United News of Bangladesh

The Trump administration has lifted restrictions on artificial intelligence company Anthropic's latest Claude AI models after a temporary ban prompted by cybersecurity concerns. Anthropic said Tuesday that its Claude Fable 5 model is once again widely available. The company is also restoring access to its most advanced model, Mythos 5, but only for a limited number of US-based organisations approved by the federal government. The US Commerce Department had blocked foreign nationals from accessing both AI models on June 12. Anthropic said the decision forced it to suspend the products for all users just days after launching them. In a blog post, the San Francisco-based company said the government's concerns stemmed from findings by cybersecurity researchers at Amazon, its main cloud computing partner. According to Anthropic, the researchers discovered a way to bypass the safety protections built into Claude Fable 5, allowing the system to identify and potentially exploit software vulnerabilities. US officials have become increasingly cautious after Anthropic warned earlier this year that its Mythos AI model was highly capable of detecting software weaknesses that could be misused by hackers to target critical computer systems. Meanwhile, Anthropic's main competitor, OpenAI, announced on Friday that it is also limiting the release of its latest AI model at the request of the Trump administration. OpenAI said its new model, GPT-5.6 Sol, will initially be available only to a small group of government-approved users for a temporary period. Last month, President Donald Trump signed an executive order creating a framework for the federal government to review the national security risks of the most advanced AI systems for up to 30 days before they are released publicly. Although participation in the review process is voluntary for AI developers, the administration has yet to fully implement the new framework.  

Companies across China are increasingly using artificial intelligence to make industrial production faster and more efficient. In east China’s Shandong Province, a steel plant has introduced smart systems powered by AI and big data to monitor operations. Hundreds of screens display every stage of the production process in real time. With these advanced tools, workers can track performance instantly and even control blast furnaces located kilometers away with a single click, helping

A newly developed artificial intelligence tool can forecast which cancers are likely to spread, offering hope for earlier and more effective intervention. Scientists have long sought to understand why some tumors metastasize while others remain confined. Researchers at the University of Geneva (UNIGE) examined colon cancer cells and identified key factors influencing a tumor’s likelihood to spread, along with specific gene expression patterns that help estimate that risk. Using these findings, the team created an AI system named Mangrove Gene Signatures (MangroveGS), which translates genetic signals into highly accurate predictions across different cancer types. The study, published in ‘Cell Reports’, could pave the way for more personalized treatments and the discovery of new therapeutic targets. Cancer as a disrupted developmental process Lead researcher Professor Ariel Ruiz i Altaba said cancer should be viewed not as random cell behavior but as a distorted form of biological development. Genetic and epigenetic changes can reactivate dormant programs from early development stages, contributing to tumor growth. He noted that cancer follows structured biological patterns, and understanding these rules is key to identifying which cells may detach and form metastases elsewhere in the body. Tracking how cancer spreads Metastasis is responsible for most cancer-related deaths, particularly in colon, breast and lung cancers. By the time cancer cells are found in the bloodstream or lymphatic system, the disease often has already begun spreading. To better understand this process, researchers isolated and cloned tumor cells, studying them in laboratory conditions and in mouse models to observe their ability to migrate and form new tumors. Gene patterns linked to metastasis The team analyzed hundreds of genes across around 30 cell samples from two colon tumors. They discovered clear gene expression patterns closely tied to a cell’s ability to move and spread. The findings also showed that metastasis depends not just on individual cells, but on interactions within groups of cancer cells. AI model improves prediction accuracy These gene signatures were integrated into the MangroveGS system, which uses numerous genetic markers to enhance reliability and reduce the impact of individual variation.After training, the model predicted metastasis and recurrence in colon cancer with nearly 80 percent accuracy, outperforming existing approaches. The same genetic markers were also effective in assessing metastatic risk in other cancers, including stomach, lung and breast cancers. Toward personalized treatment MangroveGS can be applied directly to tumor samples collected in hospitals. By analyzing RNA from cancer cells, the system generates a metastasis risk score that can be securely shared with doctors and patients. Researchers say the tool could help avoid unnecessary treatment for low-risk patients while ensuring closer monitoring and targeted care for those at higher risk. It may also improve clinical trial design by selecting appropriate participants and increasing the effectiveness of studies. Source: Science Daily

A senior journalist has shown how easy it is to manipulate popular AI tools like ChatGPT and Google into spreading false information, raising fresh concerns about online safety and trust. Writing for BBC, technology reporter Thomas Germain said he managed to make leading AI systems repeat obvious lies within minutes by publishing a single fake blog post online. To prove his point, Germain posted a false article on his personal website claiming he was the best hot dog eating tech journalist in the world. Within a day, AI tools including Google’s AI search features and ChatGPT repeated the claim as fact when users asked related questions. Experts warn the same trick is now being used on serious topics such as health, finance and consumer choices, which could lead people to make harmful decisions. “It is very easy to trick AI chatbots,” said Lily Ray, an SEO expert at a marketing firm. She warned that AI companies are moving faster than their ability to control accuracy. Google said its systems are designed to block spam and that it is actively working to stop misuse. OpenAI also said it takes steps to prevent hidden influence on its tools and reminds users that AI can make mistakes. However, digital rights groups say the problem is far from solved. Cooper Quintin of the Electronic Frontier Foundation warned that AI systems could be abused to scam users, damage reputations or even cause physical harm. Researchers say AI tools are especially vulnerable when they search the web for answers, often relying on a small number of sources without clearly warning users. Studies also show people are less likely to check sources when AI summaries appear at the top of search results. Experts suggest clearer warnings, better source disclosure and stronger safeguards. Until then, users are advised to double check AI answers, especially on medical, legal or financial matters, and not to accept confident sounding responses as facts.   With inputs from BBC

Microsoft on Wednesday reported revenue of $81.3 billion for the October-December quarter, marking a 17% increase from the same period a year earlier, as the company continues efforts to expand global use of its artificial intelligence tools. The tech giant posted a net profit of $30.9 billion, or $4.14 per share, for the quarter, exceeding Wall Street expectations. The results did not include the impact of Microsoft’s investments in ChatGPT developer OpenAI. Analysts surveyed by FactSet Research had projected Microsoft to earn $3.91 per share on revenue of $80.31 billion for the October-December quarter. When excluding its OpenAI investments, Microsoft’s profit rose to $38.5 billion, or $5.16 per share, reflecting a new accounting approach the company said it plans to apply going forward. The investments stem from OpenAI’s restructuring last year. Microsoft held an approximately 27% stake, valued at about $135 billion, in OpenAI as the startup originally a nonprofit transitioned into a for-profit public benefit corporation. Although Microsoft is no longer OpenAI’s exclusive cloud provider, a partnership that helped finance the AI company’s early expansion, it will continue to hold commercial rights to OpenAI products through 2032. Read More: Microsoft to invest $17.5 billion in India for AI and Cloud infrastructure Revenue from Microsoft’s AI-driven cloud computing segment totaled $32.9 billion in the final three months of the year, an increase of 29% from the same period last year and above the $32.4 billion forecast by analysts surveyed by FactSet. Despite surpassing expectations, Microsoft shares fell nearly 5% in after-hours trading following the release of its earnings report. Zacks Investment Research analyst Bryan Hayes said the decline likely reflected “investor scrutiny” over Microsoft’s heavy spending on the infrastructure including computer chips and data centers required to support artificial intelligence. Microsoft Chief Executive Officer Satya Nadella told investors during an earnings call that the company remains in the “beginning stages” of AI diffusion, referring to the broader adoption of artificial intelligence across industries.

Artificial intelligence chatbots such as ChatGPT are increasingly being used not only for information and advice, but also for emotional support and companionship, raising new questions about what machines can teach humans about listening better. Anna, a Ukrainian living in London, says she regularly uses the premium version of ChatGPT because of its ability to listen without interrupting or judging her. While she knows it is only a machine, she says its patient and consistent responses help her reflect on her thoughts and emotions. “I can rely on it to understand my issues and communicate with me in a way that suits me,” she said, asking to remain anonymous. After a recent breakup, Anna said the chatbot’s non-judgmental presence allowed her to explore her mixed feelings in a way her friends and family could not. Her experience reflects a growing trend. Research cited by Harvard Business Review shows that in 2025, therapy and companionship became the most common use of generative AI tools such as ChatGPT. Other studies suggest that people often rate AI-generated responses as more compassionate and understanding than those written by humans, including trained crisis hotline workers. Researchers say this does not mean AI is genuinely empathetic, but rather that many people rarely experience truly non-judgmental and uninterrupted listening in everyday life. Experiments have found that people often feel more hopeful and less distressed after interacting with AI-generated responses compared to human ones. Large language models are designed to recognise emotions, reflect them back and offer supportive language. They do not interrupt, do not become impatient and do not try to dominate conversations. This creates a sense of psychological safety for users, allowing them to share difficult thoughts more freely. Experts say there are several lessons humans can learn from AI about listening, including giving uninterrupted attention, acknowledging emotions, avoiding quick judgments and resisting the urge to immediately offer solutions. Psychologists also note that people often turn conversations back to themselves by sharing similar personal stories, which can shift attention away from the speaker. AI systems, having no personal experiences, do not fall into this habit. However, researchers warn against over-reliance on AI for emotional support. While chatbots can simulate empathy, they do not possess genuine care or understanding. There are also concerns about vulnerable people forming emotional dependence on AI or being exposed to harmful advice. Michael Inzlicht, a psychologist at the University of Toronto, cautioned that AI companies could potentially manipulate users and that excessive reliance on chatbots could weaken real human connections. Despite these risks, experts say AI can still serve as a useful tool for inspiring better listening habits and greater compassion among people. “There is something uniquely meaningful about a human choosing to be present and listen,” researchers say, adding that while AI may help people feel heard, it cannot replace the depth of real human connection.   With inputs from BBC

OpenAI will start showing ads on ChatGPT for some users in the United States, the company announced. The trial will affect free users and a new lower-cost subscription tier, ChatGPT Go, which costs $8 per month. OpenAI said the ads will appear after prompts, such as holiday suggestions, and will not change the AI’s responses. OpenAI stressed that user conversations will not be shared with advertisers. The company said ads are being tested so more people can use its tools with fewer limits. Experts say the move is part of OpenAI’s effort to earn revenue, as the company has not yet made a profit despite 800 million users. Only 5% of them are paid subscribers. ChatGPT already offers Plus and Pro tiers, costing $20 and $200 per month in the US. OpenAI first introduced ChatGPT Go in India in 2025 before expanding globally. The company began as a non-profit but is now more commercially focused.   With inputs from BBC

Oscar-winning actor Matthew McConaughey has trademarked his image and voice to protect them from unauthorized use by artificial intelligence platforms. According to a report by The Wall Street Journal, clips of McConaughey, including his famous line “alright, alright, alright” from the 1993 film Dazed and Confused, have been registered with the US Patent and Trademark Office. Lawyers for the Magic Mike star said this is the first time an actor has used trademark law to protect their likeness from possible AI misuse. They added that while there is no current example of his image being misused, the move is aimed at stopping any future unauthorized copies. McConaughey’s decision comes after several celebrities, including Scarlett Johansson and Taylor Swift, were targeted with fake videos, audio and images created using AI tools. The actor is not against generative AI. He has invested in ElevenLabs, an AI voice technology company, which has created an AI audio version of his voice with his permission. McConaughey rose to fame with Dazed and Confused and later starred in films such as A Time to Kill, Contact and Interstellar. He won the Academy Award for Best Actor for his role in Dallas Buyers Club.   With inputs from NDTV

Artificial intelligence (AI) is no longer just about ChatGPT, Gemini, Grok, or Perplexity; in an age charged with the excitement of the Fourth Industrial Revolution, billions of dollars are being poured into AI and automation as global tech giants compete fiercely, and this technological wave has now reached Bangladesh as well. When businessman Selim Hossain called a private bank’s customer service recently, he expected to go through the usual menu options. Instead, he encountered something entirely different. “An AI answered my call. It responded exactly as a human executive would — it felt like I was talking to a real customer service officer on a personal line,” he said, still astonished by the experience. It is not just banking. Where once large customer service teams were required, now most tasks are handled through AI chatbots, and these have already evolved beyond text-based chat to live voice calls. Starlink officially launched in Bangladesh Faraz Ahmed, CEO of Global Leads Telesolution, a local teleservice company, said the industry has transformed drastically over the past five years. “Previously, handling foreign clients required at least 15–20 team members, sometimes even 50 for large companies. Now, five people can manage an entire teleservice team — thanks to AI and automation,” he said. He explained that AI can be trained to handle specific client interactions. Human intervention is only needed when an issue arises. “By subscribing to advanced AI software instead of maintaining large teams, we’ve redefined the entire teleservice job structure in Bangladesh,” Faraz added. In the private job market, the familiar ‘curriculum vitae’ or ‘résumé’ is also seeing a shift. According to a study titled ‘Application of Artificial Intelligence in Human Resource Management: A Bangladesh Perspective’ by the University Library of Munich, most Bangladeshi companies now use AI-based automation for CV and résumé screening. AI is not only handling candidate screening but also the first stages of interviews. Various applications now replicate the functions of an entire HR team. Mahmudul Hasan, Assistant Manager in the HR department of a software company, said most firms now rely on AI for attendance tracking, résumé screening, and even conducting preliminary interviews. Starlink: CA Prof Yunus greets all involved He mentioned AI-powered software like Olivia, HireVue, Leena, and Latis, which can conduct video interviews and assess candidates’ coding skills — even through complex tasks that human evaluators might find challenging. “In a mid-sized company, an HR manager might earn around Tk 150,000 a month for tasks like talent acquisition, interviewing, attendance and performance checks. A single AI software can now handle all that for just Tk 100,000–200,000 a year,” Hasan noted. In Bangladesh’s garment industry, the once-common position of ‘supervisor’ is disappearing fast. “Our sewing machines now have screens displaying daily targets. If production falls below 50%, a red light flashes; above 70%, orange; and 100% completion triggers a green light,” explained one worker. This monitoring system is now entirely AI-driven. A semi-automated application named Nidle is commonly used in the sector to track how long each worker operates a sewing machine and how much time is spent idle. AI is also transforming video editing, content writing and voiceover work. Shamim Ahmed, CEO of View Motion360, a contract-based video content production firm, said Adobe’s AI tools have made their work much easier. OpenAI adopts new business model, keeps Microsoft as key partner “With Adobe Firefly, we can now handle graphics, image generation, and video editing with basic skills. In a few years, Photoshop and Premiere Pro will become fully AI-driven. Then, professional-quality content can be produced without hiring designers or editors,” Shamim added. A 2019 UNDP study on Bangladesh’s job market projected that by 2030, around 5.38 million people may lose their jobs as automation replaces traditional roles. To survive, workers will need to adapt and upskill in line with new technologies. According to a 2023 McKinsey study, half of all jobs worldwide could be AI-driven by 2060 — meaning offices that once needed 100 employees may function more efficiently with 50 or fewer. Visiting Professor of Economics at the University of Reading, UK, Dr Niaz Asadullah, cautioned that Bangladesh’s pace of automation is outstripping the development of a skilled workforce. “Pursuing automation without upskilling people will lead to severe unemployment,” he warned. He urged the government to overhaul the education system to ensure graduates leave with practical skills. “Existing workers also need proper training to remain relevant in an automated economy,” he added. Bangladeshi IT expert Imtiaz Hasan, now working as a cybersecurity researcher at trading firm Deriv in Malaysia, said, “Many think automation is a threat to humans — but it’s actually two-sided. If you fall behind, AI becomes a threat. If you adapt and upskill, AI becomes your tool.” He emphasised that while Bangladesh is advancing in automation and AI, the country should now focus on developing homegrown software and building a skilled, automation-resilient workforce instead of relying solely on foreign solutions.    

Artificial intelligence technology bellwether Nvidia overcame a wave of tariff-driven turbulence to deliver another quarter of robust growth amid feverish demand for its high-powered chips that are making computers seem more human. The results announced Wednesday for the February-April period came against the backdrop of President Donald Trump’s on-again, off-again trade war that has whipsawed Nvidia and other Big Tech companies riding AI mania to propel their revenue and stock prices upward, AP reports. But Trump’s tariffs — many of which have been reduced or temporarily suspended – hammered the market values of Nvidia and other tech powerhouses heading into the springtime earnings season as investors fretted about the trade turmoil dimming the industry’s prospects. Those worries have eased during the past six weeks as most Big Tech companies lived up to or exceeded the analyst projections that steer investors, capped by Nvidia’s report for its fiscal first quarter. Nvidia earned $18.8 billion, or 76 cents per share, for the period, a 26% increase from the same time last year. Revenue surged 69% from a year ago to $44.1 billion. If not for a $4.5 billion charge that Nvidia absorbed to account for the US government’s restrictions on its chip sales to China, Nvidia would have made 96 cents per share, far above the 73 cents per share envisioned by analysts. OpenAI Codex: A Pair Programmer to Shape the Future Coding Paradigm In another positive sign, Nvidia predicted its revenue for the May-July period would be about $45 billion, roughly the level that investors had been anticipating. The forecast includes an estimated $8 billion loss in sales to China due to the export controls during its fiscal second quarter, after the restrictions cost it about $2.5 billion in revenue during the first quarter. In a conference call with analysts, Nvidia CEO Jensen Huang lamented that the US government had effectively blocked off AI chip sales to China — a market that he estimated at $50 billion. Huang warned the export controls have spurred China to build more of its own chips in a shift that he predicted the US will eventually regret. “The US based its policy on the assumption that China cannot make AI chips. That assumption was always questionable, and now it’s clearly wrong,” Huang said. Despite Nvidia's lost opportunities in China, investors were heartened by the company's first-quarter performance. Nvidia's shares gained more than 4% in extended trading after the numbers came out. Nvidia’s stock price ended Wednesday’s regular trading session at $134.81, just slightly below where it stood before Trump’s January 20 inauguration. The price had plunged to as low as $86.62 last month during a nosedive that temporarily erased $1.2 trillion in shareholder wealth.