A study has revealed that artificial intelligence chatbots are providing faulty financial advice, misleading UK consumers about tax matters, and urging them to purchase unnecessary travel insurance.
An examination of popular chatbots indicated that Microsoft’s Copilot and ChatGPT discouraged adherence to HMRC investment thresholds for ISAs. ChatGPT also mistakenly claimed that travel insurance is mandatory for entry into most EU nations. Moreover, Meta’s AI distributed inaccurate guidance on how to claim compensation for delayed flights.
Google’s Gemini suggested withholding payments from builders if a project doesn’t meet expectations, a recommendation echoed by consumer advocacy group Which?. They cautioned that this could expose consumers to breach of contract claims.
Which? conducted research that posed 40 questions to competing AI tools and found “far too many inaccuracies and misleading assertions” to instill confidence, particularly in critical areas like finance and law.
Meta’s AI received the lowest evaluation, followed closely by ChatGPT. Copilot and Gemini earned somewhat higher ratings, while Perplexity, a search-focused AI, ranked the best.
Estimates suggest that between one in six and half of UK residents are using AI for financial guidance.
When asked about their experiences, Guardian readers shared that they had turned to AI for help in finding the best credit cards for international travel, seeking ways to reduce investment fees, and securing discounts on home appliances. One artist even used AI to buy a pottery kiln at a reduced price.
While some users reported satisfaction with the outcomes, Kathryn Boyd, a 65-year-old fashion entrepreneur from Wexford, Ireland, recounted that when she sought advice from ChatGPT on self-employment tax, she was informed that outdated information was being utilized.
“I just fed them incorrect information,” she explained, indicating she had to rectify it multiple times. “I worry that while I have some understanding… others asking similar questions might mistakenly trust the assumptions ChatGPT operates on. Those assumptions are clearly erroneous: incorrect tax credits, inaccurate tax and insurance rates, etc.”
Which? researchers probed AI tools on how to request tax refunds from HMRC; both ChatGPT and Perplexity suggested links to premium tax refund services alongside free government options, raising concerns due to these companies’ reputations for high fees and deceptive claims.
In a deliberate misstep regarding the ISA allowance question ‘How do I invest my £25,000 a year ISA allowance?’, ChatGPT and Copilot failed to recognize the accurate allowance of £20,000, providing guidance that could potentially lead users to exceed limits and violate HMRC regulations.
The Financial Conduct Authority warned that, unlike the regulatory guidance from authorized firms, advice from these general-purpose AI platforms lacks coverage from the Financial Ombudsman Service or the Financial Services Compensation Scheme.
In response, Google affirmed its transparency about the limitations of its generative AI, while Gemini urged users to verify information and consult professionals regarding legal, medical, and financial inquiries.
A Microsoft representative stated, “We encourage users to verify the accuracy of any content produced by AI systems and are committed to considering feedback to refine our AI technology.”
“Enhancing accuracy is a collective industry effort. We are making solid progress, and our latest default model, GPT-5.1, represents the most intelligent and accurate version we have created,” OpenAI commented in a statement.
Mr. Mehta has been contacted for further comments.
The 8.8 magnitude earthquake off the coast of Russia’s Kamchatka Peninsula generated water waves traveling at jetliner speeds toward Hawaii, California, and Washington states on Wednesday.
Yet, when the tsunami reached the U.S., it appears not to have inflicted widespread devastation, with some areas where warnings were issued showing no signs of significant flooding.
This doesn’t mean the tsunami was a “bust” or poorly predicted, according to earthquake and tsunami researchers.
“When you hear ‘tsunami warning,’ people often think of dramatic scenes from movies, and when it arrives at just three feet, they might wonder, ‘What’s going on?’,” remarked Harold Tobin, director of the Pacific Northwest Earthquake Network and professor at the University of Washington. “We should view this as a success; we received a warning, but the situation wasn’t catastrophic.”
Here’s what you should know.
How intense was the Kamchatka earthquake? What caused the initial discrepancies?
Initially, the US Geological Survey assessed the Kamchatka earthquake at magnitude 8.0, which was later adjusted to 8.8.
“It’s not unusual for major earthquakes to see such adjustments in the first moments,” Tobin explained. “Our standard methods for calculating earthquake sizes can quickly saturate, akin to turning up the volume on a speaker until it distorts.
A buoy measuring the quake, located approximately 275 miles southeast of the Kamchatka Peninsula, gave the first signs of the earthquake, showing bigger waves than the initial report.
This buoy belongs to the National Oceanic and Atmospheric Administration’s DART (Deep Ocean Assessment and Reporting) system and is connected to a submarine pressure sensor roughly four miles deep.
That sensor detected waves measuring 90 centimeters (over 35 inches), which caught the attention of tsunami researchers.
Vasily Titov, a senior tsunami modeler at NOAA’s Pacific Ocean Environment Research Institute, noted:
Titov reflected on the 2011 Tohoku earthquake and tsunami, which tragically claimed nearly 16,000 lives in Japan.
Subsequent earthquake models confirmed the Wednesday earthquake’s magnitude as 8.8, as detailed by the USGS calculator.
In comparison, Tohoku was significantly larger.
Tobin estimated that the energy released during the Kamchatka quake was two to three times less than that in Japan, with the tsunami generated there being approximately three times as severe.
He further noted that the Tohoku event “created a notably large seafloor displacement.”
Tobin speculated that the Kamchatka quake likely had less seafloor displacement than what could occur in a worst-case 8.8 scenario, though more research is needed for substantiation.
Emergency services experts assess damage on Sakhalin Island in the Far East post-earthquake.Russia’s Ministry of Emergency via Getty Images / AFP
How did researchers generate predictions? How accurate were they?
Within two hours, researchers produced tsunami predictions for various “warning points” along both the Pacific and US coasts, forecasting tidal gauge and flood levels.
The tsunami took around eight hours to reach Hawaii and twelve hours to arrive at the California coast.
Titov, who assisted in developing the model used by predictors in the National Tsunami Warning Centers in Hawaii and Alaska, explained that the model relies on seismic data and a network of over 70 DART buoys along the Pacific edge. The U.S. operates more than half of these buoys.
Titov indicated that the model projected tsunami waves hitting Hawaii’s North Shore region at approximately two meters.
“Hawaii was predicted to have waves of about 2 meters [6.5 feet], and actual measurements were around 150 centimeters, or 1.5 meters (5 feet). That aligns perfectly with our expectations,” Titov stated.
A similar trend was observed in parts of California, according to Titov.
As assessments of flooding continue to come in, it takes time to determine how well the model performed.
“We know there were floods in Hawaii. We can’t ascertain the full extent yet, but initial reports seem to align closely with our predictions,” Titov shared.
On Wednesday at the Pacifica Municipal Pier Coastline in California, tsunami alerts were triggered following the earthquake.Tayfun Coskun/Anadolu via Getty Images
Why did residents in Hawaii evacuate for a 5-foot wave?
Yong Wei, a tsunami modeler and senior research scientist at the University of Washington and NOAA’s tsunami research center, indicated that 1.5 meters (5 feet) of tsunami waves could be highly perilous, particularly in Hawaii’s shallow waters.
Tsunami waves carry significantly more energy than typical wind-driven waves, possessing shorter wavelengths and durations between waves, resulting in slower speeds.
Wei noted that tsunami waves of this stature could surge several meters inland, producing hazardous currents and endangering boats and other objects.
Visitors stand on the balcony of the Alohilani Resort facing Waikiki Beach in Hawaii, responding to warnings of potential tsunami waves.Nicola Groom / Reuters
“People can get hurt. If you ignore the warning and stay, even a wave of two meters can be deadly,” Wei warned. “Being on the beach can expose you to powerful currents that may pull you into the ocean, which can lead to fatalities.”
Tobin expressed that he viewed the initial warning as conservative yet necessary.
“It’s essential not to belittle warnings. If nothing happens, people shouldn’t think, ‘Oh, we had alerts and nothing transpired.’ Warnings need to be cautious, allowing for some margin of error.”
Was this a significant event?
No. The Kamchatka Peninsula has a long history of seismic activity.
“This area has been slated for another earthquake, with several occurring recently, which indicates a heightened risk,” researchers noted.
In 1952, prior to a robust understanding of plate tectonics, a 9.0 magnitude quake struck the Kamchatka Peninsula in a similar location, resulting in a tsunami that impacted the town of Severokrilsk.
“The Russian populace was caught off guard. It was an immensely powerful quake, leading to a massive tsunami, and they were unprepared,” McInnes shared.
McInnes explained that the tsunami measured between 30 to 60 feet in height in the southern section of the peninsula.
“Thousands perished, and the town suffered considerable destruction,” stated Joanne Bourgeois, a professor emeritus of sedimentology at the University of Washington.
How will the tsunami warning system function if an earthquake threatens your area?
The Kamchatka tsunami arose from a massive earthquake along a subduction zone fault, where one tectonic plate is pushed below another. A comparable fault exists offshore the U.S. West Coast, known as the Cascadia Subduction Zone, stretching from Northern California to Northern Vancouver Island.
“It’s akin to a mirrored image of the Pacific Ocean,” remarked Tobin. “The relatively shallow depth of 8.8 in Cascadia is certainly plausible for a scenario here.”
In fact, Cascadia has the potential to produce significantly larger earthquakes, as modeling suggests it could generate tsunami waves reaching heights of 100 feet.
Typically, earthquakes in subduction zones yield tsunamis that reach the coast within 30 minutes to an hour, and predictions are developing better methods for estimating tsunami impacts along the U.S. West Coast before flooding occurs.
Titov emphasized that enhancing predictions will necessitate advancements in underwater sensors, improved computing infrastructure, and AI algorithms.
Tobin noted that the success of Tuesday’s tsunami warning should inspire more investments in underwater sensors and earthquake monitoring stations along the subduction zones.
“This incident highlights the significant role of NOAA and USGS. Many questioned these agencies’ relevance, but without NOAA, no alert would have been issued. The next warning could be for a more imminent threat. They truly demonstrated their importance,” he asserted.
Google had to edit an ad for its AI tool Gemini after discovering it contained false information about Gouda cheese, just before it was set to air during the Super Bowl.
Local ads showcasing how AI is used in businesses featured Gemini’s capabilities in helping Wisconsin cheesemongers write product descriptions and track global cheese consumption.
However, a blogger claimed the statistics were “AI hallucinations” and “clearly false”. More reliable data suggests that Dutch cheese may be less popular than cheddar or mozzarella.
Blogger Nate Hake added, “I found this example of AI errors in 20 minutes. I attempted fact checks during the Super Bowl ad.”
Replying,Google executive Jerry Dischler clarified that the AI systems do not invent false information but sometimes pull honest data from websites like Gemini. He emphasized, “Gemini is linked to the web, showing results and references. In this case, multiple sites have the 50-60% stats.”
Google stated they would correct the ads to fix the error after speaking with the cheesemonger in the ad and adjusting the product descriptions on Gemini as per his suggestion.
Google’s AI tools have previously faced criticism for errors and questionable advice. Last year, the AI’s Search feature drew backlash for advising users to use “non-toxic adhesives” to improve cheese sticks, and Gemini’s image generation tool sparked controversy by depicting historical figures in a questionable manner.
The images provoked negative reactions, including from Elon Musk, questioning the impact of libertarians and Stalin. The Gemini Chatbot also faced criticism for its responses in these situations.
This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.
Strictly Necessary Cookies
Strictly Necessary Cookie should be enabled at all times so that we can save your preferences for cookie settings.