future software updates could be used as a “kill switch” to keep the AI from functioning in military operations.

District Judge Rita F. Lin was skeptical, and in her opening statements described the case as a “fascinating public policy debate” over Anthropic’s position versus the government’s military needs, but said her role wasn’t to “decide who is right in that debate.”

Rather, Lin said the real question to be decided by the court was whether the government “violated the law” when it went beyond just not using Anthropic’s AI services and finding a more permissible AI vendor to work with. 

“After Anthropic went public with this contracting dispute, defendants seemed to have a pretty big reaction to that,” Lin said. 

The reactions included banning Anthropic from ever having a government contract—excluding other entities like the National Endowment for the Arts from using it to design a website; Hegseth’s directive that anyone who wants to do business with the U.S. military sever their commercial relationship with Anthropic; and designating Anthropic as a supply-chain risk. 

“What is troubling to me about these reactions is that they don’t really seem to be tailored to the stated national security concern,” said Lin. If the concern is about chain of command, DOW could just stop using Claude and go on its way, she said.  

“One of the amicus briefs used the term ‘attempted corporate murder,’” she added. “I don’t know if it’s murder, but it looks like an attempt to cripple Anthropic. And specifically my concern is whether Anthropic is being punished for criticizing the government’s contracting position in the press.”

Parties rally behind Anthropic

The amicus, friend-of-the-court, briefs in the case have drawn a variety of voices including from Microsoft, retired military officers, and engineers and researchers from OpenAI and Google. Nearly all support Anthropic’s position seeking an injunction of the supply-chain risk designation.

The brief Lin referred to came from investors and the “Freedom Economy Business Association.” The brief referred to an X post written by Dean Ball, Trump’s former senior policy advisor for AI and emerging tech. 

“Nvidia, Amazon, Google will have to divest from Anthropic if Hegseth gets his way,” Ball wrote. “This is simply attempted corporate murder. I could not possibly recommend investing in American AI to any investor; I could not possibly recommend starting an AI company in the United States.”

The American Federation of Government Employees, a union of 800,000 federal workers, said in its amicus brief that the Trump administration had a pattern of using national security concerns as a pretext for retaliation against free speech.

Microsoft wrote that a ban on Anthropic would hurt its own business, and could chill future defense-industry investment and engagement with AI. 

The Human Rights and Technology Justice Organization brief didn’t take a position who should win in court, but argued against militarized AI broadly, and stating that its use could lead to catastrophic human rights risks.  

Lin said she’ll issue an opinion this week. 

This story was originally featured on Fortune.com

Moldova’s Parliament voted on Tuesday to impose a state of emergency in the country’s energy sector after Russian strikes on neighboring Ukraine’s energy grid disconnected a key power line linking Moldova to Romania.

The overnight strikes triggered the disconnection of the high-voltage Isaccea-Vulcanesti power line, which links southern Moldova to EU member Romania, after which Moldovan authorities urged citizens to consume electricity “rationally” during peak hours while repairs were underway.

Seventy-two lawmakers in the 101-seat legislature approved the measure that will last for 60 days. No one voted against and 18 abstained.

“What is happening in the energy sector today is not an accident,” said Moldovan Prime Minister Alexandru Munteanu. “Russia’s attacks on the civilian energy infrastructure in Ukraine represent a war crime, but also an attack on us, here in the Republic of Moldova … Russia is the only one responsible for this.”

The state of emergency will begin on Wednesday. It will allow authorities to “act faster: mobilize additional resources, protect critical infrastructure and, if necessary, take additional measures to limit the effects of the crisis,” Munteanu added. “We remain vigilant and act for the safety of every citizen … This is not a measure of panic, it’s a measure of responsibility.”

Moldova’s Soviet-era energy systems remain interconnected with Ukraine, and the country has suffered periodic outages since Russia fully invaded Ukraine in 2022. Moldova’s energy minister Dorin Junghietu said estimates to repair the damaged power line is around five to seven days.

Moscow has repeatedly targeted Ukraine’s civilian infrastructure, such as dams and river ports, throughout the full invasion. The impact of the war next door has reverberated across Moldova, a former Soviet republic with EU candidate status, since the full invasion began.

Last week, tens of thousands of Moldovans were left without water after another Russian strike on a hydroelectric plant in Ukraine resulted in oil polluting a major river that flows through both countries.

The Ukrainian plant is situated about 15 kilometers (9 miles) upstream from Moldova’s northern border with Ukraine and supplies water to about 80% of Moldova’s population of about 2.5 million. In January, Moldova experienced major power outages, including in the capital, Chisinau, after a disruption to a power line from Ukraine caused a drop in voltage.

Moldovan President Maia Sandu blamed Moscow directly on Tuesday, saying that “Russia continues to deliberately undermine the security of the Republic of Moldova and endanger the lives of our citizens.”

“After the bombing of the Ukrainian hydroelectric power plant … tonight, a new brutal attack led to the disconnection of the Isaccea-Vulcanesti line, which in certain periods provides 60-70% of our electricity consumption,” she wrote on Facebook. “All these are not accidents, but deliberate actions of Russia to weaken and leave Moldova in the dark.”

Russia has repeatedly denied it is trying to destabilize Moldova.

This story was originally featured on Fortune.com

The year was 1974 and President Richard Nixon had dispatched his Secretary of State Henry Kissinger to Saudi Arabia to strike a secret deal. Three years earlier, in August 1971, Nixon had already administered the “shock” that ended the Bretton Woods system governing global finance since World War II — suspending the dollar’s convertibility to gold in a televised address that transformed every major currency overnight. By 1973, the system had fully unraveled.

The world wouldn’t know for another 50 years what Nixon and Kissinger replaced it with, striking a deal that would quietly govern the global economy for the next half-century. Riyadh agreed to price and trade its oil in U.S. dollars and channel its petroleum windfalls back into U.S. Treasury bonds; in return, Washington promised military aid, equipment, and security guarantees—a deal that would quietly govern the global economy for the next half-century.

The existence of this secret agreement wasn’t even publicly confirmed until 2016, when Bloomberg News filed a Freedom of Information Act request with the National Archives. Other OPEC members had followed Riyadh’s lead in the years since, locking in the dollar as the indispensable currency of the modern world. The arrangement had a name only economists used: the “petrodollar” system. It was America’s greatest secret weapon—and today, in the churning waters of the Persian Gulf, it faces its most serious threat since its creation.

Getty Archive

The Chokepoint That Moves the World

The Strait of Hormuz is a sliver of water barely 21 miles wide at its narrowest point, separating Iran from Oman. It does not look like the axis of the global economy on a map. But in 2024, roughly 20 million barrels of oil and petroleum products passed through it every day—about 20% of global petroleum liquids consumption and approximately 25% of all seaborne oil trade on Earth.

Qatar and the UAE rely on the strait for virtually all of their LNG exports, representing about 20% of global LNG trade. The bulk of the crude leaving the strait heads to China, India, Japan, South Korea, and other Asian markets, which absorb the overwhelming majority of Hormuz volumes. When Iran slammed shut this door, it didn’t just disrupt shipping lanes. It placed maximum stress on the architecture of dollar dominance at its most physical chokepoint.

For weeks, President Trump has scrambled to respond. He issued a 48-hour ultimatum threatening to “obliterate” Iran’s power plants if Tehran did not reopen the strait. Iran countered by threatening to mine the Persian Gulf and target American energy infrastructure in the region. Trump then postponed his deadline amid what the White House described as diplomatic progress—a face-saving maneuver that former Defense Secretary James Mattis warned could ultimately cede the strait to Tehran’s influence. “You’d see a tax for every ship that goes through,” Mattis said during the CERAweek by S&P Global conference, as reported by Politico.

The administration has cycled through a list of increasingly desperate options, from building a naval coalition—with Trump saying he’d approached “about seven” countries—to a reported proposal to wind down the conflict without resolving the Hormuz closure. As of Monday, Trump told CNBC: “We are very intent on making a deal.”

The $39 Trillion Liability No One Is Talking About

While the gunboat diplomacy dominates the headlines, the more existential danger may be unfolding in the bond market. The U.S. national debt crossed $39 trillion on March 18, 2026, a milestone reached just weeks into the war in Iran. The speed of accumulation is staggering, and the timing could not be worse: interest costs on the debt are projected to become the fastest-growing line item in the federal budget over coming decades, and the U.S. has already suffered credit downgrades from all three major ratings agencies — S&P in 2011, Fitch in 2023, and Moody’s in May 2025.

The reason this matters geopolitically—not just fiscally—goes back to that 1974 handshake. The petrodollar system created a perpetual buyer for U.S. Treasury bonds in the form of oil-exporting nations. The mechanism was elegant in its simplicity: oil exporters accumulated vast dollar surpluses and parked them in U.S. Treasuries, which Washington was only too happy to supply. Saudi Arabia alone held $149.5 billion in U.S. Treasury securities as recently as December 2025 — a figure that, notably, rose by $12 billion over the course of last year, even as Riyadh declined to formally renew the original petrodollar agreement. That recycling loop is what allowed Washington to borrow cheaply, run persistent deficits, and still maintain the world’s reserve currency.

In 1965, French Finance Minister Valéry Giscard d’Estaing was widely credited with a memorable criticism of the Bretton Woods system that predated the petrodollar regime as an “exorbitant privilege” enjoyed by America, with the U.S. dollar serving as the world’s reserve currency. In the 1970s, once President Richard Nixon ended Bretton Woods by decoupling the dollar from gold, that privilege was revived in oil and debt, requiring every country on Earth to accumulate dollars simply to buy oil, and then reinvest those dollars back into American debt. Former Greek Finance Minister Yanis Varoufakis, a heterodox economist whose work sits outside mainstream consensus but who captures something real about the system’s coercive logic, calls this “the global minotaur,” likening the U.S. to the ancient king of Crete who held international trade captive to tribute that would feed the monster within his labyrinth.

The unfolding crisis in the Strait of Hormuz is exposing America’s privilege as a vulnerability. The speaker of Iran’s parliament delivered a warning this week that rattled bond traders: financial institutions backing the U.S. military budget were “legitimate targets,” and buyers of U.S. Treasury bonds were purchasing “an attack on your headquarters and assets.” It was theatrical. It was also a signal—that America’s $39 trillion debt load could become a pressure point in an escalating conflict.

Bernard CHARLON/Gamma-Rapho via Getty Images

The De-Dollarization Accelerant

Even before Iran closed the strait, cracks in the petrodollar system were visible—though economists caution that “cracks” is very different from “collapse.” The U.S. dollar’s share of global foreign exchange reserves has fallen to roughly 56.9% as of Q3 2025, its lowest level since 1995 and down from a peak of 72% in 2001, according to IMF COFER data. That is a real, multi-decade structural decline. But here is the critical detail most alarming headlines omit: the IMF itself found that roughly 92% of the quarterly decline recorded in mid-2025 was driven by exchange-rate movements—the dollar weakening made non-dollar holdings appear larger—not by central banks actively dumping dollars. (The weakening of the dollar is a whole other story, but Trump’s tariff regime, the exploding national debt and inflation expectations rising are all widely seen as major factors, along with the “Sell America” trade.) There is a meaningful difference between erosion and exodus.

The Chinese yuan, despite years of BRICS advocacy and aggressive promotion of yuan-denominated oil contracts, represents just 2.1% of global reserves. The euro holds second place at roughly 20%, but no single currency has emerged as a credible heir apparent. The Federal Reserve’s own 2025 assessment found that dollar reserve share has been “basically unchanged since 2022,” and that U.S. sanctions on Russia following the Ukraine invasion did not trigger the feared mass reallocation out of dollars.

Saudi Arabia, in fact, chose not to formally renew the petrodollar agreement in June 2024, but the informal, secretive nature of the deal makes it hard to evaluate whether this was a policy change. What the data actually show is that the Saudis still price the overwhelming majority of their oil in dollars; global oil markets remain structurally dollar-denominated; and the network effects that sustain that arrangement—every buyer, every trader, every swap desk globally priced in dollars—do not unwind overnight. As the Hinrich Foundation noted as recently as last week, “talk of de-dollarization is prone to hyperbole,” even as the IMF data confirms a slow, real erosion, because of the weakening dollar.

What the Hormuz crisis means isn’t an end to the petrodollar—it is a threat to accelerate a shift that was previously moving at a glacial pace by raising the geopolitical temperature around a system that had long operated below the radar. Every week the strait stays closed, Asian economies are forced to test alternative supply chains — existing bypass routes like Saudi Arabia’s East-West Pipeline and the UAE’s Abu Dhabi Crude Oil Pipeline to Fujairah absorb only a fraction of normal Hormuz volumes, meaning the pressure to find workarounds is real — and, at the margin, alternative payment mechanisms. If the crisis is resolved in weeks, those experiments are quickly abandoned. If it drags into months, habits begin to form. The dollar’s dominance is not a cliff—it is a long, slow slope—and the question the Hormuz standoff raises is not whether America falls off the edge today, but whether Trump’s handling of this crisis steepens the gradient.

There is a long slope down from this exorbitant privilege, as there remains no obvious successor to the dollar. And for all of Iran’s saber-rattling, its closure of the Strait of Hormuz is not a sophisticated financial weapon aimed at the dollar’s structural foundations. Rather, it is a desperate act of asymmetric warfare by a regime under unprecedented military pressure—a tactical move, not a strategic master plan.

Economic models analyzing the Hormuz closure project global GDP losses ranging from $330 billion in a short conflict to $2.2 trillion if it drags on. Those are serious numbers. But economic disruption is not the same as dollar displacement. If anything, crisis conditions historically drive a flight to dollars, not away from them, because the deep liquidity and institutional trust underpinning the dollar have no match.

Still, the U.S. should pay attention to its own abuses of that privilege. The consequences of sustained erosion are not abstract. The IMF has flagged that the U.S. is more fiscally imbalanced than its peers and that without reserve currency status, its credit position would be far worse. Foreign demand for U.S. Treasuries could weaken, forcing Washington to offer higher interest rates to attract buyers, which would feed directly into the cost of servicing the $39 trillion debt, creating a feedback loop of deficits and borrowing costs that could spiral well beyond the projections of today’s fiscal models. The Committee for a Responsible Federal Budget forecasts annual interest payments of $1 trillion and climbing. Add a prolonged oil shock and the ingredients for a genuine fiscal crisis in the medium term are present.

Trump has said he wants a deal. But his usual playbook, what Yale Management professor Jeffrey Sonnenfeld calls his “ten commandments,” a framework of transactional pressure tactics that served him well against conventional partners, is not working with an adversary like Iran with little left to lose. And time is the one thing the architecture of American financial dominance may no longer have in abundance. The petrodollar system was built in secret in 1974 and sustained quietly for 50 years. The Strait of Hormuz has now made its fragility visible to the entire world, whether or not Trump understood that ordering strikes on Iranian energy infrastructure and military targets would expose the financial architecture those bombs were implicitly defending.

For this story, Fortune journalists used generative AI as a research tool. An editor verified the accuracy of the information before publishing.

This story was originally featured on Fortune.com

Immigration and Customs Enforcement (ICE) agents are now doing the jobs of Transportation Security Administration (TSA) workers, but one big difference is that they’re getting paid for it.

The partial government shutdown, entering its 44th day, has left more than 50,000 TSA officers without pay, leading to more than 450 workers quitting and thousands calling out of work, according to Department of Homeland Security (DHS) data. President Donald Trump ordered ICE agents to U.S. airports to guard exits and check IDs to allow TSA agents to more quickly conduct security scans at checkpoints. Trump said ICE personnel can conduct immigration checks and arrests, though it’s not their primary purpose.

These ICE agents will continue to receive pay, even as TSA officers forgo earnings for five weeks, and the disparity has shined light on the pay differences between the two groups carrying out similar tasks. 

According to TSA Career, a nongovernment website, the starting salary for TSA agents is $34,454, with the average officer salary between $46,000 to $55,000. The highest-paid TSA employee earns around $163,000.

Meanwhile, deportation officers are paid between $51,632 and $84,277, according to a job posting on a government website. ICE agents are also eligible for a $50,000 signing bonus, often given in $10,000 per-year increments, putting total pay at nearly double that of a TSA officer.

The American Federation of Government Employees, the largest union representing federal employees and the only one representing TSA workers, claimed ICE agents were unqualified to replace and work alongside TSA officers at airports as they lacked the appropriate training. 

Everett Kelley, president of the union, demanded TSA agents be paid, rather than replaced by other government employees.

“Our members at TSA have been showing up every day, without a paycheck, because they believe in the mission of keeping the flying public safe,” Kelley said in a statement on Sunday. “They deserve to be paid, not replaced by untrained, armed agents who have shown how dangerous they can be.”

Why are ICE agents getting paid while TSA agents are not?

The reason behind why ICE agents continue to be paid while TSA agents work without paychecks comes down to where these two agencies receive their respective funding. 

Despite both being under the umbrella of the DHS, ICE received a share of its funding from Trump’s One Big Beautiful Bill Act, which pumped ICE with about $75 billion over five years. TSA is funded through DHS, which the government ceased funding in February as Democrats demanded reforms on ICE following the fatal shootings of two U.S. citizens in Minneapolis in January.

On Tuesday, the Senate closed in on a proposal that would fund much of the DHS, including providing pay to TSA agents. The funding resumption would exclude ICE operations.

The libertarian think tank the Cato Institute, called funding under the One Big Beautiful Bill Act “shutdown proof” in a February report, arguing Republicans “short-circuited the system of checks and balances” by shifting funding for immigration enforcement and defense spending outside of normal appropriations, wrestling in less oversight and greater partisanship in the budgeting process.

But the breakdown of who gets paid and who doesn’t during a government shutdown is a failure of a budget structure that goes beyond a particular administration, according to Linda Bilmes, a public finance expert and senior lecturer at Harvard University’s Kennedy School of Government. 

The decision of who is deemed essential and nonessential, for example, depends on department personnel, while salary appropriations can be impacted by lapses in the congressional budget, which occur multiple times a year.

“There is this overarching dysfunction of the entire process,” Bilmes told Fortune during the government shutdown in October 2025. (During this shutdown, law enforcement officers including both ICE and TSA agents received “super checks” as well as overtime pay). “Every time you get into one of these situations—which has been on average four times a year for the last four to five years—there is an arbitrariness in who ends up being paid for their work, who ends up working, who ends up being furloughed.

“The arbitrariness is almost inherent in this dysfunction—a feature as well as being a bug,” she added.

This story was originally featured on Fortune.com

Hello and welcome to Eye on AI. In this edition…AI’s reliability problem…Trump sends an AI legislation blueprint to Congress…OpenAI consolidates products into a super app and hires up…AI agents that can improve how they improve…and does your AI model experience emotional distress?

Like many of you, I’ve started playing around with AI agents. I often use them for research, where they work pretty well and save me substantial amounts of time. But so-called “deep research” agents have been available for over a year now, which makes them a relatively mature product in the AI world. I’ve also started trying the new crop of computer-using agents for other tasks. And here, my experience so far is that these agents are highly inconsistent.

For instance, Perplexity’s Computer, which is an agentic harness that works in a virtual machine with access to lots of tools, did a great job booking me a drop-off slot at my local recycling center. (It used Anthropic’s Claude Sonnet 4.6 as the underlying reasoning engine.) But when I asked it to investigate flight options for an upcoming business trip, it failed to complete the task—even though travel booking is one of those canonical use cases that the AI companies are always talking about. What the agent did do is eat up a lot of tokens over the course of 45 minutes of trying.

Last week, at an AI agent demo event Anthropic hosted for government and tech policy folks in London, I watched Claude Cowork initially struggle to run a fairly simple data-sorting exercise in an Excel spreadsheet, even as it later created a sophisticated budget forecasting model with seemingly no problems. I also watched Claude Code spin up a simple, text-based business strategy game I asked it to create that looked great on the surface, but whose underlying game logic didn’t make any sense.

Assessing AI agents’ reliability

Unreliability is a major drawback of current AI agents. It’s a point that Princeton University’s Sayash Kapoor and Arvind Narayanan, who cowrote the book AI Snakeoil and now cowrite the “AI As Normal Technology” blog, frequently make. And a few weeks ago they published a research paper, co-authored with four other computer scientists, that tries to think systematically about AI agent reliability and to benchmark leading AI models.

The paper, entitled “Towards a Science of AI Agent Reliability,” notes that most AI models are benchmarked on their average accuracy on tasks, a metric that allows for wildly unreliable performance. Instead, they look at reliability across four dimensions: consistency (if asked to perform the same task in the same way, do they always perform the same?); robustness (can they function even when conditions aren’t ideal?); calibration (do they give users an accurate sense of their certainty?); and safety (when they do mess up, how catastrophic are those mistakes likely to be?).

They further broke these four areas into 14 specific metrics and tested a number of models released in the 18 months prior to late November 2025 (so OpenAI’s GPT-5.2, Anthropic’s Claude Opus 4.5, and Google’s Gemini 3 Pro were the most advanced models tested). They tested the models on two different benchmark tests, one of which is a general benchmark for agentic tasks while the other simulates customer-support queries and tasks. They found that while reliability improved with each successive model release, it did not improve nearly as much as average accuracy figures. In fact, on the general agentic benchmark the rate of improvement in reliability was half that of accuracy, while on the customer service benchmark it was one-seventh!

Reliability metrics depend on the task at hand

Across the four areas of reliability the paper examined, Claude Opus 4.5 and Gemini 3 Pro scored the best, both with an overall reliability of 85%. But if you look at the 14 sub-metrics, there was still plenty of reason for concern. Gemini 3 Pro, for example, was poor judging when its answers were likely accurate, at just 52%, and terrible at avoiding potential catastrophic mistakes, at just 25%. Claude Opus 4.5 was the most consistent in its outcomes, but its score was still only 73% consistent. (I would urge you to check out and play around with the dashboard the researchers created to show the results across all the different metrics.) 

Kapoor, Narayanan, and their co-authors are also sophisticated enough to know that reliability is not one-size-fits all metric. They note that if AI is being used to augment humans, as opposed to fully automating tasks, it might be ok for the AI to be less consistent and robust, since the human can act as a backstop. But “for automation, reliability is a hard prerequisite for deployment: an agent that succeeds on 90% of tasks but fails unpredictably on the remaining 10% may be a useful assistant yet an unacceptable autonomous system,” they write. They also note that different kinds of consistency matter in different settings. “Trajectory consistency matters more in domains that demand auditability or process reproducibility, where stakeholders must verify not just what the agent concluded but how it got there,” they write. “It matters less in open-ended or creative tasks where diverse solution paths are desirable.”

Either way, Kapoor, Narayanan, and their co-authors are right to call for benchmarking of reliability and not just accuracy, and for AI model vendors to build their systems for reliability and not just capability. Another study that came out this week shows the potential real-world consequences when that doesn’t happen. AI researcher Kwansub Yun and health consultant Claire Hast looked at what happens when three different AI medical tools are chained together in a system, as might happen in a real health care setting. An AI imaging tool that analyzed mammograms had an accuracy of 90%, a transcription tool that turned an audio recording of a doctor’s examination of a patient into medical notes had an accuracy of 85%, and these were then fed to a diagnostic tool that had a reported accuracy of 97%. And yet when used together their reliability score was just 74%. That means one in four patients might be misdiagnosed!

A foolish consistency may be the hobgoblin of little minds, as Ralph Waldo Emerson famously said. But, honestly, I think I’d prefer that hobgoblin to the chaotic gremlins that currently plague our ostensibly big AI brains. 

Jeremy Kahn
jeremy.kahn@fortune.com
@jeremyakahn

Before we get to the news, I want to encourage everyone to read my Fortune colleague Allie Garfinkle’s awesome feature story about Cursor. Cursor is the AI coding startup that as recently as four months ago was a Silicon Valley darling, but which many people now think may be facing an existential threat because of new coding agents, such as Anthropic’s Claude Code, that seemingly obviate the need to use Cursor. Allie’s story lays bare all the contradictions around this company—how it has continued to see record revenue growth, even as many in Silicon Valley now harbor doubts about its survival; how it is racing to train its own coding agents, pivoting from the developer-centric coding interface that made it so popular with programmers in the first place; how its impossibly young CEO Michael Truell works under a portrait of Robert Caro, the biographer whose projects often lasted decades, while Cursor needs to operate in an industry in which a year can feel like a century. Allie’s story is definitely worth the time.

This story was originally featured on Fortune.com