Artificial Intelligence News Highlights, 12 August 2025: South Korean start-up Upstage enters global AI competition with new model

South Korean artificial intelligence start-up Upstage has developed a new large language model with performance capabilities comparable to advanced systems from the United States and China, according to a report in the Financial Times. This development marks the company's entry into the global AI competition, which has largely been dominated by developers in the US and China.

A new study indicates that the routine use of artificial intelligence to assist in colonoscopies can negatively impact the skills of healthcare experts, according to a report in the Financial Times. The research suggests that consistent reliance on AI systems, which act as a 'second pair of eyes' by highlighting suspicious polyps in real-time, may lead to a degradation of a specialist's own diagnostic abilities, a phenomenon known as 'automation-induced deskilling'. This finding comes amid the rapid adoption of these AI tools, which have been shown to improve adenoma detection rates. The study raises concerns about maintaining the quality of care in situations where AI is unavailable and highlights the need for medical training programs to adapt to ensure practitioners maintain their core competencies.

AI startup Character.AI is shifting its focus from its original goal of developing superintelligence to operating as an entertainment company, as per a report in Wired. The company's new CEO has stated that Character.AI is now focused on its entertainment offerings, which cater to a user base of 20 million. This strategic pivot moves the company away from long-term Artificial General Intelligence (AGI) research and toward a business model centered on interactive stories and character-based experiences.

OpenAI is in talks to invest in a brain implant start-up, a move that would challenge Elon Musk's Neuralink in the brain-computer interface (BCI) sector, according to a report in the Financial Times. The discussions, led by OpenAI's Sam Altman, signal a potential expansion for the AI company into hardware and direct human-computer interaction. This development would also intensify the existing rivalry between Altman and Musk, whose companies OpenAI and xAI already compete in the artificial intelligence landscape. The report clarifies that the talks are exploratory and a deal has not been finalized, and the specific start-up involved has not been named.

SoftBank Group Corp.’s shares jumped as much as 8% on Tuesday, driven by investor bets that the company can capitalize on its long-term focus on artificial intelligence, according to a report in Bloomberg. The surge in optimism is linked to signs of progress on 'Stargate,' a reported multi-billion dollar AI data center initiative. This market reaction reflects renewed confidence in SoftBank's AI-centric strategy, which is also bolstered by its majority ownership of chip designer Arm Holdings and a recent performance turnaround in its Vision Funds.

AI company Anthropic has increased the context window for its Claude model to one million tokens, which is equivalent to about 750,000 words or 2,000 pages of text, according to a report in TechCrunch. This update allows the model to process and reference vast amounts of information, such as entire codebases or extensive financial documents, in a single prompt. The development positions Claude competitively against rivals like Google's Gemini 1.5 Pro, which also offers a one-million-token window, and surpasses the capacity of other models. This enhancement enables developers and enterprises to build more sophisticated applications for tasks like complex code analysis, legal document review, and in-depth research without needing to break down large inputs into smaller pieces.

Artificial intelligence and analytics provider Fractal Analytics has filed for a $560 million initial public offering in Mumbai, according to a report in Bloomberg. The IPO filing could result in a valuation of more than $3.5 billion for the company, which is recognized as India's first AI unicorn.

Leonis Capital has secured $25 million for a new fund aimed at investing in emerging artificial intelligence talent, according to a report in Bloomberg. The firm, which is linked to an unnamed former OpenAI researcher, has a stated goal of helping to "uncover the next OpenAI," indicating a focus on backing individuals and small teams working on foundational AI models and research.

Growing investor fears that generative artificial intelligence tools could disrupt major software companies have triggered a significant selloff across the sector, wiping out billions in market capitalization, according to a report in Bloomberg. The concern is that AI's ability to automate tasks like coding, content creation, and data analysis may reduce demand for traditional software applications. This has put pressure on large, incumbent software firms to rapidly integrate AI capabilities or risk losing market share, leading investors to re-evaluate the valuations of companies perceived as slow to adapt.

Indian tech entrepreneur Bhavin Turakhia is partnering with GoDaddy Inc. to launch his AI-led email service, Titan, to GoDaddy's customer base, according to a report in Bloomberg. The service is priced at $2 per user and is positioned as a competitor in the enterprise email market currently dominated by Microsoft and Google. This partnership provides Titan with immediate access to GoDaddy's extensive distribution channel, targeting small and medium-sized businesses with an affordable, AI-enhanced email alternative.

Bitcoin mining company MARA Holdings Inc. has agreed to a $168 million deal to acquire a majority stake in Exaion from Electricite de France SA (EDF), according to a report in Bloomberg. The acquisition is a strategic move by MARA, identified as the largest Bitcoin miner, to extend its operations into the artificial intelligence sector. By purchasing the EDF unit, MARA aims to become an infrastructure provider for AI, leveraging its existing large-scale computing capabilities.

AI search startup Perplexity has made an unsolicited $34.5 billion offer to acquire Google Chrome, a bid significantly higher than its own reported valuation of approximately $18 billion, according to a report in The Verge. A company spokesperson confirmed the bid, which was made on Tuesday. Perplexity's Dmitry Shevelenko stated that multiple large investment funds have agreed to fully finance the transaction. The offer follows a previous statement by the startup that it would purchase Chrome if antitrust regulators forced Google to sell the browser, though Google has not indicated any willingness to sell.

Nvidia's rise to a $4 trillion company is attributed to foundational work from its research lab, which positioned the company to dominate the artificial intelligence market, as per a report in TechCrunch. A key historical development was the creation of the CUDA platform, which allowed its GPUs, originally for gaming, to be used for general-purpose parallel computing. This technological groundwork proved essential for the subsequent boom in deep learning and Large Language Models (LLMs), which created massive demand for Nvidia's AI accelerator chips. The company's advantage is further solidified by a comprehensive software ecosystem built around its hardware, creating a high barrier to entry for competitors and making its technology a critical strategic asset in the global AI landscape.

Artificial intelligence firms are competing for government contracts by offering their products at heavily discounted rates, according to a report in The Verge. Within the last week, both OpenAI and Anthropic have introduced special pricing for government-specific versions of their AI chatbots, offering federal agencies access to ChatGPT and Claude, respectively, for a nominal fee of $1 for a one-year term. Anthropic is reportedly targeting all three branches of the US government with its offer. These recent moves follow a similar initiative from xAI, which announced its "Grok for Government" platform in mid-July 2025.

Western Australia is suffering the worst coral bleaching on record, caused by what scientists describe as the most intense marine heatwave ever recorded in the region, according to a report in BBC. While not a factor in the event itself, this environmental crisis underscores the increasing reliance on Artificial Intelligence. Scientists are deploying AI and machine learning models to improve the accuracy of marine heatwave forecasts, providing crucial early warnings. Furthermore, AI-driven computer vision is used to automate the large-scale analysis of underwater imagery to assess the extent of bleaching and overall reef health, offering more timely and comprehensive data than traditional methods.

Computer science graduates are reporting increased difficulty in securing tech jobs due to a combination of widespread industry layoffs and the growing adoption of AI-powered coding tools, according to a report in The New York Times. The trend is primarily driven by a broader tech industry downturn that began in 2022, with the integration of AI serving as an accelerant. Companies are leveraging AI tools to boost developer productivity, which has led to a shift in hiring priorities toward candidates with specialized expertise in AI and machine learning. Consequently, graduates seeking generalist software engineering positions face a more competitive market, while those with specific AI-related skills may still find robust opportunities.

GitHub CEO Thomas Dohmke is stepping down as the company faces intensified competition in the AI developer tool market, TechCrunch says in a report. The departure occurs as GitHub's flagship AI tool, GitHub Copilot, is challenged by significant advancements from competitors. These include Google's development of sophisticated AI coding systems powered by its Gemini model and the rise of AI-native code editors such as Cursor, which build AI functionality directly into the development environment.

Google is rolling out a new feature that enables users to define their top sources, which will result in seeing more stories from those selected sites in search results, according to a report in TechCrunch. This function gives users a direct method to influence the output of Google's AI-driven search algorithms by prioritizing content from publishers they trust. The development is a response to the increasing volume of AI-generated content and aims to enhance user control over information quality.

An investigation by The Markup and CalMatters has found that dozens of data brokers are intentionally hiding their personal data opt-out pages from search engines like Google, according to a report in Wired. The companies use methods such as 'noindex' tags to prevent these crucial privacy pages from being discoverable, effectively hindering individuals from exercising their legal right to data deletion. This practice has significant implications for artificial intelligence, as personal data that individuals want removed can remain in datasets used to train AI models, complicating efforts to ensure ethical AI development and data governance.

The UK's tax authority, HMRC, is deploying artificial intelligence to analyze the social media accounts of individuals suspected of tax evasion, according to a report in BBC. The technology is designed to identify potential discrepancies by comparing a person's declared financial status with their lifestyle as portrayed on public social media posts. HMRC has clarified that the AI system will serve as an investigative tool and will not replace human decision-making, ensuring that human officials remain responsible for final judgments. The agency also stated that the use of this AI is subject to legal oversight to comply with data protection regulations.

Former Uber Freight CEO Lior Ron has been appointed as the new chief operating officer of Waabi, an AI-powered self-driving truck startup, according to a report in TechCrunch. The move comes as Waabi prepares to scale its commercial operations in anticipation of launching its autonomous trucks on public highways later this year. Ron's appointment is aimed at strengthening the company's leadership as it transitions from AI technology development to commercial deployment.

Google is rolling out a new feature called “preferred sources” that allows users to manually choose which news outlets they want featured most prominently in Search’s “top stories” section, as per a report in The Verge. The feature, now available in the US and India, lets users select their favorite outlets, which will then appear more frequently in top stories and in a new “from your sources” section. This development is noted as a non-AI personalization tool, introduced as Google continues to address reports that its AI-driven search features are negatively impacting web traffic to publisher sites. Google has stated that click volume has remained “relatively stable” and that users are increasingly navigating to websites with diverse content like forums and videos.

Despite common misconceptions that it is a data broker or data miner, Palantir is a software company that provides advanced platforms for data analysis, according to a report in Wired. The company's core offering, including its Artificial Intelligence Platform (AIP), enables large government and commercial clients to integrate their own vast and disparate datasets. Using AI and machine learning, the software helps these organizations analyze their information to identify patterns, make predictions, and support complex operational decisions in fields ranging from national security to supply chain management.

OpenAI is actively modifying its strategy for its next major model rollout, reportedly referred to as GPT-5, to address several critical challenges, according to a report in VentureBeat. The company is focusing on stabilizing its underlying technical infrastructure to handle the demands of more advanced AI. Concurrently, OpenAI is working to fine-tune personalization features for users and is developing methods to effectively moderate what the report describes as more "immersive interactions." These adjustments reflect an iterative development process that is responsive to technical and safety considerations ahead of a full public release.

Elon Musk has threatened to sue Apple, accusing the company of unfairly favoring rival AI firm OpenAI through its planned operating system integration. The public threat is a response to Apple's recent announcement of "Apple Intelligence," which will see OpenAI's ChatGPT technology deeply embedded into iOS, iPadOS, and macOS. Musk, whose company xAI competes with OpenAI, argues that this system-level integration provides an unfair advantage over other AI applications, such as his own chatbot, Grok, and has previously raised security concerns over the partnership.

The increasing integration of AI agents into the workplace is producing mixed results, with businesses seeing efficiency gains while employees express concerns about job security and the need for new skills, as per a report in ZDNet. This trend involves the widespread embedding of AI "copilots" into enterprise software and the use of intelligent automation for complex cognitive tasks. While companies stand to benefit from increased productivity, they also face challenges with implementation, data privacy, and managing employee morale. Consequently, there is a growing focus on upskilling the workforce with AI literacy and developing governance frameworks to ensure responsible deployment.

Apple's push into generative artificial intelligence, highlighted by its 'Apple Intelligence' suite, poses a significant challenge to its 2030 net-zero climate goal, according to a report in Wired. The energy-intensive nature of AI, particularly the 'Private Cloud Compute' component that relies on large data centers, is projected to substantially increase the company's electricity consumption. This development complicates Apple's environmental ambitions at a time when it also faces pressure from US-China trade tariffs to diversify its manufacturing supply chain, potentially disrupting established sustainability programs and making it more difficult to reduce emissions across its entire value chain.

Nvidia has added support for its AI-powered DLSS 4 Multi Frame Generation technology to Grand Theft Auto V Enhanced and Senua’s Saga: Hellblade II Enhanced, as per a report in The Verge. The feature, which requires Nvidia's upcoming RTX 50-series graphics cards, utilizes artificial intelligence to generate additional frames to create smoother gameplay motion. A new Game Ready driver (580.97) has also been released to enable the new support and provide performance improvements for both titles.

YouTube is set to begin testing a new age-verification system in the U.S. that uses artificial intelligence to infer a user's age based on the types of videos they watch, AP says in a report. The initial tests will be limited to a small segment of logged-in users in the country. The system, which is an expansion of a similar technology already deployed in other regions, will be implemented more broadly if the U.S. trials prove successful.

A US medical journal has issued a warning about the dangers of using ChatGPT for health information, according to a report in The Guardian. The warning comes after the Annals of Internal Medicine published a case study of a 60-year-old man who developed bromism, or bromide toxicity, a rare condition. The man reportedly developed the condition after consulting the AI chatbot for advice on removing table salt from his diet, highlighting the risks of relying on general-purpose AI for medical guidance.

Chinese regulators are demanding that major tech companies, including Alibaba and ByteDance, provide justification for their purchases of Nvidia's H20 AI chips, as per a report in Financial Times. The H20 chip is a less powerful processor specifically designed by Nvidia to comply with US export controls targeting China. This governmental scrutiny places Chinese firms in a difficult position, caught between needing advanced hardware for AI development and Beijing's strategic push for semiconductor self-sufficiency and control over key technology acquisitions.

OpenAI is reportedly addressing a user revolt following widespread complaints about a perceived performance degradation in its latest AI models, according to a report in Wired. The backlash is evident on social platforms like Reddit, where users describe recent updates as an "erasure" of capabilities rather than an innovation, pointing to a "dumbing down" or increased "laziness." While the report's title mentions the unreleased GPT-5, the user feedback specifically references the current GPT-4o model, indicating the complaints are focused on its perceived decline in performance since its initial release.

Elon Musk announced the shutdown of Tesla's Dojo AI supercomputer project, which was developed in-house to train the company's Full Self-Driving (FSD) software, as per a report in TechCrunch. In a post on the social media platform X, Musk stated that the project was an "evolutionary dead end" and that the decision was made after it became "clear that all paths converged to AI6," an apparent internal reference to a new AI development path. The shutdown also involved "tough personnel choices," indicating staff changes related to the project.

Nearly one in ten of the fastest-growing YouTube channels globally now consist entirely of mass-produced, surreal AI-generated content, according to a report in The Guardian. This increase in artificial videos, spurred by recent breakthroughs in text-to-video AI technology, includes content such as animated cat soap operas and videos of babies in space. Several of these channels have rapidly gained millions of subscribers, highlighting a significant new trend on the video platform, although at least one such channel mentioned in the report has since been closed.

New research highlights significant security and privacy risks associated with the rapid advancement of "OS agents"—AI systems capable of controlling computers and phones like a human, according to a report in VentureBeat. These agents, powered by breakthroughs in large language models, can interpret and execute complex, multi-step tasks across an operating system. The study warns that their inherent need for broad system access creates major vulnerabilities. If compromised, these agents could be used to exfiltrate sensitive data, execute unauthorized commands, or facilitate sophisticated social engineering attacks, posing a new class of threat to both individual users and enterprise cybersecurity.

Seoul-based Datumo has raised $15.5 million in a funding round with backing from Salesforce to challenge competitor Scale AI, TechCrunch said in a report. The investment comes as many organizations report being unprepared to use generative AI in a safe and responsible manner. Citing a recent McKinsey report, the article highlights AI explainability as a key concern, noting that while 40% of organizations view it as a significant risk, only 17% are actively addressing it.

Anthropic has launched a memory function for its Claude chatbot, allowing users to ask the AI to search and summarize past chats to continue projects, The Verge said in a report. The feature, which is not a persistent memory and does not build a user profile, is rolling out to Claude's paid subscription tiers and works across web, desktop, and mobile platforms. According to Anthropic spokesperson Ryan Donegan, Claude will only retrieve past chats when a user explicitly requests it. This development is part of an ongoing competition with OpenAI, which previously launched its own memory feature, as both companies compete to increase user engagement and market share.

Video platform Rumble Inc. is considering an all-stock acquisition of German data center firm Northern Data AG to build out its artificial intelligence cloud computing services, as per a report in Bloomberg. The proposed deal would value Northern Data at approximately $1.17 billion. This strategic move is aimed at boosting Rumble's capabilities as a cloud computing provider, specifically targeting the high-demand market for AI infrastructure and compute capacity.

India is undertaking a major national effort to develop artificial intelligence capable of handling the country's vast linguistic diversity, which includes hundreds of languages and thousands of dialects, according to a report in BBC. The project aims to create a public digital infrastructure for Indian languages through government initiatives like 'Bhashini' and academic contributions from groups such as AI4Bharat. These efforts focus on building models for translation, speech-to-text, and text-to-speech to improve digital inclusion, access to government services, and educational resources for its citizens.

A company in Malaysia has created a new AI large language model specifically designed to serve the global Muslim population, which is estimated at around 2 billion people, according to a report in Bloomberg. The project leverages foundational open-source AI technology from the Chinese AI firm DeepSeek. This development is part of a wider trend in which AI systems are being localized to meet the unique cultural, linguistic, and religious needs of specific communities, addressing concerns about potential biases in general-purpose AI.