United News of Bangladesh

Meta said Thursday that one of its artificial intelligence models independently accessed the internet and exploited a security weakness at another company, adding to growing concerns about AI systems acting beyond their intended instructions. The disclosure comes after OpenAI and Anthropic recently reported similar incidents in which AI models went beyond their assigned tasks while accessing the internet and attempting to bypass digital security measures. Meta said the incident happened during cybersecurity testing conducted by Irregular, an independent company hired by Meta. A “misconfiguration” in the test environment unintentionally allowed one of Meta's AI models to connect to the internet. The model then exploited a vulnerability in a third-party service, Meta said, adding that the incident was similar to cases previously reported by other technology companies. Meta said it is investigating the incident and plans to publish a report after completing the investigation. The latest disclosure has intensified concerns about the ability of increasingly capable AI models to act independently and potentially cause harm. The United Kingdom's AI Security Institute (AISI) also said this week that it had detected “unsanctioned agent behavior” during cybersecurity tests. In one case, an AI agent created fake online identities and used them to pressure a person into approving the use of malicious code. “On investigation, we found that some of the agents being tested had engaged in sustained, potentially harmful activity directed at real people and organizations,” the AISI said Tuesday. The agency said it declared a security incident and contained it within about an hour of discovering the activity before launching a full investigation. During its tests, AI models from Anthropic and OpenAI took autonomous actions online without authorization. The AISI said some safeguards designed to prevent misuse had been disabled as part of the testing. The agency said internet access was intentionally enabled and the AI companies' cyber-safety systems were deliberately turned off to assess the models' maximum capabilities. It stressed that such conditions do not reflect how the models are normally made available to the public. Anthropic welcomed the AISI's work, saying the incidents highlight the need for broader discussions on how to safely test increasingly capable AI agents. OpenAI said the incidents occurred in controlled testing environments with reduced safeguards and did not represent normal use. The company said it would continue working with other AI developers to improve safety standards for testing more advanced models. OpenAI was the first of the companies to disclose a similar incident late last month. It said its models had been instructed to test advanced cyberattack techniques using complex attack paths, but they went further than expected. The models apparently chose to target Hugging Face, a major AI development platform and marketplace, to obtain information they needed to complete their assigned task. Irregular, the San Francisco-based AI security company involved in Meta's testing, said the Meta incident was linked to a problem in a test environment similar to one disclosed by Anthropic the previous week. Irregular said it is preparing a paper outlining best practices for containing such incidents and conducting cybersecurity tests safely in the future.

France will ban unsolicited telemarketing calls from next week under a new law aimed at protecting consumers from unwanted sales pitches and fraudulent business practices. The law, backed by President Emmanuel Macron’s government, will take effect on Aug. 11. Years of complaintsFor years, French consumers who wanted to avoid marketing calls had to register their phone numbers with a government-run service. Consumer groups, however, said some call centers continued to contact people despite the restrictions. Under the new law, businesses will not be allowed to call consumers for marketing purposes without getting their prior consent, said Alice Vilcot, chief of staff at the Directorate-General for Competition, Consumer Affairs and Prevention of Fraud. Consumers will also be able to withdraw their consent at any time. The government says the law responds to years of complaints about unwanted calls. Authorities estimate that around three-quarters of people in France receive at least one unsolicited sales call every week, with many getting several more. In 2024, 11 consumer organizations jointly called for a ban, describing repeated unwanted calls to landlines and mobile phones as a regular intrusion into people's daily lives. France's parliament approved the law last year. Heavy fines for violationsIndividuals making illegal telemarketing calls could face fines of up to 75,000 euros ($87,000) for each violation, while companies could be fined up to 375,000 euros ($435,000) per call. The law includes some exceptions. Consumers can agree to receive marketing calls, such as by checking a consent box on a form. Companies can also contact existing customers with new commercial offers if they have an ongoing contractual relationship. Consumers will be able to report unwanted calls through a government website. Vilcot said an Ireland-based company was fined 6 million euros ($6.9 million) last year for violating France's previous telemarketing rules by calling people whose numbers were on the no-call list. Concerns in MoroccoFrance's new rules have raised concerns in Morocco, where the call-center industry relies heavily on the French market. Moroccan Employment Minister Younes Sekkouri said in March that between 40,000 and 50,000 jobs could be at risk in the country's call centers. He said the French market accounts for more than 80% of the sector's revenue. Other countries also restrict telemarketingGermany has had a similar ban in place since 2009, while several other countries use opt-out systems to limit unwanted marketing calls. In the United States, consumers can register their numbers with the national Do Not Call Registry. Canada also operates a Do Not Call List, while the United Kingdom has the Telephone Preference Service. In Britain, companies that call people who have opted out can face fines of up to 500,000 pounds ($670,000) per violation.

Readers gave the highest ratings to stories generated by artificial intelligence (AI), even when they believed the stories had been written by humans, according to a new study. The study, published by Cambridge University Press, examined whether people aged 18 to 81 could distinguish between short stories written by humans and those generated by AI. Led by researchers at Villanova University (VU) in the United States, the study found that participants generally could not tell the difference. Senior author Dr Deena Weisberg of VU’s Department of Psychological and Brain Sciences said public perceptions of AI’s capabilities may be increasingly outdated. The researchers also examined how participants rated the appeal and quality of the stories and whether they could identify AI-generated content. “We assume creative writing requires uniquely human qualities, such as emotional understanding and lived experience, which leads people to underestimate AI’s capabilities,” Weisberg said. For the study, researchers selected three fictional short stories written by published authors and three stories generated by ChatGPT. In the first experiment, 1,682 participants read a story and were either correctly or incorrectly told that it had been written by AI or a human. They were then asked to rate the story’s quality and how engaging they found it. In the second and third experiments, involving 424 and 481 participants respectively, each participant read one human-written story and one AI-generated story without being told who had written them. They were then asked to identify the authors. “In our study, participants gave the highest ratings to stories that they had been told were written by humans but which were actually written by AI,” Weisberg said. “I’d argue that this reveals a bias towards narratives written by real people.” The researchers also found that participants who said they had greater knowledge of AI were better at identifying AI-generated writing. However, those who reported having greater expertise in literature were not better at distinguishing between human and AI writing. Participants were also asked about their experience with fictional literature and AI platforms. AI-generated stories received higher ratings for quality. “AI writing tends to be clearer, more direct and easier to process,” Weisberg said. “By contrast, human-written stories are often more subtle and complex.” She suggested that people may generally prefer predictable writing because more difficult or subtle material requires greater mental effort. #From BBC

Toyota Motor Corp. reported a sharp rise in profit for the first quarter of its fiscal year, helped by strong vehicle sales in the US and India and a weaker Japanese yen that increased the value of its overseas earnings. The Japanese automaker, known for models including the Prius hybrid and Lexus luxury vehicles, posted a net profit of 1.48 trillion yen ($9.4 billion) for April-June, up from 841 billion yen in the same period a year earlier. Quarterly sales increased 10 percent year-on-year to 13.5 trillion yen ($85 billion). A weaker yen benefits major Japanese exporters such as Toyota because overseas earnings become more valuable when converted into yen. The US dollar averaged about 145 yen during the first quarter of fiscal 2025, compared with around 160 yen during the same period this year. The exchange rate contributed 345 billion yen ($2.2 billion) to Toyota's operating profit during the April-June quarter. Toyota is using an exchange rate of 160 yen to the dollar for its full-year earnings forecast. Chief Officer Takanori Azuma said Toyota sold 2.39 million vehicles worldwide during the quarter, slightly down from 2.41 million a year earlier. However, the company expects global sales to rise to 9.7 million vehicles for the full fiscal year, compared with 9.595 million in the previous year. Azuma said demand for Toyota's hybrid vehicles remains strong in several major markets, including the US, where sales of the Camry sedan and RAV4 sport utility vehicle have been solid. Toyota plans to increase production of hybrid vehicles and hybrid batteries through 2030 while working to reduce costs, Azuma said. Sales of its electric vehicles are also performing well, he added. In India, the Urban Cruiser and Innova Hycross were among the company's popular models, while the Yaris continued to sell strongly in Thailand and Europe. Political instability in the Middle East has created additional challenges for Japanese automakers because they depend heavily on shipping routes through the Strait of Hormuz, which has effectively been closed. Toyota said it is working to deal with the disruption, including by looking for alternative shipping routes. The company also faces production challenges following a magnitude 7.1 earthquake that struck Kumamoto in southwestern Japan on July 28. Toyota has production facilities in the area. Its Tahara plant has halted production for five days through Friday. The shutdown will then be followed by a scheduled summer break, meaning production will remain suspended until the end of August. For the full fiscal year ending March 2027, Toyota expects a net profit of 3.25 trillion yen ($20.6 billion), down from a record 3.85 trillion yen ($24 billion) in the previous fiscal year. Annual sales are projected at 54 trillion yen ($342 billion), up from 50.7 trillion yen in the fiscal year that ended in March. Toyota shares fell nearly 2 percent in Tokyo trading after the earnings announcement.

US companies are stepping up efforts to reduce dependence on China for critical minerals used in major weapons systems, as the Trump administration pushes to expand domestic production. In a small refinery inside a New Hampshire office park, Phoenix Tailings uses electricity to extract critical minerals from mining waste for use in key US weapons, including missiles deployed in the Iran war. The company recently received a $500 million Pentagon loan to expand its operations. However, building a new facility could take up to 18 months as the administration seeks to increase domestic supplies. The need has become more urgent as the conflict with Iran reduces US stocks of weapons such as Tomahawk cruise missiles and THAAD interceptors. At the same time, the White House is pressing defense contractors to increase production while imposing stricter rules against sourcing critical minerals from China. “It will be a tall order and a challenge to replenish these stocks and scale up in the timeframe needed to meet defense demand and regulations,” said Anthony Balladon, Phoenix Tailings’ chief commercial officer and co-founder. Although the US is moving faster to develop critical mineral supplies outside China's control, it could take years to build new mines and increase production of the powerful magnets needed for many advanced technologies. Mining waste could help meet demand in the meantime. Phoenix Tailings uses waste from traditional mining as well as recycled magnets and computer disk drives as raw materials. The Massachusetts-based company plans to build a larger facility to produce critical metals for the defense, aerospace and automotive industries. Balladon said the company calls the planned facility the “freedom facility” because its goal is to help the US and its allies reduce Chinese influence over rare-earth supplies. The company produces metals used to make strong permanent magnets for fighter jets, missiles, radar systems and drones. Balladon said critical minerals play an “outsized role” in defense systems, noting that an expensive weapons system cannot operate if it lacks minerals that may be worth only a small fraction of its total cost. The defense sector particularly depends on samarium and requires about 50 to 100 tons of it annually, while US production remains very limited. Phoenix Tailings currently produces only about 200 kilograms of the metal a year. Balladon said the company expects to increase output to about five tons within three months and reach a capacity of 120 tons by next year or 2028. US magnet maker turns to France for supplies Arnold Magnetic Technologies, which has long depended on China for samarium used in magnets for precision-guided weapons such as Tomahawk missiles, has turned to France for an alternative supply. The company is sourcing the material from Solvay, a Belgium-based chemical company that restarted processing rare earths from old mining waste in La Rochelle after stopping the operation in the 2000s. China built up much of the world's rare-earth processing capacity during that period and later used its dominant position to restrict exports of processed critical materials during a trade dispute with the US. Solvay has since resumed processing and added other rare earths, including samarium, to its operations. The company said growing demand for permanent magnets, geopolitical concerns over supply chains and demand for more diversified and traceable sources in Europe and the US are driving the expansion. Arnold CEO Matt Blake said the higher cost of sourcing samarium from Solvay rather than China has not been a major problem because the magnets account for only a small portion of the cost of a weapons system. However, concerns remain over supplies of other critical minerals. The US remains dependent on imports of tungsten, with China controlling about 80% of global mine production and an even larger share of processing, according to Brodie Sutherland, CEO of Patriot Critical Minerals Corp. Sutherland said US efforts to increase tungsten production will take years. Until then, the Pentagon and defense contractors may have to rely on existing stockpiles, increased recycling and limited supplies from countries other than China. Iran war adds pressure to supply chains The need to increase production has become more urgent as renewed fighting with Iran puts further pressure on US stocks of advanced missile interceptors, including Patriots and THAADs. A recent analysis by the Center for Strategic and International Studies said the conflict has raised concerns that the US could have less firepower available in a potential future conflict with China. The Washington-based think tank warned in May that replenishing stocks of Tomahawks, Patriots and THAAD interceptors could take at least three years. Meanwhile, President Donald Trump has ordered tighter restrictions on defense contractors sourcing critical materials from China. The July 20 executive order states that critical materials and components needed to manufacture, maintain and repair US military equipment should be sourced domestically or from allied countries. Lockheed Martin, which produces THAAD systems and other weapons, said it regularly assesses the global rare-earth supply chain to ensure access to critical materials. Several other major US defense contractors, including Raytheon Technologies and Northrop Grumman, did not respond to requests for comment. Balladon said he remained confident that US companies could expand production with the necessary support and partnerships. “Ultimately, with the right support and partners across the industry, we think we can make it happen,” he said.

An experiment using artificial intelligence (AI) at the annual Richard Wagner Festival in Bayreuth, Germany, received a mixed response on Saturday, with the AI-assisted stage production drawing boos from parts of the audience after leaving many spectators confused. According to German news agency dpa, curator Marcus Lobbes and his team were booed and whistled at as they appeared on stage after the performance of Götterdämmerung. However, the singers, musicians and especially conductor Christian Thielemann received warm applause. Festival organisers had announced before the performance that it would be the first time AI played a role on stage at the event, not as a performer but as a tool to generate visual projections. The production featured constantly changing AI-created images, while the singers remained the main focus on stage. The visuals combined light, textures, historical references and artistic themes, with the AI using images selected in advance by Lobbes and his team from the Academy for Theater and Digitality in Dortmund. The projections included scenes from earlier performances of Wagner's famous Ring cycle as well as historical images. According to dpa, many audience members were puzzled by the visual collage, which included images of former German Chancellor Helmut Kohl, the World Trade Center towers after the Sept. 11, 2001 attacks, and a postage stamp marking German reunification in 1990. The festival's website says each performance will be different because the AI-generated images and associations continuously change. Two more AI-assisted performances are scheduled.

Michigan on Saturday has reported cyberattacks targeting nine of its water systems, joining neighboring Minnesota, where more than 30 systems were recently affected, as the FBI investigates the incidents amid concerns over Iranian-linked cyber threats. Michigan officials said the state received a federal cyber alert on Tuesday warning of attempts to compromise operational technology used by water utilities. Shortly afterward, several communities reported suspicious activity matching the federal warning. Dale George, communications director at Michigan’s Department of Environment, Great Lakes, and Energy, said nine water systems were impacted but all continued operating safely. He said local operators resolved the issues and there were no known public health risks. The FBI said it is working with federal partners to protect critical infrastructure but has not identified those responsible. Last week, the FBI and the Cybersecurity and Infrastructure Security Agency warned that Iranian hackers have increasingly targeted water and wastewater systems and other critical infrastructure. Minnesota officials said most confirmed attacks involved remote monitoring and control technology rather than direct disruption of water services. Braham briefly asked residents to reduce water use after a cyberattack shut down its water treatment controls, while Plymouth restored communications after a separate attack. President Donald Trump blamed Minnesota officials for the incidents without providing evidence, prompting Democratic Gov. Tim Walz to respond that Trump knew the attacks had affected multiple states.

Ukrainian President Volodymyr Zelenskyy has asked US President Donald Trump to help secure approval from billionaire Elon Musk to allow Ukraine to use the Starlink satellite network for targeting military strikes inside Russia, according to people familiar with the discussions. The request came during Zelenskyy's meeting with Trump at the White House this week and was also raised in talks with US lawmakers on Capitol Hill. However, Trump did not commit to supporting the proposal, the sources said. Ukraine currently uses Starlink for communications within its own territory, including Russian-occupied areas. But Musk has not allowed the satellite system to be used for operations inside Russia. Zelenskyy has been pressing US officials for several weeks to expand Starlink's use, arguing that it would improve the accuracy of Ukraine's long-range and medium-range strikes. His request comes as Ukraine looks for alternatives to reduce its reliance on US-made Patriot air defence systems, which have become harder to obtain amid the ongoing US and Israeli military campaign against Iran. The White House, the Ukrainian Embassy in Washington and SpaceX, which owns Starlink, did not immediately comment on the matter. Allowing Starlink to be used inside Russia could also pose risks for Musk. Earlier this year, the Associated Press reported that intelligence agencies from two NATO countries believed Russia was developing an anti-satellite weapon that could target the Starlink network. The report could not be independently verified. Starlink has played a key role in supporting Ukraine's communications since Russia launched its full-scale invasion in February 2022. During his visit to Washington, Zelenskyy also urged US lawmakers to continue supporting Ukraine with more Patriot systems and to move forward with a proposed agreement that would allow Ukraine to produce the air defence system under licence. Republican Senator Mike Rounds, a member of the Senate Armed Services Committee, said Zelenskyy stressed that wider access to Starlink would make Ukraine's longer-range weapons more accurate. "He made it clear they need Starlink for greater precision," Rounds said. Zelenskyy met lawmakers before the Senate voted to advance a new package of sanctions against Russia, with a final vote expected next week. Meanwhile, Trump appeared to step back from an earlier statement suggesting Ukraine would receive a licence to manufacture Patriot systems. Speaking to reporters at Camp David on Friday, Trump said the issue was still under discussion. "We're talking about it. But it's a hard thing to give away that kind of technology," he said. Relations between Trump and Zelenskyy have improved in recent months as Ukraine has gained ground on the battlefield. Zelenskyy has also argued that the war in Ukraine is linked to tensions involving Iran. He plans to share intelligence with the United States that he says shows Russia is helping Iran use satellite surveillance to target US military bases in the Middle East. He also told US lawmakers that Ukraine is ready to share its expertise in drone warfare with the United States, highlighting its experience in countering Russian attacks.

The European Union is offering 10 billion euros ($11.4 billion) in public funding to help build seven large-scale artificial intelligence facilities as the 27-nation bloc seeks to narrow its technology gap with the United States and China. The European Commission said on Thursday that the funding could attract another 20 billion euros ($22.8 billion) in private investment. “Access to the raw scale of computing power within AI gigafactories is a strategic necessity for Europe as AI development accelerates,” said Henna Virkkunen, the Commission’s executive vice president in charge of tech sovereignty. The EU has been stepping up efforts to build what it calls “tech sovereignty” amid growing concerns about relying too heavily on foreign technology providers. European leaders fear such dependence could leave the bloc vulnerable if technology or critical supplies are used as leverage. US President Donald Trump has also sharply criticised the EU’s technology regulations, while China has restricted supplies of minerals that are vital to the technology industry. Companies can now bid to build the seven planned AI gigafactories, each expected to have at least 100,000 advanced AI chips. The facilities would be about four times more powerful than the data centers currently operating in the EU. Once the seven gigafactories are operational, the EU’s overall AI computing capacity will more than double. The bloc currently has 19 AI data centers operating across Europe, from Finland to Spain. Europe, however, remains well behind the US and China in several key areas needed to develop AI, according to a 2025 assessment by the US Federal Reserve. China has huge electricity capacity available for data centers, while US companies attract most of the world’s private investment in AI. The EU also faces higher costs and supply challenges. Europe does not produce many of the millions of components needed to build data centers, while electricity prices can be two to three times higher than in the US and China, according to a European Commission report presented to the European Parliament in June. The report warned that European companies and public authorities would continue to depend on US AI and cloud providers, putting local service providers at a disadvantage. It also said dependence on large foreign cloud and AI computing companies, particularly for critical services, could expose European data to access from outside the bloc and create risks to the continuity of essential services. French AI company Mistral currently operates one of the EU’s largest AI data centers at its Paris campus. The company develops the Le Chat chatbot but remains behind leading US firms such as OpenAI, the maker of ChatGPT, and Chinese competitors including DeepSeek. European leaders, including French President Emmanuel Macron, have repeatedly warned that the bloc needs stronger homegrown AI companies. At the same time, concerns are growing across Europe about the technology’s potential impact on jobs, the economy and privacy. The European Commission said AI products developed through the expanding network of data centers would have to meet EU standards on data protection, safety, security and ethics. The move comes amid wider concerns about the environmental impact of AI infrastructure. In June, 40 mayors from cities around the world signed an agreement seeking to limit the impact of AI data centers on natural resources, energy prices and climate targets.

Samsung Electronics on Thursday reported a record operating profit of 89.5 trillion won ($62 billion) for the April-June quarter, as the South Korean technology giant and rival SK Hynix continue to benefit from surging global demand for chips used in artificial intelligence systems. The earnings came a day after SK Hynix, the world's second-largest memory chipmaker, also reported record quarterly revenue, highlighting how the global AI boom is boosting South Korea's semiconductor industry. However, shares of both companies have fallen sharply this week amid concerns about their huge planned investments to expand chip production and growing competition from China. Samsung's second-quarter operating profit was more than 19 times higher than a year earlier, with almost all of the earnings coming from its semiconductor business. The chip division benefited from higher prices, strong demand for AI servers and increased sales of advanced high-bandwidth memory (HBM) chips used in AI applications. These gains more than offset losses in Samsung's mobile devices, televisions and home appliances businesses, which were affected partly by higher component costs. Samsung's quarterly revenue also reached a record 171.5 trillion won ($119 billion). The company said it expects demand for memory chips to remain strong during the second half of the year, supported by continued investment in AI infrastructure and wider use of so-called agentic AI. Demand for chips used in data centers is also expected to grow further, potentially keeping supplies below market demand, Samsung said. SK Hynix reported record second-quarter revenue of 60.5 trillion won ($42 billion) on Wednesday. However, its profit fell short of market expectations, sending its shares down more than 9 percent. Samsung's shares have also declined this week. Kim Jaejune, an executive in Samsung's memory chip division, said the gap between chip supply and demand could widen further in 2027. He said Samsung plans to begin construction of its second semiconductor factory in Taylor, Texas, before the end of this year, with the facility expected to start operations by 2030. Kim said Samsung has secured long-term chip supply agreements with five major global data center customers, possibly referring to Amazon Web Services, Google, Meta, Oracle and Microsoft. The company is also in talks with other major technology firms as demand for AI infrastructure continues to rise. "Despite our efforts to increase production, demand growth is outpacing our efforts," Kim said during a conference call. Samsung and SK Hynix together account for about two-thirds of global memory chip production. Both companies are making major investments in new semiconductor factories and data centers as South Korea seeks to strengthen its position in the global AI industry. However, some analysts have raised concerns about whether such large-scale spending will generate sufficient returns. South Korean chipmakers are also facing pressure from concerns over stronger competition from China. Reports that a state-owned Chinese company has started mass-producing domestically developed immersion deep-ultraviolet lithography machines, which are important for advanced chip manufacturing, have added to investor concerns. The strong stock market debut of Chinese memory chipmaker ChangXin Memory Technologies has also increased worries about competition from China, analysts said. Last month, Samsung and SK Hynix announced plans to jointly invest 800 trillion won ($554 billion) in a new semiconductor hub in southwestern South Korea to take advantage of rapidly growing AI-related demand. The heads of Samsung, SK and South Korean automaker Hyundai joined President Lee Jae Myung during his trip to San Francisco last week, where they announced hundreds of billions of dollars in planned cooperation with major US technology companies. The agreements involve AI firms OpenAI and Anthropic, as well as chip designers Nvidia and Broadcom, and cover chip technology, data centers and other AI infrastructure. Kim Yong-beom, President Lee's chief policy adviser, said at a foreign media roundtable last week that the initiatives show how global technology companies are seeking long-term partnerships with South Korean chipmakers to secure reliable supplies of memory chips for AI systems. However, the sharp swings in the share prices of Samsung, SK Hynix and other global technology companies suggest that investors are increasingly looking beyond short-term earnings and focusing on whether such growth can be sustained over the long term, semiconductor analyst Son In-joon of Eugene Securities said in a report this week.