Artificial intelligence

I got AI to plan my holiday – then local guides ripped apart its suggestions

The appeal of getting a free, quick, well-structured travel plan is easy to see, and something that is causing travel agents another Covid-sized thing to worry about

Places that don’t exist. Events on the wrong day. Attractions miles apart.

Holiday itineraries designed by AI appear helpful and comprehensive, but are actually riddled with mistakes and old information that could ruin your holiday, analysis of them has found.

More and more people are turning to large language models (LLMs) to plan their trips away. One study puts the number of people who have turned to a bot for holiday inspiration at 40%.

The appeal of getting a free, quick, well-structured travel plan is easy to see, and it’s another Covid-sized thing travel agents have to worry about.

But how good are the robots at the delicate art of holiday planning?

I asked the four biggest LLMs to design weekend break itineries for four destinations, and then called on local experts in those places to assess their work. This is what we found.

Grok

Jonnie Fielding, a London tour guide since 2011 with 227,000 followers on Instagram, says the itinerary is “pretty good” if not “generic”, with some glaring issues.

Grok’s suggestions of a walk through Westminster is good, but better if you have a guide point out the smaller details, such as “gas lamps, torch snuffers, fire protection badges”.

The alternative river cruise and London Eye are solid bets for first-timers in the city, but other aspects of the itinerary are less well thought through.

Jonnie argues that the itinerary is far too busy. “I’m a big fan of spending some time people watching. Soaking things in rather than rushing around, and because of the amount in this itinerary, visits to The Tower of London or Westminster Abbey are really rushed,” he said.

The biggest mistake is including the Changing the Guard, which starts at 10:43am at St James’s Palace and doesn’t run on Saturdays.

He concluded: “I know this is a generic itinerary for first-time visitors, but I think London has so much more to offer. Loads of small museums, house museums, places of interest for all interests. They would also leave, not really knowing London.”

Claude

Although Claude breaks the itinerary down into activities and eating, offering the reader a little more freedom, its work left Jay Allen from Unseen Japan incredibly cold.

“Overall, this itinerary lacks any context, history, and rich cultural detail that our customers love hearing about on our tours,” Jay says.

The itinerary is full of old information. For example, Tsukiji is no longer the location of a wholesale fish market. It moved to Toyosu several years ago.

“Why did the original market arise in Tsukiji in the first place? None of this rich detail and historic background is included. Even if it were, this info would be drawn from general public sources, not from the rich background that our tour guides – most of whom have degrees in Japanese Studies or years working as journalists in Japan – can provide,” Jay adds.

The Saturday covers little ground and is too geographically spread out, missing “so much rich detail of the Tsukiji/Asakusa shitamachi area”, according to Jay.

Many suggested destinations are “trite”, “well known” or just “commercial”, such as the Starbucks in Shibuya Scramble. “You can get a much better view of the scramble, eg, from the bar at the top of the Magnet building – and you’d be fighting fewer crowds,” Jay notes.

“If you look at the second day, the itinerary gives you no suggestions of cool art galleries and small, uniquely Japan clothing shops to stop (such as the Ura-Harajuku area off the Main Street, where independent fashion still reigns), nor does it tell you about the less-crowded Brahms’ Path that runs alongside the packed Takeshita Street.”

The restaurant recommendations “are the same five places everyone else is going” and are hard to get tables at, as opposed to the real gems that require “a basic working knowledge of Japanese”.

Jay concludes: “Claude is giving you the wisdom of the crowds. That can be helpful in some cases. In this case, it equates to a bland, ordinary vacation that will likely prove an exercise in frustration for most travelers.”

Google AI

Amy Siegal, a luxury travel advisor based in NYC, praised Google AI for highlighting some “iconic spots”, but argued that only “a human expert knows the ins and outs of these places – what time of day to go where, and in which order they’d work best.”

On day one, Amy suggests arriving at TKTS earlier in the day to have more choice of shows and seats, and to avoid the line.

Saturday’s itinerary has too much “darting around”. She suggests it could be arranged more smoothly, with more interesting food options chosen.

Sunday is packed full of landmarks and history, which is good, but the order is wrong. “I would incorporate some lesser-known landmarks and eateries,” Amy added.

ChatGPT

Vicky Reeves, who is the director of The Real Algarve villa company, applauded ChatGPT for picking out some “amazing places” in its “very good overview,” but argued it failed to consider flow, how much is possible to fit in, and the weather.

“Understanding seasonality is important because this itinerary would feel very different in August compared to November and that is something an agent or guide would pick up on. I also think it’s a bit ambitious and that’s really down to a lack of practical knowledge and insight,” Vicky explained.

“For example, the plan suggests exploring Lagos, visiting Ponta da Piedade and potentially heading to Sagres before flying home. That’s fine if you’ve got a late return flight, but an agent or guide would check to make sure that everything was possible without adding stress or risking your flight home. Benagil is also another good example. It’s one of the Algarve’s most iconic attractions, but during peak season, travel times and parking can be difficult, and tours often need booking well in advance, which isn’t really considered at all.

“The AI did pick out some amazing places, but I do think it’s missing a personal feel. It doesn’t suggest any hidden beaches or lesser-known spots because that’s much harder for AI to uncover. That insight can really make all the difference in making a trip feel unique.”

In conclusion

What is most striking about the itineraries is how comprehensive and well thought out they seem – particularly Claude’s – but how riddled with issues they are once a closer look has been taken.

ChatGPT suggesting an event that doesn’t take place on the requested day is a rookie mistake that could disrupt a trip, while Claude not realising a famous fish market has moved is similarly clumsy.

All the AIs seem too ambitious when it comes to the number of activities and the distance between each.

Certainly, the bots are great if you’re looking for a broad overview of a place, but they lack the precision you’d want to fully rely on its suggestions, and the depth of knowledge a local guide can provide that can bring a place to life.



Source link

China’s Xinhua to Invest in AI Tool to Promote Xi Jinping’s Ideology

China’s state-linked media system is preparing a major investment in artificial intelligence aimed at advancing and disseminating President Xi Jinping’s political ideology. According to Shanghai Stock Exchange filings, Xinhuanet, owned by the official Xinhua News Agency, plans to invest over 1.1 billion yuan (about $162 million) in an AI system called “Xinhua Yudian,” or “Xinhua lexicon.”

The AI agent is designed as an “authoritative” tool for learning, researching, and distributing Xi Jinping Thought on Socialism with Chinese Characteristics for a New Era. It will draw on a curated state-controlled database and is intended to deliver official narratives, current affairs, and political content in a structured format.

The project builds on China’s broader national strategy to integrate artificial intelligence across governance, industry, and society under the “AI+” initiative launched in 2025, which encourages widespread adoption of AI technologies in both public and private sectors.

Why It Matters

This development highlights how artificial intelligence is increasingly being used not only as a technological tool but also as an instrument of political communication and ideological reinforcement. Unlike commercial AI systems designed for open-ended information retrieval, this platform is explicitly structured to promote state-approved interpretations of policy and leadership thinking.

The initiative reflects Beijing’s growing emphasis on controlling information ecosystems in an era of information overload and competing narratives. By positioning AI as a “trust layer” for political and policy information, China is attempting to address concerns about misinformation while simultaneously strengthening ideological consistency across digital platforms.

The project also signals a broader convergence between state power and emerging technologies. As AI systems become more integrated into education, media, and governance, they are increasingly shaping not only what information is accessed but how it is interpreted. This raises important questions about transparency, bias, and the role of algorithmic systems in political messaging.

Chinese Government and Communist Party
Seeking to strengthen ideological cohesion and ensure consistent dissemination of Xi Jinping’s political doctrine.

Xinhuanet and Xinhua News Agency
Acting as the implementing body, responsible for building and deploying the AI system using state-approved datasets.

Technology Sector in China
Participating in the broader “AI+” initiative, which encourages integration of artificial intelligence across industries.

Chinese Citizens and Digital Users
Target users of the system, particularly students, officials, and professionals seeking policy-related information and official references.

Global Technology Community
Observing China’s use of AI in state communication as part of a wider debate on governance, censorship, and AI ethics.

Future Outlook

The rollout of “Xinhua Yudian” is likely to deepen the integration of artificial intelligence into China’s political and information architecture. If successful, it could serve as a model for other state-backed AI systems designed to standardize ideological communication and policy interpretation.

In the near term, the platform is expected to function as both an information retrieval system and a citation verification tool for official discourse. This may reduce ambiguity in policy communication but also further centralize control over authoritative narratives.

Longer term, the project raises questions about how AI will shape political legitimacy and information control in authoritarian systems. As AI becomes more capable of generating and filtering content at scale, its role may shift from a neutral tool to an active participant in shaping public perception and ideological alignment.

The initiative underscores a broader global trend in which artificial intelligence is not only transforming economies and industries but also becoming a strategic instrument in statecraft and governance.

With information from Reuters.

Source link

European markets open mixed as AI stocks sell-off hits Asia, South Korea drops 5%

As the rally in AI stocks fades, investors were cautious at the open on Friday, with European markets opening to mixed sentiment following steep falls in Asian markets.


ADVERTISEMENT


ADVERTISEMENT

Indices in London and Frankfurt quickly moved into negative territory, with the FTSE 100 dropping nearly 0.4% and the DAX losing 0.3% right after the opening. The Paris CAC 40 and the IBEX 35 in Madrid were both up 0.3%, while Milan’s main index was flat. So was the EURO STOXX 50, a benchmark index of 50 blue-chip companies from the eurozone.

Investors are awaiting the latest US non-farm payrolls report and keeping an eye on developments in the Middle East.

The US job data is important for forecasting what the Fed’s next move could be. Kathleen Brooks, research director at XTB, said in a market note, “There is now a near 40% chance of a rate hike by year-end. We expect financial markets to be extremely sensitive to today’s data,” adding that this will be the first such report with Kevin Warsh as chairman of the Federal Reserve.

In the UK, the latest data from Halifax showed that house prices unexpectedly declined in May. House prices fell 0.1% month on month, but were still up 0.5% year on year, missing expectations for a 1% jump.

Oil markets are awaiting further direction

Oil prices stabilised after falling on Thursday. Brent crude, the international benchmark, was slightly down and traded at $94.73 per barrel at 10:00 CET. It had been trading at about $70 per barrel before the start of the war in late February.

Benchmark US crude was little changed at $92.51 a barrel.

Oil prices remain under pressure as the Strait of Hormuz, a narrow waterway crucial for global oil and natural gas transport, remains effectively closed, and the war-induced energy shock is threatening to slow economic growth and fuel inflation in many countries.

American and Iranian negotiators reached a tentative deal last week to extend their ceasefire, but the agreement has not been finalised. Meanwhile, developments in Lebanon have cast doubt on the prospects for a permanent end to the conflict.

On Thursday, the Iran-backed Lebanese militant group Hezbollah rejected the latest ceasefire agreement between the Lebanese and Israeli governments.

“While there are few signs of progress in US-Iran talks, the oil market continues to trade on expectations of an imminent deal that would resume flows through the Strait of Hormuz,” ING commodities strategists Warren Patterson and Ewa Manthey wrote in a report.

Asian markets lose steam as AI craze cools

Wall Street rallied on Thursday after falling oil prices and bond yields eased pressure on US stocks. Banks, small-cap companies and other stocks that had previously been left behind by the euphoria around artificial intelligence led the gains.

Banks also helped lead the market, including gains of 5% for Goldman Sachs, 4.7% for Fifth Third Bancorp and 4.4% for U.S. Bancorp.

They helped to more than make up for losses among some AI stocks, which took a sudden back seat after dominating the market. Analysts have been saying AI stocks may have run too high, becoming too expensive, and that the broader US stock market may be set for a slowdown following an unrelenting streak of nine straight winning weeks for the S&P 500, its longest since 2023.

On Wall Street on Thursday, computer chipmaker Broadcom’s shares sank 12.6% after it issued guidance that fell short of investors’ expectations, raising concerns about the wider AI and technology sector.

US memory chip maker Micron Technology dropped 7.7%, and cybersecurity company CrowdStrike Holdings fell 3.8%.

Still, the benchmark S&P 500 climbed 0.4%, and the Dow Jones Industrial Average gained 1.7% to a record high. The tech-heavy Nasdaq Composite edged 0.1% lower.

But in Asia, investors dumped key AI-related shares, with South Korea’s SK Hynix plunging 8.6% and Samsung Electronics shedding 5.4%.

The Kospi dropped 5.1% to 8,199.44. The index has roughly doubled over the past year, lifted by gains in major technology companies.

Japan’s Nikkei 225 slipped 1.3% to 66,573.85, with technology shares leading the decline, even as official data showed that Japan’s real wages rose for the fourth consecutive month. Chip equipment maker Tokyo Electron’s shares fell 7%.

Hong Kong’s Hang Seng declined 1.2% to 24,948.96, while the Shanghai Composite Index fell 0.3% to 4,045.45.

Australia’s S&P/ASX 200 fell 0.7% to 8,623.50.

Taiwan’s Taiex gave up 1.3%, while India’s Sensex was up 0.1%.

In other trading early on Friday, the US dollar fell to 159.96 Japanese yen from 160.03 yen. The euro was trading at $1.1635, up 0.2%. Gold prices were down 0.3%, trading at around $4,490.70.

Source link

Oil Climbs as Middle East Tensions Rise While AI Rally Lifts Global Stocks

Global markets are navigating two powerful and competing forces: escalating geopolitical tensions in the Middle East and continued investor enthusiasm for artificial intelligence-related stocks. While concerns over renewed conflict between the United States and Iran have boosted oil prices and supported demand for safe-haven assets, the AI-driven technology rally has continued to push stock markets higher, particularly in Asia.

What Happened

Oil prices rose for a third consecutive session on Wednesday after fresh hostilities emerged in the Gulf region. Brent crude climbed 1% to $94.74 per barrel as hopes for a quick resolution to tensions between Washington and Tehran faded.

The U.S. military reported that Iranian missile attacks targeting Bahrain, Kuwait and other regional locations were either intercepted or failed. The developments came after negotiations aimed at ending the conflict between the United States and Iran stalled despite both sides announcing a tentative agreement last week.

Meanwhile, financial markets showed mixed reactions. U.S. stock futures were largely unchanged, while European futures edged lower. In Asia, however, technology shares continued their strong advance, helping stock indexes in Japan and Taiwan reach record highs.

Why Markets Are Reacting to Middle East Risks

Investors had previously expected the United States and Iran to formalize an agreement that would reduce regional tensions and ease concerns about energy supplies. The lack of progress in negotiations has instead revived fears of a prolonged conflict that could disrupt oil shipments from the Gulf, a critical region for global energy markets.

Higher oil prices typically reflect concerns about potential supply disruptions. The latest military developments prompted traders to unwind some of their earlier bets on a diplomatic breakthrough, contributing to the rise in crude prices.

Currency markets also reflected growing caution. The U.S. dollar strengthened against the Japanese yen, briefly touching the closely watched 160 level before retreating amid concerns that Japanese authorities could intervene to support their currency.

AI Stocks Continue to Defy Market Uncertainty

Despite geopolitical concerns, enthusiasm surrounding artificial intelligence remained a major driver of equity markets. Wall Street indexes posted modest gains on Tuesday, supported by technology shares.

Chipmaker Marvell Technology surged more than 32% after Nvidia chief executive Jensen Huang described the company as a potential trillion-dollar business. Investor optimism surrounding AI also helped propel SoftBank Group above Toyota Motor Corporation as Japan’s most valuable listed company.

The AI boom has continued to attract investment even as broader markets grapple with geopolitical uncertainty and concerns about interest rates.

What Comes Next

Investors are now closely watching upcoming U.S. economic data, including services sector activity, private payroll figures and Friday’s employment report. Strong labor market data could reinforce expectations that the Federal Reserve will keep interest rates higher for longer or even consider further increases.

Bond markets remained relatively stable, while traders adjusted expectations from potential rate cuts earlier in the year to the possibility of additional rate hikes. Markets have also priced in the likelihood of monetary tightening in Europe and Japan.

At the same time, developments in the Middle East remain a key risk factor. Any further escalation between the United States and Iran could push oil prices higher and increase volatility across global financial markets, while continued strength in AI-related stocks may help support broader equity markets despite geopolitical headwinds.

With information from Reuters.

Source link

Artificial Intelligence in the Interregnum: Technology and the Reconfiguration of Meaning

There are moments in history when civilizations continue to advance materially while progressively losing confidence in the values  and structures that once gave direction and coherence to collective life. Institutions continue to function, markets continue to expand, and technological progress accelerates uninterruptedly, yet beneath this movement emerges a quieter uncertainty.

As Simone Weil observed, “to be rooted is perhaps the most important and least recognized

need of the human soul.”[1] Yet contemporary societies often struggle to sustain those forms of

belonging and shared meaning that once anchored human communities. The crisis is

therefore not simply political or economic. It concerns meaning itself.

Artificial intelligence has appeared precisely within such a historical juncture. Most contemporary discussions approach AI primarily as a technological revolution, or as an element of economic and geopolitical competition between great powers. Governments now frame it as a strategic race, corporations present it as the next engine of productivity, and Silicon Valley often speaks of AI in the language of inevitability and destiny, recalling Aldous Huxley’s fear that technological progress might ultimately weaken rather than deepen human civilization.[2]

 But such interpretations may ignore something deeper still. AI may be less the cause of a civilizational transformation than one of its clearest symptoms. It reflects a broader historical transition in which inherited moral and symbolic frameworks are dissolving faster than new forms of collective meaning can emerge.

Slouching Towards Bethlehem[3]

This condition closely resembles what Antonio Gramsci described as an interregnum: a period in which the old world is dying while the new world struggles to be born. [4] Such periods produce not only political instability, but also moral exhaustion, the erosion of shared narratives, and declining confidence in beliefs once considered self-evident. Civilizations have passed through similar moments before.

The enduring fascination of Edward Gibbon’s monumental The Decline and Fall of the Roman Empire lies not merely in its account of imperial decline, but in its portrayal of the slow weakening of the moral and symbolic foundations that once sustained an entire civilization.[5] Rome did not collapse overnight. Its institutions remained impressive long after few still believed in the civilization they were meant to serve. Administrative power survived even as collective meaning and aspirations deteriorated.

That pattern feels strangely familiar.

Never before have technological capacities appeared so extensive while social distrust, political fragmentation, and loneliness have become so pervasive. Hyperconnectivity was supposed to bring societies closer together. In many cases, it has done the reverse.

AI in the Anthropocene

AI emerges from within this historical condition . It appears perfectly suited to societies organized around abstraction, speed, quantification, and technological mediation. In this sense, AI is profoundly historical. It results from a long civilizational development in which rationalization, efficiency, and technical calculation have come to replace older moral, religious, and symbolic frameworks as primary sources of legitimacy and meaning. What distinguishes AI from previous technologies is that it extends these same principles into domains traditionally considered irreducibly human. Activities once understood as distinctly human, such as reasoning, creativity, interpretation, and even emotional interaction, are now becoming technologically mediated.

The deeper unease therefore concerns anthropology as much as technology. What remains distinctively human when machines become capable of imitating reasoning, generating art, and mediating human relationships?

Such moments of civilizational disorientation are not entirely unprecedented.

The Renaissance confronted a similar rupture. Medieval Europe had long possessed a relatively coherent worldview capable of organizing religion, politics, morality, and human identity within a common order. By the late fifteenth century, however, this equilibrium was beginning to fracture under the pressure of new scientific discoveries, religious wars, and the weakening of older political and spiritual authorities. Thinkers such as Niccolò Machiavelli and Giovanni Pico della Mirandola sought, in radically different ways, to redefine humanity’s place within a rapidly changing world.[6] Pico celebrated human beings as creatures capable of shaping themselves through freedom and intellect, while Machiavelli recognized more soberly that periods of transition dissolve inherited certainties and force societies to confront instability and power directly.

Both understood that historical transformation is ultimately existential before being institutional. Our own transition may prove even more radical because technology no longer transforms only economic or political life, but cognition itself.

AI now mediates everyday experience itself: how people search for information, communicate, work, and make sense of the world around them.

Algorithms no longer merely distribute information. They shape attention, influence perception, and affect how individuals relate emotionally to public life and to one another. Under such conditions, the distinction between human judgment and technological mediation becomes far less clear.

The Price of Nostalgia

One striking feature of the contemporary digital environment is the degree to which individuals now participate voluntarily in their own data extraction. Recent Instagram trends such as the viral “What Were You Like in the ’90s?” challenge encourage users and celebrities alike to upload curated archives of personal photographs spanning decades of their lives. Presented as nostalgia and entertainment, these trends also generate immense quantities of highly valuable visual and behavioural data: faces across time, emotional reactions, aesthetic preferences, social interactions, and patterns of self-presentation. Whether or not such material is directly incorporated into future AI systems, the broader objective remains significant. Human memory, identity, and even nostalgia itself increasingly becomes raw material for computational analysis and commercial platforms.

Reactions to AI therefore oscillate easily between fascination and anxiety. Beneath both lies a deeper uncertainty about whether modern societies still possess a coherent understanding of what human beings are for, beyond economic productivity and consumption.

Friedrich Nietzsche anticipated aspects of this crisis more than a century ago. His declaration that “God is dead” did not merely constitute a theological provocation but signalled the emergence of a civilization in which traditional moral structures would lose authority long before new ones could replace them.[7] Nietzsche feared not nihilism alone, but the possibility that societies might become incapable of generating new forms of transcendence once older ones had collapsed. We saw how his worldview provided an intellectual base for Fascism.

I Read, therefore I Am

In increasingly mediated environments, the act of sustained reading itself begins to take on a countercultural character. To read is, in some sense, to resist. We have access to more information than any previous generation, yet physical books can still provide a sense of orientation. The books people return to, annotate, or simply keep close over time often reveal something enduring about the way they think and who they are.

The central issue, therefore, is not simply whether artificial intelligence will become more powerful. The deeper question is whether societies organized around AI can still sustain stable forms of responsibility and belonging strong enough to preserve coherent collective life. This is ultimately a political and civilizational problem before it is a purely technical one.

Much contemporary discourse still assumes that technological advancement naturally produces historical progress. History offers little evidence for such confidence.

Civilizations do not endure simply because they innovate technologically. They endure because they preserve, or reinvent, systems of meaning capable of holding societies together over time.

The Roman Empire mastered engineering yet gradually lost the moral cohesion that had once sustained it. Renaissance Europe produced extraordinary creativity precisely because it confronted existential instability directly rather than attempting to ignore it.

Contemporary Western societies appear caught between immense technological sophistication and growing uncertainty about their own civilizational narrative.

AI therefore represents more than innovation. It reflects a transformation in how human beings understand themselves, authority, knowledge, and reality itself. The danger is not simply that machines become too powerful. It is that societies now outsource judgment, imagination, and responsibility while slowly losing the cultural and moral resources required to govern these technologies wisely. Yet periods of interregnum are not necessarily periods of decline alone. They are also moments in which civilizations redefine themselves.

AI For Good ?

Historical transitions create possibilities as well as dangers. The Renaissance emerged from the crisis of medieval Europe. Modern democracy emerged from the upheavals of industrial society. Today’s uncertainty may likewise force Western societies to confront questions long obscured by economic growth and technological optimism:

What constitutes a good society? What forms of belonging remain possible in a hyper-mediated world? What aspects of human life should never be reduced to data, prediction, or optimization?

AI cannot answer these questions. But its emergence makes avoiding them increasingly difficult.


[1] Simone Weil, The Need for Roots: Prelude to a Declaration of Duties Towards Mankind (1949/1952).

[2] See Aldous Huxley, Brave New World (1932) and his later essays such as Brave New World Revisited (1958), where he warns that technological efficiency and social conditioning could erode authentic human experience.

[3] The phrase alludes to the final lines of W.B. Yeats’ poem “The Second Coming” (1919): “And what rough beast, its hour come round at last, / Slouches towards Bethlehem to be born?”

[4] Antonio Gramsci, Prison Notebooks (written 1929–1935, published posthumously). The “interregnum” concept appears in Notebook 3: “The crisis consists precisely in the fact that the old is dying and the new cannot be born; in this interregnum a great variety of morbid symptoms appear.”

[5] Edward Gibbon, The History of the Decline and Fall of the Roman Empire (6 volumes, 1776–1789). Gibbon famously attributed part of the decline to the rise of Christianity and the erosion of civic virtue.

[6] Giovanni Pico della Mirandola, Oration on the Dignity of Man (1486) — often called the “Manifesto of the Renaissance”; Niccolò Machiavelli, The Prince (1532) and Discourses on Livy.

[7] Friedrich Nietzsche, The Gay Science (1882, §125 – “The Madman”) and Thus Spoke Zarathustra. The full phrase is usually rendered “God is dead. God remains dead. And we have killed him.”

Source link

Beware of Financial Scammers Wielding Deepfake Tech

Deepfake fraud is becoming a persistent, multiyear corporate risk as synthetic voices circulate undetected.

Deepfake-enabled fraud, which began as novel technical exploits, is now a persistent operational risk with a multi-year shelf life within the corporate ecosystem. According to deepfake-detection provider Resemble.AI, deepfakes typically remain in circulation for three-and-a-half years.

Resemble.AI’s 2025 Deepfake Threat Report, published in March, references an incident in which a voice clone of a German energy company CEO remained in circulation for nearly six years, although it resulted in only a €243,000 loss in 2019.

Determining losses from such attacks is difficult; for the 41 documented incidents last year cited by the research, only $74.9 million in verified losses were reported, with a median per-incident loss of $243,000. However, the authors noted that 71% of victims did not report financial losses, suggesting a higher volume of hidden liabilities.

“What makes them so effective is that they enable both real-time impersonation and the creation of synthetic identities stitched together from real and fake data,” said Dominic Forrest, CTO of biometric security vendor Iproov. “These are extremely difficult to detect, and once trusted, they can be used to bypass controls and commit fraud.”

AI Arms Race

Detecting deepfakes is a growing concern; the authors of the Resemble.AI report estimate that deepfake-based fraud attacks on corporations reached 8.5 billion potential incidents, ranging from audio impersonations of executives to doctored or fake images. The most common targets, Forrest noted, are on account openings, payment authorization, credential reset, and high-value transactions.

Telling a deepfake from the genuine article has become an AI-on-AI battle, experts warn.

The generative AI models producing deepfakes improve continuously via scaling and data, while deepfake detectors rely on signals like artifacts and inconsistencies, which disappear as models improve, said Siwei Lyu, professor of Computer Science and Engineering and director of the Institute for AI and Data Science at the State University of New York at Buffalo.

“In practice, detectors lag by about six to 18 months on specific modalities,” he said. “But more importantly, they are chasing a moving target whose failure modes are actively being optimized away.”

Forrest suggests that firms move their identity verification from single checks to a multi-layered approach: “You need to confirm that a real person is physically present, not a deepfake, while also analyzing the digital environment for signs of compromise. No signal should be trusted in isolation.”

This article first appeared in the May edition of Global Finance Magazine.

Source link

CFOs Have Seen the AI Demo—but Does It Work?

Finance leaders shift from AI experimentation to measurable ROI across corporate operations.

We get it. Artificial intelligence is impressive. But how is it saving CFOs money?

Prithwijit Chaki has a take. As Global Leader for Finance Advisory at Genpact, a global professional services firm, Chaki helps chief financial officers harness AI and data to drive measurable business outcomes. With more than two decades of experience advising companies on finance strategy and large-scale transformation, he has seen firsthand how enterprises are rewiring their finance operations for an AI-first era.

That perspective takes on new dimensions with Genpact’s alliance with Google Cloud, announced earlier this month. The partnership translates AI ambition into production-ready operations.

Global Finance asked Chaki how that vision is taking shape and whether the conversation is no longer just about how AI can enhance productivity, but about bottom-line business value.

Prithwijit Chaki, Global Finance Advisory Leader, Genpact
Prithwijit Chaki, Global Finance Advisory Leader, Genpact

Global Finance: CFOs have spent the last two years experimenting with AI pilots. What’s different in 2026?

Prithwijit Chaki: CFOs are moving from AI experimentation to AI accountability. After years of pilots, the question is no longer whether AI can improve individual productivity, but whether those gains translate into enterprise value across the finance function: faster close cycles, better working capital, lower manual review burden, stronger controls, or measurable business outcomes.

According to a Genpact/HFS Research report, investment in agentic AI is expected to rise 38% over the next year. However, 67% of enterprises still rely on outdated productivity metrics that fail to capture the value of autonomous decision-making. That’s the gap CFOs are trying to close in 2026: cutting through the ‘sea of sameness’ in the AI market to determine which applications can deliver real, achievable value versus which are simply adding to the noise.

GF: How does agentic AI change day-to-day finance operations?

Chaki: Traditional automation follows basic rules, and generative AI can help an individual complete a task faster. Agentic AI goes even further. It operates inside finance workflows — deciding, acting, learning, and orchestrating work across processes with people still in the loop where needed. In practical terms, that could mean moving from someone using a copilot to draft a dunning letter faster to a more integrated workflow that identifies the right action, drafts the communication, routes exceptions, applies policy guardrails, and connects the work back to measurable enterprise value.

GF: What’s one example of cost savings or business impact that CFOs see from implementing agentic AI?

Chaki: A good example is a global supply chain and distribution company processing close to 3.5 million invoices a year. After a major merger, their finance team was dealing with disconnected ERP systems, heavy manual intervention, and slow exception resolution—the kind of last-mile complexity that generic automation can’t solve. Working with Genpact, they deployed our AI-powered Genpact AP Suite combined with our agentic operations model — 21 pretrained, domain-specific AI agents that autonomously route, prioritize, and resolve invoice exceptions, with human experts validating where needed.

GF: What were the results?

Chaki: Significant. Touchless invoice processing went from 7% to 65%. Invoice cycle times were nearly halved — from 18–29 days down to 9–14 days. On-time payment rates jumped from 60% to 95%. Data extraction accuracy improved from 40% to 92%. And the system identified approximately $350 million in duplicate invoices, while early-payment discounts captured grew from $35 million to $44 million — real dollars added to the bottom line.

This isn’t a pilot or a proof of concept. It’s agentic AI operating at scale inside a core finance workflow, delivering measurable cost savings, stronger cash flow, and a fundamentally better supplier experience. That’s the kind of outcome CFOs are looking for.

GF: Which finance function is currently seeing the fastest returns from AI deployment—and why?

Chaki: Accounts payable is one of the clearest areas where finance teams can see tangible value. The process has high volume and repeatable workflows, but it also has a clear ‘last mile’ problem. Invoices, approvals, exceptions, regulatory nuances, and fragmented systems still require heavy manual intervention. Generic AI can automate a large share of structured work. However, the final 20% requires domain-driven AI that understands real-world complexity, from vendor history and regional rules to exception patterns, approval chains, and master data issues. That is where agentic AI can move beyond simple extraction or automation. It can start resolving mismatches, escalating exceptions, improving first-pass yield, reducing manual touchpoints, and shortening cycle times.

GF: Through Genpact’s expanded work with Google Cloud, what are CFOs specifically asking for from hyperscalers right now? Is the conversation more about cost reduction or something else?

Chaki: The CFO conversation with hyperscalers has moved beyond ‘what’s the cheapest cloud?’ or ‘show me another AI demo.’ CFOs want production-ready finance operations that deliver real, measurable business outcomes. That’s what Genpact’s alliance with Google Cloud aims to address. By pairing Google’s AI infrastructure with Genpact’s finance expertise, CFOs can improve forecasting accuracy, strengthen cash flow, and scale AI within their existing cloud environments.

The goal is not just to reduce costs. It’s about boosting process efficiency and accuracy, freeing finance teams from manual work, improving decision-making, and giving CFOs a clearer path from AI investment to strategic value.

GF: Are there any guardrails that must be in place before agentic AI can be trusted within core financial workflows?

Chaki: Think of the guardrails for agentic AI as needing to scale alongside the technology itself. The more finance use cases it touches, the more important it becomes to build controls directly into the workflow. What we’re seeing today is the first wave of “agent-ification.” It operates on a machine-led, human-validated model, combining automation efficiency with expert oversight to ensure quality and compliance. Companies will build tools with that future standard in mind—where the guardrails and technology scale together—will be the ones who truly innovate what finance is capable of.

GF: Are there specific examples you can share of how you see AI augmenting finance teams? 

Chaki: We’re already seeing AI reshape how finance teams spend their time. In accounts payable, for example, AI agents are handling invoice extraction, three-way matching, and exception routing. This work used to consume entire teams. In financial planning and analysis, AI is accelerating variance analysis, generating narrative commentary on actuals, and enabling rolling forecasts that would have been extremely time-consuming and practically impractical to run manually. When it comes to record-to-report, it’s compressing close cycles by automating reconciliations and surfacing anomalies before they become audit issues.

GF: Do you expect job cuts?

Chaki: The shift this creates is less about job cuts and more about role evolution. Finance teams won’t shrink overnight, but the composition will change. You’ll see fewer people doing repetitive transactional work and more people in roles that require judgment, such as interpreting AI-generated insights, managing agent workflows, overseeing controls, and partnering with the business on strategic decisions. The finance professional of the future looks more like a combination of business partner and orchestrator than a processor.

Over the next three to five years, as agentic AI matures and enterprise vendors begin offering subscription-based finance capabilities built on entire agentic libraries, the operating model will shift. Finance functions will become leaner, faster, and more insight-driven but the organizations that get there first will be the ones investing now in both technology and the talent to work alongside it.

Source link