The order, one of Beijing’s strongest moves yet against a foreign purchase of a Chinese tech company, came on Monday from China’s National Development and Reform Commission (NDRC), which said foreign investment in Manus would be prohibited under Chinese law, and the deal must be unwound.

Beijing is now concentrating on AI talent, software and intellectual property, and areas once taken over by chip restrictions now include artificial intelligence, as competition between China and the United States gets stronger

Chinese authorities began examining the acquisition in January, shortly after Meta completed the purchase in December. The review later intensified, and in March, Manus co-founders Xiao Hong and Ji Yichao were reportedly called to Beijing for talks with regulators and then barred from leaving China.

Neither founder publicly responded to requests for comment.

Meta has also not issued a public response.

Manus had drawn attention in China after launching what it described as a general AI agent in 2025. State-backed media had commended the company as a possible successor to DeepSeek, one of China’s most-watched AI firms.

Unlike model developers who build large language systems from scratch, Manus focused on agent software designed to complete multi-step tasks with limited human input. These tasks include coding, research and workflow automation.

Before the takeover, Manus raised $75 million in funding led by Benchmark in May 2025.

The company later shut its China offices and moved operations to Singapore, where its parent company, Butterfly Effect, was restructured. That move was seen as an attempt to attract foreign capital while easing both U.S. and Chinese restrictions.

Chinese regulators now appear determined to challenge that route.

The practice, sometimes called “Singapore washing”, involves Chinese-founded startups shifting legal structures or operations abroad while keeping roots in China. The latest development with Beijing reveals that strategy may no longer guarantee protection from investigations.

Startups moving overseas may not be enough as authorities may now demand proof of where management is headquartered, where research is done, where data is stored and who controls the company’s technology.

The China ruling could also create some problems for Meta, as some Manus staff had already moved into Meta’s Singapore offices, while parts of the startup’s work were reportedly being integrated into Meta projects.

Any reversal may now require separating teams, contracts and technology already tied together.

This is coming weeks before a planned summit in Beijing between U.S. President Donald Trump and Chinese President Xi Jinping in mid-May.

That meeting was expected to cover trade and technology tensions, but this issue now adds another case.

China has previously criticised foreign-linked deals involving strategic assets, but forcing the breakup of a completed transaction is rare.

China does not want core AI assets leaving its reach, no matter where a company later relocates.

The launch also includes Zia Agent Studio, a no-code agent builder, Zia Agent Studio, over 25 deployable Zia agents, and a Model Context Protocol (MCP) server to open up Zoho’s vast library of actions to third-party agents.

 “Today’s announcement emphasises Zoho’s longstanding aim to build foundational technology focused on protection of customer data, breadth and depth of capabilities because of the business context, and value,” said Kehinde Ogundare, country head, Zoho Nigeria. “Our LLM model is trained specifically for business use cases, keeping privacy and governance at its core, which has resulted in lowering the inference cost, passing on that value to the customers, while also ensuring that they are able to utilise AI productively and efficiently.”

Zia LLM: Custom-Built AI for the Enterprise

Zoho has successfully launched its own large language model, Zia LLM, built completely in-house by leveraging NVIDIA’s AI accelerated computing platform.

Trained with Zoho product use cases in mind, ranging from structured data extraction, summarisation, RAG, and code generation, Zia LLM comprises three models with 1.3 billion, 2.6 billion and 7 billion parameters, each separately trained and optimised for contextual applicability that benchmark competitively against comparable open source models in the market.

The three models allow Zoho to always optimise the right model for the right user context, striking the balance between power and resource management.

This focus on right-sizing the model is an ongoing development strategy for Zoho. In the short term, Zoho will scale Zia LLM’s model sizes, starting with the first set of parameter increases by the end of 2025.

While Zoho supports many LLM integrations for users, including ChatGPT, Llama, and DeepSeek, Zia LLM continues Zoho’s commitment to data privacy by allowing customers to keep their data on Zoho servers, leveraging the latest AI capabilities without sending their data to AI cloud providers.

The model is currently testing for internal use cases across Zoho’s broad app portfolio, and will be available for customer use in coming months.

Zoho also announced two proprietary Automatic Speech Recognition (ASR) models for speech-to-text conversion for English and Hindi.

Optimised to perform on a low computer load without compromising on accuracy, the models benchmark up to 75% better than comparable models across standard tests. Zoho will expand language support for ASR models to enable more inclusive AI adoption across diverse regions.

It will also introduce a reasoning language model (RLM).

Prebuilt Agents, Agent Studio and Marketplace Designed for Real-World Impact

To enable immediate adoption of agentic technology, Zoho has developed a roster of AI agents contextually baked right into its products.

These agents can be used across various business activities, handling relevant actions based on the role of the user.

These include Customer Service Agent for Zoho Desk that can process incoming customer requests, understand the context, and either answer directly or triage them to a human rep, providing an efficient first line of assistance.

Ask Zia, Zoho’s platform-wide conversational AI assistant, is bolstered with additional BI skills, tailored to data engineers, analysts, and data scientists, while supporting all users within an organisation.

It can build end-to-end data pipelines for engineers, analyse data, create reports and dashboards in an interactive conversation mode for analysts, or help jump start building ML models for data scientists.

First announced earlier in 2025, Zoho has further simplified the Zia Agent Studio experience to be fully prompt-based (with the option to use low-code) and includes ready-made access to over 700 actions across Zoho’s products.

Agents built by users can be deployed autonomously, triggered by button click, with rule-based automation, or even summoned within customer conversations.

At the time of deployment, an agent can be provisioned as a digital employee, maintaining the user access permission structure defined within the organisation.

Admins can perform behavioural audits as well as performance and impact analyses on digital employees, ensuring that every agent is working as effectively as possible and within clear guardrails.

Several pre-built agents are now available for users, such as Candidate Screener, which identifies and ranks the most suitable candidates for a specific job opening based on role requirements, skills, experience, and other key attributes; Deal Analyser, which can analyse deals and provide insights such as win probability, next best action, and follow-up suggestions, and Revenue Growth Specialist, which suggests opportunities for upsell and cross-sell for existing customers.

These agents are available in Agent Marketplace from where customers can easily deploy them. Ecosystem partners, ISVs, and individual developers will be able to create agents and host them on the Zia Agents Marketplace in coming months.

The company plans to add more skills to Ask Zia, allowing it to act as an assistant to Finance teams, Customer Support teams to start with.

Support for the Agent2Agent (A2A) protocol will be implemented, allowing Zia Agents to interact and collaborate with each other, as well as collaborate with agents on other platforms.

Interoperability and Governance

Zoho’s adoption of the Model Context Protocol (MCP) allows clients to securely access and interact with workflows and actions across over 15 Zoho applications. MCP is  available in early access, and integrations via Zoho Flow extend its reach to third-party tools.

Zoho Analytics now also supports a local MCP server, allowing for advanced contextual AI use cases while preserving data security and compliance

This is another step in Europe’s investigation of Chinese tech firms accused of flouting privacy rules.

The Berlin Data Protection Commissioner, Meike Kamp, invoked Article 16 of the EU’s Digital Services Act (DSA) in a formal notice to both tech giants, declaring DeepSeek’s app as illegal content under European law. 

At the heart of the matter is the company’s transfer of user data to China, without any of the legal safeguards mandated by the EU’s General Data Protection Regulation (GDPR).

Kamp stated, “DeepSeek has not been able to provide my agency with convincing evidence that German users’ data is protected in China to a level equivalent to that in the European Union.”

That alone would be enough to raise red flags. But the issue runs deeper.

Germany’s investigation found that DeepSeek violated Article 46 of the GDPR, which governs international data transfers. China does not have an EU adequacy decision, a prerequisite for transferring personal data outside Europe without further protections. 

Yet DeepSeek allegedly failed to implement even basic legal mechanisms, such as Standard Contractual Clauses (SCCs), that could have made such transfers lawful.

According to DeepSeek’s own privacy policy, the app stores an alarming range of personal data, including search queries, chat histories, uploaded documents, and location data, on servers based in China. Nowhere in the policy is GDPR mentioned. No safeguards are outlined. No clarity is given.

This isn’t the first time the company has faced European resistance. Italy’s data protection authority banned DeepSeek earlier this year after it failed to explain how it collects and processes user data. The Netherlands followed shortly after, warning the public not to submit sensitive information through the app.

In the United States, lawmakers are drafting legislation that would prohibit federal agencies from using AI models developed in China. A senior U.S. State Department official told Reuters that “DeepSeek is actively supporting China’s military and intelligence operations, including providing services to PLA research institutions.” 

The same report revealed that the company allegedly used shell companies in Southeast Asia to bypass U.S. export controls and acquire Nvidia H100 chips, restricted hardware used for training advanced AI models.

These concerns have been amplified by DeepSeek’s rapid ascent. The company claimed in January that its AI models, such as DeepSeek-R1 and V3, rival those of OpenAI and Meta—at a fraction of the cost. It reportedly trained its large language model for just $5.6 million, a figure many experts find highly questionable. 

Nevertheless, the apps have surged in popularity, topping download charts across multiple countries, and exposing users’ data to foreign jurisdictions.

Despite repeated requests from German authorities since May, DeepSeek refused to adjust its data practices or withdraw voluntarily from the app stores. With that deadline now past, the commissioner’s office has moved decisively.

Apple and Google are expected to act quickly, but neither company has responded to requests for comment. 

The new model is now available on Hugging Face, an AI development platform, where it has already gained attention for its improved reasoning and coding capabilities. 

DeepSeek claims that this version surpasses the previous model in multiple benchmarks, particularly in mathematical problem-solving and software development.

One of the improvements is its performance on the American Invitational Mathematics Examination (AIME), where it scored 59.4, a notable jump from the previous model’s 39.6. 

Similarly, on LiveCodeBench, a coding assessment, it gained 10 points to reach 49.2. These improvements suggest a more capable AI system for both research and practical applications.

With 685 billion parameters, DeepSeek-V3-0324 slightly surpasses the earlier V3 model’s 671 billion. Unlike old models, which used a proprietary commercial license, the latest version is distributed under the MIT license, making it more accessible to developers worldwide.

DeepSeek’s development has caught the attention of experts. “Anthropic and OpenAI are in trouble,” said Kuittinen Petri, a lecturer at Häme University of Applied Sciences, on X.

He tested the model by instructing it to “create a great-looking responsive front page for an AI company,” and it successfully generated a fully functional, mobile-friendly website with 958 lines of code.

Apple research scientist Awni Hannun ran the model on a 512GB M3 Ultra workstation. While it processed over 20 tokens per second, he noted that memory usage peaked at 381GB, which, though high, remained within expectations for a model of this scale.

DeepSeek’s progress in AI development has been commendably fast. After launching its V3 model in December and the R1 model in January, speculation is already building about the release of R2, a possible follow-up to its reasoning-focused series. “The coding capabilities are much stronger, and the new version may pave the way for the launch of R2,” said Li Bangzhu, founder of AIcpb.com.

Jasper Zhang, a University of California, Berkeley graduate and maths Olympiad gold medallist, also tested the model using an AIME 2025 problem. “It solved it smoothly,” he noted, asserting that DeepSeek’s models are closing the gap with their Western competitors.

Fahd Mirza, lead cloud and AI engineer at Australian construction materials company Boral, described DeepSeek-V3-0324 as “mind-blowing.” On his YouTube channel, he shared a demonstration of the model tackling complex coding and mathematical tasks, calling its performance “outstanding.”

DeepSeek’s approach to AI development has focused on efficiency and accessibility. Unlike heavily funded rivals, it operates with significantly fewer financial resources. Petri pointed out, “DeepSeek is doing all this with just [roughly] 2 per cent [of the] money resources of OpenAI.”

This means the tech giant is working to reduce dependence on external suppliers like Nvidia and lower infrastructure costs as it bolsters AI-driven innovations.

According to sources familiar with the matter, Meta has deployed the chip on a small scale and will expand production if testing proves successful. The chip is part of the company’s AI strategy, which includes a heavy focus on recommendation systems and generative AI products.

Developing its own AI chips, Meta seeks to control expenses and optimise performance. The company has projected expenses between $114 billion and $119 billion for 2025, with up to $65 billion allocated to capital expenditures, largely for AI infrastructure.

The new chip, specifically designed for AI workloads, is expected to be more power-efficient than conventional GPUs, which are typically used for AI training. Unlike general-purpose chips, Meta’s new hardware focuses solely on AI-specific tasks, potentially improving efficiency and cost-effectiveness.

Production of the chip is being handled by Taiwan Semiconductor Manufacturing Company (TSMC), a major player in the global semiconductor industry.

The test deployment began after Meta completed a critical stage in chip development known as “tape-out,” where an initial design is sent to a manufacturing plant. This process, which can cost tens of millions of dollars, is a key milestone in chip production.

Meta previously abandoned a similar project after early tests failed, opting instead to invest billions in Nvidia GPUs in 2022. 

Nonetheless, Meta is still one of Nvidia’s largest customers, using its hardware to train AI models that power recommendation systems, advertising tools, and its Llama series of foundation models.

The company has already deployed an earlier version of its in-house chip, known as the Meta Training and Inference Accelerator (MTIA), for AI inference—where AI systems generate responses based on user inputs. 

Meta now aims to expand its use of proprietary chips for AI training, which involves feeding massive amounts of data into models to improve their accuracy and capabilities.

Speaking at the Morgan Stanley technology, media, and telecom conference last week, Meta’s Chief Product Officer Chris Cox described the company’s chip strategy as progressing in stages. “We’re working on how would we do training for recommender systems and then eventually how do we think about training and inference for gen AI,” he said. He added that while the company is still in the early phases, the first-generation MTIA chip for recommendations was seen as a “big success.”

Meta’s goal to reduce reliance on Nvidia comes when the AI chip market is changing. The dominance of large language models trained on vast datasets has been challenged by new approaches focused on computational efficiency.

Chinese startup DeepSeek recently launched low-cost AI models that rely more on inference rather than extensive training, raising talks about the long-term sustainability of scaling up AI models with massive amounts of data. This change briefly led to a sharp decline in Nvidia’s stock value earlier this year, though the company has since regained most of its losses.

While Meta’s in-house chip build could eventually reduce costs and enhance AI performance, the company still has challenges in matching Nvidia’s advanced hardware capabilities. 

According to the research, 74.2% of DeepSeek-R1’s outputs share stylistic fingerprints with OpenAI’s technology, raising talks about possible reliance on OpenAI’s model during training.

This revelation has also led to discussions around data sourcing, intellectual property rights, and transparency in AI development. If DeepSeek-R1 was trained using OpenAI-generated content without disclosure, it could cause legal and ethical risks, including reinforcing biases and limiting diversity in AI-generated text.

The study employed an advanced text attribution method, utilising three independent AI classifiers trained on outputs from OpenAI, Gemini, Claude, and Llama. To ensure accuracy, a classification was only confirmed when all three classifiers reached the same conclusion. This approach resulted in a 99.88% precision rate, with a false-positive rate of just 0.04%.

During testing, DeepSeek-R1’s texts were found to align with OpenAI’s writing style in 74.2% of cases. In contrast, Microsoft’s Phi-4 model exhibited a 99.3% disagreement rate with existing AI-generated texts, indicating independent training.

Shai Nisan, Copyleaks’ chief data scientist, commented on the importance of the findings, stating, “With this research, we have moved beyond general AI detection as we knew it and into model-specific attribution, a breakthrough that fundamentally changes how we approach AI content.”

The research team, led by Yehonatan Bitton, Shai Nisan, and Elad Bitton, adopted a rigorous “unanimous jury” approach to ensure reliability of their findings. Their method went beyond identifying known AI models to also detecting previously unseen ones by analysing unique stylistic markers.

If DeepSeek-R1’s model was developed using OpenAI’s work without proper attribution, it could mislead investors and stakeholders about the originality of its technology. 

This ultimately points to cautiousness about AI governance, competitive fairness, and the risks of intellectual property infringement in the industry. Transparency in model training and attribution is highly important in maintaining trust and ensuring ethical development practices.

So far, the company has been at the top in the AI semiconductor market, a key driver of its advantage. However, recent developments in lower-cost AI models have pushed down its value, leaving investors to ponder over the indispensability of its premium chip and how the company can maintain its grip on the AI sector

Chinese AI firm DeepSeek has introduced models that claim to match the performance of Western alternatives at way lower costs. This development sent shockwaves through the industry earlier this year, leading to Nvidia suffering a record $593 billion single-day market value loss.

Nonetheless, Nvidia’s revenue is expected to show strong growth. Analysts predict a 72% rise in revenue for the fourth quarter, reaching $38.05 billion, though this would be its slowest growth rate in nearly two years. The forecast for the first quarter is positive, with projections of a 60% revenue increase.

Tech giants, including Microsoft and Meta, have reiterated their commitment to expanding AI-driven infrastructure, ultimately sustaining demand for Nvidia’s high-performance chips. “The CapEx plans communicated by Meta, Microsoft, Google and Amazon… paint a very positive picture of the near-term demand backdrop for Nvidia,” said John Belton, portfolio manager at Gabelli Funds, which holds shares in the company.

One of the biggest factors in Nvidia’s performance will be the rollout of its Blackwell series chips. These advanced AI chips, including the GB200 NVL72 system, represent a well-thought-out transition from selling standalone GPUs to offering fully integrated computing systems. 

However, the production ramp-up has been complex and costly, squeezing profit margins. Analysts estimate that Nvidia’s adjusted gross margin could decline by over three percentage points to 73.5% in the fourth quarter.

Manufacturing challenges have added to Nvidia’s issues. Taiwan Semiconductor Manufacturing Company (TSMC), its primary chip supplier, has faced difficulties expanding its capacity for advanced packaging—an essential process for AI chip production. Initial Blackwell shipments were also delayed due to design flaws and low yield rates, though these issues have since been resolved.

Nvidia however stated in November, that Blackwell’s revenue would go beyond initial projections of several billion dollars in the fourth quarter.

“Blackwell has been a complicated set of products to launch,” noted Belton. “With the magnitude of out-performance that investors have become used to—Nvidia’s delivery could be smaller this time around, just given some of these dynamics with the Blackwell launch.”

The surge was driven by strong demand for AI-powered computing infrastructure, helping the company offset current challenges in the global PC market. 

Net profit more than doubled to $692.7 million for the December quarter, although a substantial portion—$282 million—came from deferred tax credits.

Lenovo is capitalising on the growing demand for artificial intelligence by incorporating DeepSeek’s cutting-edge AI models into its products, including PCs and tablets. 

The Chinese startup has gained attention for providing AI models with high inferencing efficiency and lower computational costs, making AI adoption more accessible.

“DeepSeek has improved AI efficiency. The new models with a high inferencing efficiency and low computing power costs will democratise access to AI,” Lenovo CEO Yang Yuanqing stated during an interview with Reuters.

The company anticipates that AI-enabled PCs will make up 25% of its total shipments by 2025, potentially growing to 80% by 2027. This transition is expected to drive further demand for GPU servers, which are important for AI applications.

Infrastructure and Enterprise Segments Show Strong Growth

Lenovo’s infrastructure solutions division, which includes its AI server business, posted a 59% revenue increase, pointing to the thriving reliance on AI-driven computing. The solutions and services group, which provides enterprise cloud-based software, also saw a 12% rise in revenue, reaching $2.3 billion.

Even with its progress in AI, Lenovo is still competing well in the PC market. Its intelligent devices division, encompassing PCs, smartphones, and tablets, recorded a 12% revenue gain, showing a steady commercial PC replacement cycle.

Following the earnings report, Lenovo’s Hong Kong-listed shares initially surged but later declined by over 4%. However, the company’s stock has gained 17% this year, largely driven by investor optimism over its AI advancements and strategic partnerships with firms like DeepSeek.

Beyond PCs, Lenovo’s non-PC businesses now account for 46% of total revenue, with expansion into AI-powered computing solutions strengthening its competitiveness in AI-driven infrastructure.

Eager to provide more insight, the experts at AIPRM have revealed the five essentials people should be aware of before using DeepSeek:

1. What is DeepSeek designed for?

Real-time problem solving is at the core of DeepSeek-R1 – the free AI app, which was released in January 2025. It has the ability to analyse big complicated data sets in a quick and efficient manner, extracting key insights. DeepSeek also provides users with quick access to information specifically relevant to their needs, by focusing not only on keyword matching, but meaning and context as well.

2. Is your personal data safe?

A more thorough analysis of DeepSeek’s privacy policy reveals that its partners, including advertisers, share information with DeepSeek about your actions outside the AI app, such as your activity on other websites, or information on any products, or services you purchase online.

The AI tool also collects “keystroke patterns and rhythms”, which may never be deleted. This means DeepSeek tracks every button you’ve pressed on your keyboard, including for how long and the exact time you’ve done so.

Image

When it comes to the storage of all your personal data, the AI platform says they store it “for as long as necessary”, but doesn’t provide exact information on how they protect your data from unauthorised access, or whether they encrypt it.

3. Should you be worried about information bias? 

With DeepSeek being a Chinese company, it is regulated by Chinese law, which means it can censor any topics the Chinese government deem to be politically sensitive. Applying such censorship can in turn have an effect on the objectivity and accuracy of the information the platform provides. Certainly an important aspect to keep in mind before using the app.

4. Does DeepSeek pose a safety risk?

Recent research discovered DeepSeek’s safety measures can often be easily bypassed by users, which can result in the app providing harmful content, such as hate speech or threats, criminal and even self-harm materials. Furthermore, a whopping 83% of the bias tests researchers ran led to discriminatory output from the app. The tool’s high vulnerability to manipulation suggests it can prove dangerous and therefore, pose a safety risk.

5. Has the AI app been banned anywhere?

The number of countries that have imposed a ban on the use of DeepSeek continues to rise, as data privacy concerns are being raised. Italy was among the first countries to ban the app just last week.

Currently the following countries have imposed some type of ban on the usage of the AI tool:

• Italy– full ban on DeepSeek usage
• Taiwan– government agencies banned from using DeepSeek
• USA– banned in the Pentagon, the U.S. Congress, the U.S. Navy, NASA and Texas
• Australia– DeepSeek banned from all government devices.

Christoph C. Cemper, Founder at AIPRM, comments on the use of DeepSeek:

“While it’s true that current AI tools collect and store some type of user data, DeepSeek has been raising quite a few red flags with its privacy policy. This is one of the core reasons countries have begun to ban the app’s usage, with the US even proposing a law that could lead to people using DeepSeek receiving fines or possible jail time. One should be cautious of using the app, but in case you’re looking to start making use of it, the best course of action would be to thoroughly research  DeepSeek and familiarise yourself with the details of their privacy policy beforehand.”

[Featured Image Credit]

The pace of AI development is accelerating, and with it comes a profound question: will we rise to meet the opportunities and challenges, or will we let this transformation spiral beyond our control?

The world’s greatest thinkers agree – we are on the verge of something extraordinary. The emergence of AI capable of reasoning, learning, and even self-improvement will redefine what it means to be human.

But as we rush toward this future, we’re also standing on the edge of risks that could undo us entirely. The stakes couldn’t be higher.

The Accelerating Wave of Progress

History shows us that technological progress isn’t linear – it’s exponential. As futurist Ray Kurzweil’s Law of Accelerating Returns explains, each new innovation builds upon the last, compressing centuries of progress into decades and, eventually, years. What once took lifetimes now happens in the blink of an eye.

Take the 20th century as an example. In just a hundred years, we went from horses to cars, from telegrams to the internet, from rudimentary medicine to life-saving antibiotics. Now consider this: the technological leap of the 20th century is being repeated every few decades – and soon, every few years.

AI is the engine behind this acceleration. In just the past decade, we’ve seen tools like OpenAI’s GPT models, Google’s DeepMind solving decades-old scientific problems, and the rise of models like China’s DeepSeek, which has proven that cutting-edge innovation no longer requires a Silicon Valley address.

DeepSeek, for instance, achieved performance comparable to the best Western AI systems at a fraction of the cost, upending the global tech balance and sending ripples through markets.

But the story of AI isn’t about one breakthrough or company – it’s about a trajectory that’s racing faster than we can predict.

The AI Evolution: From ANI to AGI to ASI

To understand where we’re heading, let’s break AI into three stages:

1. Artificial Narrow Intelligence (ANI): This is where we are now. AI excels at specific tasks like language translation, driving, or predicting stock movements. Think Siri, Tesla, or Google Translate – powerful, but specialized.
2. Artificial General Intelligence (AGI): This is the next step. AGI will match human intelligence across all domains. It won’t just respond to questions; it will reason, learn new skills, and solve problems in ways that rival human ingenuity.
3. Artificial Superintelligence (ASI): This is where it gets both thrilling and terrifying. ASI would surpass human intelligence by orders of magnitude, solving problems we can’t even articulate today. Imagine a system that could cure every disease, reverse climate change, or even make death optional. But ASI could also lead to catastrophic outcomes if its goals don’t align with ours.

Here’s the kicker: the leap from AGI to ASI might happen in days, hours, or even minutes. Once an AGI can improve itself – a concept called recursive self-improvement – its intelligence could increase exponentially. What starts as a tool designed by humans could quickly surpass us in every conceivable way.

The Promise of AI &  The Risks We Can’t Ignore

It’s easy to fixate on the risks, but AI’s potential for good is staggering. Imagine a world where healthcare is revolutionized, education is democratized and climate change is tackled.

The promise of AI isn’t just technological – it’s deeply human. It’s the chance to solve problems that have plagued us for millennia and unlock possibilities we haven’t dared to imagine.

But with great power comes great responsibility. AI doesn’t share our values – unless we program it to. And even then, what happens if those values are misaligned or misunderstood?

The paperclip maximizer thought experiment illustrates this perfectly. Imagine an AI tasked with maximizing paperclip production. It might conclude that the most efficient path is turning all of Earth’s resources – including humans – into paperclips. It’s a simple example, but it highlights the risks of creating systems that operate on goals detached from human priorities.

Even today, we’re seeing glimpses of these challenges. Models like DeepSeek, for instance, reportedly avoid politically sensitive topics like Taiwan or Tiananmen Square, reflecting the biases – or agendas – of their creators.

As AI becomes more powerful, who decides what it values, and whose interests it serves?

The Urgency of Now

Here’s the truth: AI isn’t waiting for us to figure this out. The future is hurtling toward us, and we’re woefully underprepared.

To navigate what lies ahead, we need to act now. AI isn’t a competition – it’s a shared responsibility. Ensuring AI systems share human values – is humanity’s most urgent challenge. If we get this wrong, the consequences could be irreversible.

AI isn’t just a technical issue – it’s a societal one. Everyone, from policymakers to everyday citizens, needs to understand what’s at stake.

The rise of AI is inevitable, but its trajectory is not. It could usher in a golden age of human flourishing or lead us into disaster.

The difference lies in the choices we make today. History has handed us the pen to write the next chapter of human progress. Let’s make sure it’s a story worth telling.

*Written by: Heath Muchena, Founder of Proudly Associated and author of Artificial Intelligence Applied and Tokenized Trillions.