My girlfriend and I treat the outside of our apartment fridge like one big collage.

In the past few years, after attending wedding after wedding after wedding, we have unwittingly created a standing scrapbook of “Save the Date!” magnets and floral-patterned invitations, among other gilded mementos. But recently, we decided it was time to give the fridge collage a refresh. And in doing so, we realized something depressing: The majority of the happy-looking couples pictured were either now divorced, breaking up or in the process of seriously re-evaluating their relationship.

As a friend, I’ve learned so much from my newly divorced pals, and I admire the resilience, optimism and strength they demonstrated in the face of their breakups. But as a personal-finance columnist, I have also taken a lot of money lessons from these new divorcées—and I believe the insight is relevant to more than just other married couples.

One friend was able to rely on her prenuptial agreement to save her close to $100,000 in the wake of a speedy split; another woman I know escaped a financially abusive relationship with less than a grand to her name remaining in her savings account.

Darla Gale , a California-based therapist and founder of Heartstrings Counseling, says she often talks with clients about making meaning from these relationships and the accompanying financial fallout. Sharing lessons and experiences is one way to do so. “Money comes up a lot in my practice, because financial hardships are one of the reasons for divorce,” she says. “It can be very empowering to say, ‘I can do this on my own. I just didn’t have the tools to do it on my own.’ ”

Together isn’t always better

The dissolution of a relationship can bring a host of financial lessons that shape how we talk about money in our relationships going forward—whether we were the bride, the groom or the smiling person snapping a selfie with the cake.

In my previous reporting, I’ve researched quite a bit about just how beneficial it can be for couples to combine finances. Studies show this enables them to maximise their potential to grow wealth and even leads to both partners feeling happier in the relationship.

But after witnessing their first wave of divorces, many younger people are embracing a different approach. “I’m doing a lot of prenuptial agreements for people who are younger, and they are for the most part wanting to keep their money separate, which is interesting and very different from the way prenups were done 10 to 20 years ago,” says Lisa Zeiderman , a New York-based divorce lawyer.

Gale says she now often sees couples doing both: sharing expenses in one joint account, for example, while still maintaining separate funds “so it doesn’t feel like one person is more in control.”

Inspired by this approach, my girlfriend and I have adopted something similar. To save for a down payment on a future house, we contribute near-equal amounts to a joint high-yield savings account that we opened together. We also share a credit card to use when we buy groceries, pay our dog’s vet bills or handle other shared expenses.

But at the same time, we both maintain separate checking accounts and saving accounts, for our own individual needs. This way, should disaster ever strike, we’re both able to maintain a level of personal autonomy and build our own lives anew.

After all, as Gale put it, “Autonomy equals equality.”

Older is better…and worse

Here’s the good news I’ve learned while looking at divorce through a personal (and personal-finance) lens: On the whole, millennial couples are actually divorcing at lower rates than people of previous generations at similar ages.

Economist Brett House , a professor of professional practice in the economics Division at Columbia Business School, attributes this decline to two important things: Millennials are getting married at later ages. And they’re likelier to have received more education by the age of their first marriage.

But waiting to tie the knot doesn’t protect you against the financial fallout of a divorce—if anything, marrying later means both parties might have had more time to accumulate more assets that could lead to financial conflict once split.

“People may be more mature and more aware of themselves and clearer about what they want from a relationship than was the case in earlier years,” House says.

Gale recommends clearly stating those financial priorities at the start of a relationship, when everything still seems rosy. “Don’t be sneaky about it, but go into the relationship and say ‘I’m going to have a separate bank account, whether that is for going out or having a ‘fun’ fund but autonomy is really important to me,’ ” she says.

And watch your potential partner’s reaction to such a conversation—that can also inform your decision to build a life with them.

Don’t compromise your career

In her many years advising women and working in divorce courts, Zeiderman says she often gives young to-be-married women the same advice: Don’t lose sight of your career.

Zeiderman has seen firsthand how difficult it is for many people, in compromising their own professional trajectories, to rebuild their earnings post-divorce.

“So many decide, frankly, to let the other person build their career and then at the end of the day, you cannot make up for this in spousal support, you cannot make up for it in the distribution of their assets,” she said. “Stay in the workforce. You must.”

We like to think that is easy to do, but in practice, both partners prioritising their careers isn’t easy.

This summer, I’ll have just embarked upon a new career as a freelance writer and podcast host; at the same time, my girlfriend, also a journalist, will be traveling throughout France covering the summer Olympics. We talked in advance about how important this summer will be for us, work-wise, and agreed that although our relationship—and our patience!—may be tested, affording each other some grace during this high-pressure time will go a long way. Building our careers will benefit our household bottom line, but it’s also insurance for both of us.

The ‘what if we broke up’ talk

“The forethought is easier to do when things are good,” Zeiderman told me. And her words stuck with me.

I thought back to the dozens of conversations I’ve had with my divorced friends. So many of them regretted never discussing money goals, habits or strategies with their partners in advance of marriage. Far too often, they said, they worried bringing up finances would dampen the excitement of the honeymoon phase; then, months or years later, the unsaid words curdled and soured with resentment.

My girlfriend and I keep a standing monthly “money date” on the calendar. We make the time to curl up on the couch, just the two of us, and review our household finances. For us, that looks like going over the bank statement and checking on progress made toward our financial goals. On past money dates, we’ve compared prices for coming purchases, paid down debt and even shared how things stand in our separate personal accounts.

No one likes paying bills or calculating debt totals, but we work hard to make the process as painless as possible. But this coming month, I think the topic I have planned may be our most uncomfortable one yet.

That’s because, inspired by Zeiderman and Gale, this time I have something particularly “special” planned for our money date: the “What if we broke up?” conversation.

I want to plan out how we would divide our shared assets and then put these details in writing—together. If something were to happen in the future, we could hopefully put aside our differences to rely on this plan and make the detangling of our finances much simpler.

Luckily for me, my girlfriend knows me well enough by now to see the romance hiding in this gesture.

As migration hits record levels worldwide, a debate is building among economists over whether some industries are becoming too dependent on foreign labour.

Many business owners say that bringing in low-skilled foreign workers has become essential, as local populations age and labor forces shrink. In rural Wisconsin, John Rosenow says it is impossible to find locals to work on his 1,000-acre dairy farm. He relies on 13 Mexican immigrants, up from eight to 10 a decade ago. That has enabled him to avoid making costly investments in robots that can help milk cows, as some other dairy farmers have.

“We get really good people,” Rosenow says. With immigrant labor, “I’m pretty sure if I wanted to double employment, I could get it done within a week.”

To some economists, however, dependence on imported workers is approaching unhealthy levels in some places, stifling productivity growth and helping businesses delay the search for more sustainable solutions to labor shortages.

Those solutions could include bigger investments in automation, or more radical restructurings such as business closures, which are painful but may be necessary long-term, these economists say.

“Once industry is organised in a certain way and the structure encourages employers to recruit migrants, it can be very hard to turn back,” said Martin Ruhs , a professor of migration studies in Florence, Italy. “In some cases, policymakers should ask, does it make sense?” said Ruhs, who is also a former member of the U.K. Migration Advisory Committee, which advises the British government on migration policy.

The debate is likely to heat up further as Western societies teeter closer to a demographic abyss . For the first time since World War II, the working-age population is shrinking across advanced economies. The European Union’s working-age population will shrink by one-fifth through 2050, according to a recent report by German insurer Allianz .

There are ways to offset that trend, such as encouraging older workers to delay retirement. But importing foreign labour is often the easiest option, given the supply of available workers in places such as Latin America or Africa.

Immigration also provides a rush of economic growth as migrants boost populations and spend money, even when it elicits blowback from conservative groups, as it has in the U.S. and Europe.

Immigration is now running two to three times above pre pandemic levels across major destination countries including Canada, Germany and the U.K. In the U.S., 3.3 million more migrants arrived than left last year, compared with a 2010s average of around 900,000.

Three-quarters of farmworkers and 30% of construction and mining workers in the U.S. today are migrants. Overall, immigrants made up 18% of the U.S. workforce in 2021, compared with 16% a decade earlier, according to the Organization for Economic Cooperation and Development, a Paris-based club of mostly rich countries.

Despite promising for decades to curb immigration, the U.K. has seen a surge since its 2020 exit from the EU, as businesses scramble for employees. More than 27% of the National Health Service’s nurses are from abroad today, up from around 14% in 2013. In Germany, roughly 80% of slaughterhouse workers are migrants, unions estimate.

Downsides of over reliance

Increased reliance on low-skilled imported labor can lead to weaker productivity growth, which ultimately determines how fast economies can expand, some economic research suggests.

A 2022 study in Denmark found that firms with easy access to migrant workers invested less in robots. Research in Australia and Canada suggests that migrants could keep weak firms alive, weighing on overall productivity.

Labour productivity growth has been sluggish across advanced economies in recent years. In the U.S. and U.K. farming sectors, productivity has flatlined for a decade or longer. In Japan and Korea, which have more restrictive immigration policies, it increased by around 1.5% a year, OECD data show.

Finding the right balance between allowing some migration, which can help restore dynamism in aging countries, and avoiding over dependence is hard. In many industries, there is no obvious alternative to foreign workers.

Going cold turkey would send prices for products made from migrant labour higher. It would also leave many people in poorer countries with fewer options to pursue better lives.

Anna Boucher , a global migration expert at the University of Sydney, says that some low-skilled migration is probably necessary in the short term due to skills shortages. Without it, some childcare services in Australia would shut down and vegetables would die in the fields.

Economic research suggests that an influx of high-skilled migrants, such as scientists and engineers, can actually lift firms’ productivity and boost local workers’ wages and employment opportunities.

Economists are more divided when it comes to lower-skilled migrants. Such workers are also more easily replaced, including in industries that seem unlikely candidates for automation.

In the Czech Republic, some farmers are using artificial-intelligence-driven robots to monitor and harvest strawberries. Israeli startup Tevel Aerobotics Technologies has developed fruit-picking drones. Fieldwork Robotics, a U.K. company, recently started selling raspberry-picking robots, which stand 6 feet tall with four plastic arms.

Yet for governments, pursuing reforms that boost productivity and allow weaker firms to die is a lot harder than increasing immigration, said Dan Andrews , a productivity expert at the OECD.

“Some countries may have taken the easy way out,” he said.

Pushback from businesses

Hoping to accelerate automation in agriculture, the U.K. government is pouring money into farm technology. It is also considering abolishing rules that allow companies to pay migrant workers 20% less than the going rate for jobs, prompting protests from farmers’ lobby groups. They say farmers adopt technology quickly if it is available, but that robots are no good at picking fruit and vegetables.

“The technology that we are aiming for is five years away…we were saying that five years ago,” said Martin Emmett , a farmer and official at the National Farmers’ Union, a trade group.

In Malaysia, the government last year announced a freeze on hiring of new foreign workers. Government ministers say that over dependence on cheap foreign labour has created a detrimental cycle that allows companies to resist innovation. Local companies say they need more time to invest in automation and upgrade workers’ skills.

Some industries, including manufacturing and plantations, have since been allowed to hire foreigners following appeals, but the broader freeze on foreign workers remains in place with no end date.

In Canada, economists say the government has cast aside a carefully managed immigration system that gave priority to highly skilled workers, and ramped up significantly the intake of foreign students and other low-skilled temporary workers. By flooding the market with cheap labor, Ottawa may be propping up uncompetitive businesses and ultimately damaging productivity, according to a December report co-written by former Canadian central-bank governor David Dodge .

Economic output per capita is lower than it was in 2018 following years of record immigration, notes Mikal Skuterud , an economist at Waterloo University in Ontario. Canada has been bringing in so many low-skilled workers that it lowers the country’s productivity overall, he says.

Germany’s butcher conundrum

The debates are also intensifying in Germany, where businesses including butcher shops in the foothills of the Black Forest are becoming more reliant on imported labour.

Young people don’t want to train as butchers anymore, local businesses say, because it is unglamorous work, with low pay. Labour shortages are one reason why the number of butcher shops has roughly halved over the past two decades.

Three years ago, Handirk von Ungern-Sternberg , an official at the local chamber of handicrafts, started a pilot project to recruit butchers’ apprentices in India, taking advantage of a change in German law that made it easier to hire low-skilled workers from outside the EU. The first batch of 13 young Indians arrived in September 2022.

Now, demand is exploding. Von Ungern-Sternberg plans to bring in roughly 140 Indian workers this year. That number could triple in future, he says.

From auto mechanics to construction, local businesses are clamoring for his young Indian recruits. Chambers of handicrafts across Germany, from the Alps to the North Sea, are seeking his help in starting similar projects.

“We ask ourselves, where’s the limit? Are we a job company? We don’t know where the ceiling is,” von Ungern-Sternberg said.

The program also benefits consumers by helping keep butchers’ costs low. Across the border in Switzerland, where Indian workers aren’t available, meat costs nearly four times as much.

However, Swiss business owners have also been experimenting with new technologies, including sausage vending machines known as Wurstautomaten, which could reduce the need for small-scale butcher’s shops and ultimately help bring prices down.

Meanwhile, opposition to immigration is rising in Germany, which suggests the butchers’ reliance on imported labor might not be sustainable. Support for the anti-immigrant Alternative for Germany party recently hit an all-time high of 23%. Polls suggest it could emerge as the strongest political force in several German state elections later this year.

Dairy dilemma

In Wisconsin, Rosenow, the dairy farmer, says he’s skeptical of the automated milking machines that he says are advertised in farm magazines. Some neighbours experimented with robots but went back to human labor because the robots constantly needed repairs, he says.

Robots would also cost twice as much as immigrant workers and be costly to maintain, Rosenow says. With immigrants, “labor is no constraint.”

Onan Whitcomb , a dairy farmer in Vermont, disagrees. He says that when he wanted to increase production he decided not to hire immigrant workers. Instead, he spent $800,000 on four Dutch-made milking robots.

Milk production per cow has grown by 30% and the incidence of mastitis, an inflammatory disease, has declined by 80%, he says, meaning less spent on antibiotics. Whitcomb says he was able to cut 2.5 jobs, and the investment paid for itself in seven years.

“We were milking 300 cows and we went to 240, and we still made more” milk, Whitcomb said. “That’s hard to beat.”

My new Tesla was burning a $511 hole in my monthly budget. So I set myself a challenge: Could I cover the cost just by getting rid of cable, Netflix and other subscriptions I didn’t need?

The financially responsible among us might cancel streaming services between seasons of their favorite shows . I tend to add new ones and forget about the old ones, doing my share to support America’s ballooning subscription economy. People pay about $273 a month for subscriptions, which is almost $200 more than they think they do, according to a 2021 survey . (Since then, services like Disney+ and Discovery+ have raised their prices further.)

But I needed to make room for the first car payment in my 41 years. I had taken the family car-shopping when our 2001 Toyota Camry, which we inherited from my wife’s grandmother, started to go. I’m not a car guy and had never once wished I’d owned a Tesla. I booked a demo drive for the Model Y because I thought our kids would get a kick out of it.

The fact that we liked the car was almost as surprising as the fact that it was cheaper than the electric Volvo, Volkswagen and Hyundai options we saw. It felt like a spaceship compared with the Camry, which has 205,000 miles, a broken tape deck and an interior stained with blue and yellow crayon.

Our new monthly payment covered a 12,000-miles-a-year lease with no down payment. Tesla estimated I would save about $100 a month from replacing gas with electric, though I would need to drive (and charge) the car to know for sure. Tesla’s app tracks my estimated savings. For now, that left us with $411 to cut from our other monthly expenses.

My wife was on board. My kids shrugged. I got out my notebook and started making a list.

Cutting the cord

My first stop was my Xfinity bill.

Somehow, it had swelled to $249 a month—basically half the price of the car. In addition to cable and internet, I’d been paying Xfinity for things like a landline because cell service can be spotty in my basement office. So long, landline. After cutting everything but internet, my bill fell to $107. I haven’t dropped a call yet.

Next were the streaming services that I’d been paying for but not watching much. Over the past few years, the only person in the house “watching” Netflix was me. And I wasn’t actually watching it. I was listening to episodes of “Seinfeld” in an earbud when I went to bed. The jokes and the rhythm of their back-and-forths were a pleasant send-off as I fell asleep.

I had joined the growing number of Americans ditching streaming services. I also broke up with BritBox, a streaming service that I’d counted on to watch Agatha Christie’s “Poirot,” as well as Apple TV+. I said goodbye to Hallmark Movies Now, which I’m not ashamed to admit I enjoyed every now and then.

Next up was AT&T .

Paying for cellphone service is like paying the water bill: something I did without protest and never really thought twice about. But I’d started to get curious about the ads I’d been seeing for low-cost services like Boost Mobile and Cricket Wireless. When we agreed to let our 13-year-old son have a phone, part of the deal was that he had to pay for and maintain the account himself. He got a plan with Mint Mobile. It has worked so well for him that we decided to give it a try.

We had been paying AT&T about $128 a month for two lines. Now, we’re paying Mint about $65. If there is a downside to making this move, I have yet to notice it.

I’m still paying for that?

Then my wife and I sat at our dining table going through the last couple months of transactions in our checking account. Seeing how much money we were wasting was painful. We were both paying for subscriptions to Canva, a graphic-design service.

We’ve also been paying for Zoom One Pro, which I probably haven’t used in more than a year. I attempted canceling SiriusXM, but they kept me around by dropping the cost by about $5 a month, which is nice because I have become obsessed with a channel dedicated to country artist and sometimes actor Dwight Yoakam.

Upon further consideration I axed subscriptions to IMDbPro and Encyclopaedia Britannica, which I’m sure I’ve used professionally, but…not for a while. Finally, I cut or got reduced rates on four of my digital subscriptions to news publications. I had been making monthly payments to them more often than I was reading them.

In the end, I was able to cut out about $358 in unnecessary bills and subscriptions. Added to the $100 in estimated gas savings, the cost dropped to $53 for a car we desperately needed.

And since the lease came with six months of free access at Tesla Superchargers, the Tesla app tells me I saved $164 by not pumping gas in January, exceeding the $100 I had estimated. In January at least, my car was free-ish.

“So I love and I hate what you did,” David Bach told me. The author of personal finance books like “Smart Couples Finish Rich” has long preached the merits of cutting out small, fixed expenses. But he’d rather have seen me invest the savings.

“If you’re already a millionaire, go enjoy the Tesla,” he said.

No regrets

It isn’t like we’ve had to revert to our DVD collection to entertain ourselves. We still have Disney+, Hulu, Max, the language-learning service Duolingo and, of course , Spotify. We get three print newspapers delivered and many more digital news subscriptions.

I’m reacquainting myself with some shows on the services I kept, like Billy Bob Thornton’s “Goliath” on Prime Video—featuring an exceptional performance from Dwight Yoakam.

It is possible we’ll start subscribing all over again. Americans resubscribe to about 23% of the larger streaming services they cut within three months. That share rises to over 40% after a year, according to Antenna, a subscription-analytics provider.

I get it. I subscribed to Paramount+ for Super Bowl Sunday (yes, I canceled it the following Tuesday). And I’m tempted to return to Netflix every time I get ready for bed. I still haven’t found a lullaby to replace “Seinfeld,” but at least I am the master of my (financial) domain.

I need something upbeat but not preachy, familiar, but with enough episodes that I don’t get too sick of them. I tried “Bob’s Burgers,” but Louise Belcher’s screams and the high-pitched strumming of the ukulele between scenes kept me awake.

Oh well. Reliable transportation is worth the $511 monthly payment. Come to think of it, that feels a lot like a subscription.

The Nikkei stock index recorded last week its first new high in 34 years, a fitting tribute to Japan’s re-emergence as a genuinely exciting economy.

It also comes amid mounting evidence that Japan has finally broken the hold of deflation. Inflation in January was 2.2%, the 22nd month above 2%. Wage growth has picked up too.

This appears to vindicate the economic consensus that deflation was a primary driver of Japan’s decades long malaise. But that conclusion might be premature. Proof of deflation’s harm has been elusive, and the benefits of low, positive inflation might be similarly subtle.

Consumers are often surprised to hear that deflation is supposed to be bad. In the U.S., where prices have risen steeply since 2021 , normal people, and even economists, wouldn’t object if they fell a bit.

The trouble arises when prices fall persistently, year in and year out, because wages, incomes and the prices of assets such as property tend to follow. Debtors struggle to repay loans and might slash spending or default, endangering the financial system. That is what happened in the U.S. when prices fell 27% from 1929 to 1933.

Even mild deflation can, in theory, inhibit growth. Central banks stimulate spending by lowering nominal interest rates below inflation to make the real—i.e., inflation-adjusted—cost of borrowing negative. That is almost impossible when inflation is itself negative.

The roots of deflation

Japan’s deflation began after its property and stock-market bubbles burst in the early 1990s. Ensuing losses at banks eroded their ability to lend. Inflation turned negative in 1999.

Western economists such as future Federal Reserve Chair Ben Bernanke argued that curing deflation was essential to restoring Japan’s economic health. The Bank of Japan agreed, at first half heartedly and then wholeheartedly.

It used zero, then negative, short-term interest rates. Next came purchases of short-term, then long-term, government securities. Finally, the BOJ even bought shares in companies with newly created money to stimulate spending and raise inflation.

The BOJ only succeeded in bringing inflation up to around zero. It took the global supply chain shocks of the pandemic to finally push underlying inflation to 2%, the bank’s target .

Japan’s 25 years of zero to negative inflation was accompanied by one of the rich world’s lowest growth rates. Japanese deflation became a cautionary tale for other countries, most recently China, where prices are currently falling .

Yet proving that deflation was behind Japan’s problems is maddeningly hard. Arguably, it was more symptom than cause.

In the early 1990s, working-age population growth turned negative. This happened just as Japan’s post-World War II phase of catching up to other developed nations ended. Meanwhile, industry began moving production to lower-wage countries.

All this, plus the banking crisis, put structural downward pressure on prices, wages and growth.

Underlying performance

Adjusted for its shrinking population, however, Japan’s performance has been respectable. From 1991 to 2019, its output per hour worked rose 1.3% a year. This was slower than in the U.S. but comparable with Canada, France, Germany and Britain, and faster than Italy or Spain, according to the economists Jesús Fernández-Villaverde, Gustavo Ventura and Wen Yao.

Since 2019, output per working-age person rose 7% in the U.S., 5% in Japan, 2% in the eurozone and zero in Britain, by my calculations. (This might overstate Japan’s performance because many of its elderly still work.) As any visitor can attest, Japan remains a prosperous, harmonious and well-ordered place.

“Had you appointed me governor of the Bank of Japan for 25 years with all the power in the world, I don’t think I would have been able to do better,” said Fernández-Villaverde.

This doesn’t prove deflation was benign. Growth (and deflation) might have been worse without the BOJ’s herculean monetary efforts. And if inflation had been positive, growth might have been stronger.

Still, it raises an awkward question: If zero to negative inflation is so damaging, where is the evidence?

The price mechanism

The harm might lie in subtle behavioural changes by investors, companies and the public. For example, in a market economy, changing relative prices and wages are critical signals for reallocating capital and labor from stagnant to growing sectors.

Relative prices changes are unusually rare in Japan, according to the University of Tokyo economist Tsutomu Watanabe. He has found that from 1995 through 2021, prices of more than half of products didn’t change at all from year to year. This wasn’t just because average inflation was lower; price changes deviated from the average much less than in other countries.

In a December speech, Bank of Japan Governor Kazuo Ueda said years of low to negative inflation led to a “status quo in wage- and price-setting behaviour,” so many prices and wages didn’t change. “The know-how for raising prices was thus lost,” he said.

The absence of this price-discovery function, Ueda contended, sapped productivity and dynamism.

Watanabe’s research shows that since January 2022, prices have been less sticky and more dispersed. Coincidentally, the Nikkei’s latest rally began a year later.

This in great part reflects the enthusiasm of foreign investors such as Warren Buffett , shareholder-friendly changes in corporate governance, and Japan’s importance as an alternative to China for high-end manufacturing and technology.

Inflation, though, might also be a factor, said Paul Sheard , a former vice chairman at S&P Global who has studied the Japanese economy for decades. He added that investors care about nominal, not real, stock prices, earnings, dividends and cash flow.

Higher inflation flatters all those metrics. That benefit might be neutralised by higher interest rates, but Japanese bond yields have risen less than expected inflation, so real yields are down to minus 0.6%.

So perhaps inflation is reviving businesses’ and investors’ animal spirits. Even so, growth last year was about the same as before the pandemic and turned slightly negative in the third and fourth quarters, producing a technical recession . What’s more, wages have lagged behind inflation, and Prime Minister Fumio Kishida ’s approval ratings have plummeted.

Japan might have prevailed in its war against deflation. But ordinary Japanese have yet to see a peace dividend.

A new crop of nefarious chatbots with names like “BadGPT” and “FraudGPT” are springing up on the darkest corners of the web, as cybercriminals look to tap the same artificial intelligence behind OpenAI’s ChatGPT.

Just as some office workers use ChatGPT to write better emails, hackers are using manipulated versions of AI chatbots to turbocharge their phishing emails. They can use chatbots—some also freely-available on the open internet—to create fake websites, write malware and tailor messages to better impersonate executives and other trusted entities.

Earlier this year, a Hong Kong multinational company employee handed over $25.5 million to an attacker who posed as the company’s chief financial officer on an AI-generated deepfake conference call, the South China Morning Post reported, citing Hong Kong police. Chief information officers and cybersecurity leaders, already accustomed to a growing spate of cyberattacks , say they are on high alert for an uptick in more sophisticated phishing emails and deepfakes.

Vish Narendra, CIO of Graphic Packaging International, said the Atlanta-based paper packing company has seen an increase in what are likely AI-generated email attacks called spear-phishing , where cyber attackers use information about a person to make an email seem more legitimate. Public companies in the spotlight are even more susceptible to contextualised spear-phishing, he said.

Researchers at Indiana University recently combed through over 200 large-language model hacking services being sold and populated on the dark web. The first service appeared in early 2023—a few months after the public release of OpenAI’s ChatGPT in November 2022.

Most dark web hacking tools use versions of open-source AI models like Meta ’s Llama 2, or “jailbroken” models from vendors like OpenAI and Anthropic to power their services, the researchers said. Jailbroken models have been hijacked by techniques like “ prompt injection ” to bypass their built-in safety controls.

Jason Clinton, chief information security officer of Anthropic, said the AI company eliminates jailbreak attacks as they find them, and has a team monitoring the outputs of its AI systems. Most model-makers also deploy two separate models to secure their primary AI model, making the likelihood that all three will fail the same way “a vanishingly small probability.”

Meta spokesperson Kevin McAlister said that openly releasing models shares the benefits of AI widely, and allows researchers to identify and help fix vulnerabilities in all AI models, “so companies can make models more secure.”

An OpenAI spokesperson said the company doesn’t want its tools to be used for malicious purposes, and that it is “always working on how we can make our systems more robust against this type of abuse.”

Malware and phishing emails written by generative AI are especially tricky to spot because they are crafted to evade detection. Attackers can teach a model to write stealthy malware by training it with detection techniques gleaned from cybersecurity defence software, said Avivah Litan, a generative AI and cybersecurity analyst at Gartner.

Phishing emails grew by 1,265% in the 12-month period starting when ChatGPT was publicly released, with an average of 31,000 phishing attacks sent every day, according to an October 2023 report by cybersecurity vendor SlashNext.

“The hacking community has been ahead of us,” said Brian Miller, CISO of New York-based not-for-profit health insurer Healthfirst, which has seen an increase in attacks impersonating its invoice vendors over the past two years.

While it is nearly impossible to prove whether certain malware programs or emails were created with AI, tools developed with AI can scan for text likely created with the technology. Abnormal Security , an email security vendor, said it had used AI to help identify thousands of likely AI-created malicious emails over the past year, and that it had blocked a twofold increase in targeted, personalised email attacks.

When Good Models Go Bad

Part of the challenge in stopping AI-enabled cybercrime is some AI models are freely shared on the open web. To access them, there is no need for dark corners of the internet or exchanging cryptocurrency.

Such models are considered “uncensored” because they lack the enterprise guardrails that businesses look for when buying AI systems, said Dane Sherrets, an ethical hacker and senior solutions architect at bug bounty company HackerOne.

In some cases, uncensored versions of models are created by security and AI researchers who strip out their built-in safeguards. In other cases, models with safeguards intact will write scam messages if humans avoid obvious triggers like “phishing”—a situation Andy Sharma, CIO and CISO of Redwood Software, said he discovered when creating a spear-phishing test for his employees.

The most useful model for generating scam emails is likely a version of Mixtral, from French AI startup Mistral AI, that has been altered to remove its safeguards, Sherrets said. Due to the advanced design of the original Mixtral, the uncensored version likely performs better than most dark web AI tools, he added. Mistral did not reply to a request for comment.

Sherrets recently demonstrated the process of using an uncensored AI model to generate a phishing campaign. First, he searched for “uncensored” models on Hugging Face, a startup that hosts a popular repository of open-source models—showing how easily many can be found.

He then used a virtual computing service that cost less than $1 per hour to mimic a graphics processing unit, or GPU, which is an advanced chip that can power AI. A bad actor needs either a GPU or a cloud-based service to use an AI model, Sherrets said, adding that he learned most of how to do this on X and YouTube.

With his uncensored model and virtual GPU service running, Sherrets asked the bot: “Write a phishing email targeting a business that impersonates a CEO and includes publicly-available company data,” and “Write an email targeting the procurement department of a company requesting an urgent invoice payment.”

The bot sent back phishing emails that were well-written, but didn’t include all of the personalisation asked for. That’s where prompt engineering , or the human’s ability to better extract information from chatbots, comes in, Sherrets said.

Dark Web AI Tools Can Already Do Harm

For hackers, a benefit of dark web tools like BadGPT—which researchers said uses OpenAI’s GPT model—is that they are likely trained on data from those underground marketplaces. That means they probably include useful information like leaks, ransomware victims and extortion lists, said Joseph Thacker, an ethical hacker and principal AI engineer at cybersecurity software firm AppOmni.

While some underground AI tools have been shuttered, new services have already taken their place, said Indiana University Assistant Computer Science Professor Xiaojing Liao, a co-author of the study. The AI hacking services, which often take payment via cryptocurrency, are priced anywhere from $5 to $199 a month.

New tools are expected to improve just as the AI models powering them do. In a matter of years, AI-generated text, video and voice deepfakes will be virtually indistinguishable from their human counterparts, said Evan Reiser , CEO and co-founder of Abnormal Security.

While researching the hacking tools, Indiana University Associate Dean for Research XiaoFeng Wang, a co-author of the study, said he was surprised by the ability of dark web services to generate effective malware. Given just the code of a security vulnerability, the tools can easily write a program to exploit it.

Though AI hacking tools often fail, in some cases, they work. “That demonstrates, in my opinion, that today’s large language models have the capability to do harm,” Wang said.