Day: July 22, 2026

Singapore is moving to sharpen the rules around how companies use personal data in generative AI, as the country tries to square two competing realities: businesses want more data to build AI systems, while users are increasingly asked to trust tools they barely understand.

At the Singapore Data Festival on Monday, Minister for Digital Development and Information Josephine Teo announced three new initiatives: advisory guidelines on the use of personal data in generative AI, transparency guidelines for AI chatbots, and a Digital Twin for Enterprises Playbook aimed at helping companies deploy real-time virtual models of physical operations.

Also Read: Beyond the hype: What generative AI is actually changing in startups

The announcements are not a dramatic regulatory crackdown. They are closer to a tightening of expectations. But for startups, banks, telcos, SaaS firms, customer service platforms, and AI application developers in Southeast Asia, they point to where Singapore wants the market to go: more disclosure, clearer consent, and fewer vague privacy notices hiding broad data use.

“Without good data, even the best systems will struggle to produce useful outcomes,” Teo said. “That is why data governance matters more, not less, in the age of AI.”

Consent can no longer hide in boilerplate

The most consequential move is the Personal Data Protection Commission’s Advisory Guidelines on the Use of Personal Data in Generative AI.

Under Singapore’s Personal Data Protection Act, organisations already need consent to collect, use or disclose personal data unless exceptions apply. The new guidelines clarify what that means when companies use personal data to develop, improve or fine-tune generative AI models.

Teo used the example of a customer service team wanting to train a generative AI model on call recordings. Such recordings often contain names, addresses, billing information and other personal details. Under the new guidance, companies should not rely on generic wording buried in privacy policies. They should state clearly that customer data may be used to train or improve AI models.

That matters because the generative AI supply chain is messy. A startup may build an app using a third-party large language model, fine-tune it with customer conversations, host it on cloud infrastructure, and integrate analytics from another vendor. When something goes wrong, such as data leakage, hallucinated advice or inappropriate use of public datasets, accountability can quickly become diluted.

The PDPC guidelines also address roles and responsibilities across the AI value chain, as well as due diligence when organisations rely on publicly available data.

Also Read: Without governance, AI agents risk becoming enterprise chaos engines

For Southeast Asia, this is not an abstract compliance problem. The region’s digital economy is projected to reach around US$1 trillion by 2030, according to regional policy and industry estimates, and much of that growth will depend on cross-border data flows, platform trust and AI-enabled services. Yet data protection rules remain uneven across ASEAN, with Singapore, Malaysia, Thailand, Indonesia, Vietnam and the Philippines at different stages of enforcement and regulatory maturity.

Singapore is effectively trying to set the operating standard before bad practices become entrenched.

Chatbots may soon come with ‘information cards’

IMDA is also launching Generative AI Chatbot Transparency Guidelines, beginning as a voluntary framework.

The key idea is a Chatbot Information Card. Teo compared it to the label on medicinal products: not a full technical manual, but a plain-language summary of what the chatbot is for, what it is not for, how data may be handled, and how users can report issues.

This targets a real gap. Most users do not read terms of service documents. Even if they do, the relevant information is often scattered across privacy notices, AI disclaimers and product documentation. For consumer-facing AI services, the result is a dangerous grey zone: users may disclose sensitive information without understanding how it is stored, reviewed or used to improve systems.

DBS, Google, Meta, OCBC and Singapore Airlines are among the early adopters that will use the guidelines as a reference point. Google is expected to consolidate key information about its Gemini app, while Meta will provide clearer information on how users interact with its AI-powered products.

The competitive implications are broader. Gemini competes directly with OpenAI’s ChatGPT, Anthropic’s Claude and Microsoft Copilot, all of which are fighting for enterprise and consumer adoption in Asia. Meta AI is being pushed through social platforms with massive regional reach, particularly in markets where Facebook, Instagram and WhatsApp remain default digital infrastructure.

For banks such as DBS and OCBC, the pressure is also regional. Rivals including UOB, Maybank, CIMB and Kasikornbank are all experimenting with AI across fraud detection, customer engagement and operations. A chatbot transparency norm in Singapore could quickly become a benchmark for financial institutions operating across ASEAN.

Digital twins move beyond large enterprises

The third initiative, IMDA’s Digital Twin for Enterprises Playbook, is aimed at a different but related problem: helping companies turn operational data into useful AI systems.

Also Read: Why emerging markets need AI governance infrastructure before AI scale

A digital twin is a real-time virtual representation of physical assets, systems or processes. Large industrial companies, logistics operators, airlines and Formula One teams have used such systems for years. The government now wants smaller enterprises to see them as practical tools rather than futuristic toys.

Teo cited Exceltec, a Singapore facilities management company that built a digital twin drawing on sensor data from more than 70 customer sites. The system monitors issues such as air-conditioning faults or unusual water usage that may indicate leaks. According to Teo, the system saves each team about 45 minutes a day on each inspection.

Exceltec operates in a crowded facilities management and building services market that includes players such as CBM, C&W Services, ENGIE Services, Sodexo and Surbana Jurong-linked service providers. For smaller operators, digital twins could become a way to compete on predictive maintenance rather than manpower-heavy inspection routines.

Across Southeast Asia, the timing is relevant. Cities are adding sensors to buildings, utilities and transport networks, while property owners face rising energy costs and pressure to improve sustainability reporting. In markets such as Singapore, Malaysia, Thailand and Vietnam, digital twins are likely to be pulled into smart building, manufacturing and logistics use cases.

But the playbook’s “legal guide” framing is telling. The government is not just pushing adoption; it is warning companies that data architecture, consent, security and accountability cannot be bolted on later.

ASEAN context: AI safety is local, not universal

Teo also pointed to January’s AI Safety Red Teaming Challenge, where more than 80 experts from all ASEAN countries, as well as China, India, Japan and Korea, tested whether generative AI applications could leak protected data.

The findings underline a problem global AI companies often underplay: model safety does not travel neatly across languages and cultures. Some harmful requests refused in English were answered in Khmer. Casual local phrasing could bypass safeguards that worked against formal prompts.

Also Read: Safeguarding your organisation in the age of increasing AI

That is a serious issue for Southeast Asia, where hundreds of languages and dialects sit alongside uneven digital literacy and fast AI adoption. Guardrails built primarily for English-speaking users may fail in local contexts.

“None of us can build a trusted data ecosystem by looking only within our own borders,” Teo said.

Singapore will assume the ASEAN Chairmanship next year and has signalled that trusted data use will be part of its regional digital agenda. The challenge will be moving beyond voluntary frameworks and high-level alignment. For startups and enterprises, the direction is already clear: if AI is trained on user data, users need to be told plainly; if chatbots interact with the public, their limits must be visible; and if companies want to extract value from operational data, governance has to start before deployment, not after the first breach.

The post Singapore turns AI scrutiny towards chatbots, personal data, and digital twins appeared first on e27.

Kaspersky researchers say they have uncovered a long-running cyber-espionage campaign targeting government and diplomatic organisations in Southeast Asia, using a Go-based remote access trojan designed less for smash-and-grab theft than for patient intelligence collection.

The campaign, which Kaspersky’s Global Research and Analysis Team has dubbed GoSerpent, was identified in July 2026. According to the company, the operation uses a customised toolset that includes the GoSerpent backdoor, Stowaway, and TmcLoader, suggesting a campaign built for persistence, stealth, and staged data exfiltration.

Also Read: After cyber attacks, silence can be the biggest brand killer: Penta’s Dan La Russo

The finding lands in a region where state-linked cyber operations have become part of the wider geopolitical weather. Southeast Asia sits between major powers, hosts critical shipping lanes, has a dense web of diplomatic missions, and is rapidly digitising public services. For attackers seeking policy intelligence, trade positions, defence information, or diplomatic cables, ministries and embassies in the region are high-value targets.

A campaign built around patience

Kaspersky described GoSerpent as a sophisticated remote access trojan that has been active since at least 2021, with the latest known variant deployed this year. Written in Go, the malware uses persistence mechanisms and filenames that mimic legitimate system processes, a familiar but effective trick to reduce visibility inside compromised systems.

What makes this campaign notable is not just the tooling but the tempo. Rather than immediately deploying every payload after gaining access, the attackers appear to wait before moving to secondary tools used for exfiltration.

“What stands out about GoSerpent is the deliberate dwell time. Usually, attackers want to move quickly once they get a foothold, but this group drops the initial backdoor and waits,” said Noushin Shabab, Lead Security Researcher in Kaspersky GReAT. “They let the dust settle for weeks before deploying their secondary exfiltration tools like TmcLoader.”

That delay, she added, is designed to outlast standard log retention policies and automated security sweeps, making it harder for defenders to connect the first compromise with the later theft of data.

For under-resourced government agencies, this is a serious problem. Many public-sector systems in the region still run on uneven security budgets, fragmented vendor environments, and legacy infrastructure. Even where agencies have endpoint protection and monitoring in place, long dwell times can expose gaps in logging, incident response, and cross-agency threat sharing.

Why Southeast Asia remains a prime target

Kaspersky said the victims were government and diplomatic entities in Southeast Asia, though it did not name specific countries or agencies. That omission is typical in cyber-espionage reporting, where disclosing victims may trigger diplomatic fallout or reveal ongoing investigations.

The regional context matters. ASEAN has placed cybersecurity on the policy agenda through efforts such as the ASEAN Cybersecurity Cooperation Strategy 2021-2025 and the ASEAN-Singapore Cybersecurity Centre of Excellence. Singapore, Malaysia, Thailand, Indonesia, Vietnam, and the Philippines have all expanded national cyber agencies or regulations in recent years. Yet the region remains uneven: Singapore has mature cyber governance, while several neighbours continue to grapple with shortages of skilled personnel, decentralised IT procurement, and weak incident disclosure cultures.

Also Read: Hackers using AI to mask identity behind cyber attacks, researchers say

That unevenness gives advanced threat actors room to operate. A compromised diplomatic network in one country can become a stepping stone to broader intelligence on ASEAN negotiations, defence partnerships, trade deals, maritime disputes, or foreign investment decisions. In the context of the South China Sea, supply-chain realignments, semiconductor policy, and digital trade rules, diplomatic inboxes and internal files are not mere administrative targets; they are intelligence assets.

For Southeast Asia’s startup ecosystem, the lesson is indirect but important. Govtech vendors, cloud service providers, cybersecurity startups, system integrators, and managed service providers increasingly sit inside public-sector supply chains. Attackers do not always need to breach a ministry head-on if a contractor with weaker controls offers a cleaner route in.

Possible link to TetrisPhantom

Kaspersky said it suspects a connection between GoSerpent and the TetrisPhantom threat actor, citing overlaps in victimology, technical capabilities, and operational methods. The company has not made a definitive attribution and said further investigation is continuing.

TetrisPhantom has previously been associated with cyber-espionage activity against government entities in Asia-Pacific, including campaigns that drew attention because of their focus on highly specific targets and operational discipline. A possible link, if eventually confirmed, would reinforce the view that GoSerpent is not a commodity malware campaign but part of a more targeted intelligence operation.

Attribution in cyber-espionage remains a messy business. Security vendors typically rely on infrastructure overlaps, malware similarities, victim profiles, operational timing, and tradecraft. None of these alone is conclusive. Groups also reuse tools, borrow techniques, and deliberately plant false flags. Kaspersky’s cautious wording is therefore notable: the company is flagging similarities without declaring a firm actor behind the campaign.

Beyond tools: the hard part is visibility

Kaspersky’s advisory urges organisations to watch for GoSerpent indicators of compromise and strengthen detection, response, email security, digital footprint monitoring, compromise assessments, and incident response readiness. Stripped of the product language, the underlying point is straightforward: agencies need better telemetry and longer memory.

A campaign that waits weeks between initial access and exfiltration is betting that defenders will lose the trail. Short log-retention windows, siloed security teams, and over-reliance on automated alerts all work in the attacker’s favour. For diplomatic and government networks, where the data value is high and intrusions may be quiet, security teams need the ability to reconstruct events across endpoints, servers, identity systems, and email environments over extended periods.

Also Read: Are cyber attacks more life-threatening than we think?

The GoSerpent disclosure is not a mass-market ransomware story. There is no public claim site, no splashy ransom note, and no immediate operational shutdown. That makes it less visible but arguably more consequential. In espionage campaigns, success is measured by what remains unknown: how long the attacker stayed, what they read, and which decisions they influenced before anyone noticed.

The post GoSerpent exposes the quiet cyber war against Southeast Asian governments appeared first on e27.

For years, much of Southeast Asia’s digital economy has been built around a quiet compromise: if users wanted access to the best technology, they often had to meet it in English. That bargain is beginning to shift.

New usage data around Google’s Gemini suggests that generative AI in the region is increasingly being used in local languages, not just by urban professionals writing emails or developers debugging code, but by farmers, older users, students, creators and small business owners who are more comfortable thinking, speaking and selling in their mother tongue.

Also Read: Beyond the chatbot: How Gen Z pioneers are leading ASEAN’s new AI revolution

The clearest signal comes from Vietnam, where 89 per cent of Gemini prompts are now written in Vietnamese. Thailand is close behind, with 87 per cent of prompts in Thai, while Indonesia records 84 per cent in Indonesian.

The figures point to a deeper change in how AI is being adopted across Southeast Asia: usefulness is no longer simply about model size or speed, but about whether the system understands the language, idioms and cultural context of the person using it.

That matters in a region where English has long been treated as the default language of technology, even though it is not the default language of daily life for hundreds of millions of people.

The rise of native-language AI

The shift is not just anecdotal. The report cites the Southeast Asia Holistic Evaluation of Language Models, or SEA-HELM, a benchmark that assesses how large language models perform across regional languages. Gemini is ranked as the best-performing large language model overall for Southeast Asian languages in the evaluation, which covers Burmese, Filipino, Indonesian, Malay, Tamil, Thai, and Vietnamese.

For founders and developers, the implications are practical. A chatbot that works well in English may serve a bank’s urban customers, but it will not necessarily help a farmer in northern Vietnam, a shopkeeper in rural Thailand or a student in an Indonesian public school. To reach those users, AI products need to understand not only grammar and vocabulary, but also local phrasing, intent and cultural references.

This is where native-language performance becomes more than a technical milestone. It expands the addressable market for startups building education tools, financial services, health access platforms, customer support agents, creator tools and productivity apps. In a region as fragmented as Southeast Asia, language has often been a barrier to scale. Better multilingual AI could turn it into a distribution advantage.

Malaysia shows a slightly different pattern. English remains dominant for professional and coding-related tasks, reflecting the country’s multilingual workforce and its long-standing role as a regional services hub. But prompts in Malay have doubled in early 2026, according to the report. That suggests users are not abandoning English so much as switching languages depending on the task: English for work, Malay for learning, creativity or cultural expression.

A tea farmer and the economics of translation

The most compelling example in the source material comes from Lao Cai, a mountainous province in northern Vietnam known for its highland communities and ancient Shan Tuyet tea trees.

A small-scale tea farmer there once depended on middlemen to reach foreign buyers. The problem was not merely logistics; it was language. Selling premium tea to customers in Europe or the US requires more than listing weight and price. It involves storytelling, product descriptions, tasting notes, invoices and trust-building communication. Without fluent English, the farmer was stuck at the edge of the value chain.

Using Gemini, he can now describe his thoughts in Vietnamese and ask the AI to rewrite them in English “like an expert tasting a fine wine”. The output gives him polished descriptions that help position his tea for international buyers while preserving the authenticity of his own story.

This is a small example, but it captures why language-capable AI could matter for Southeast Asia’s small businesses. Translation has traditionally been treated as a support function. In practice, it can determine who captures value. If a farmer, craft producer, homestay owner or independent creator can communicate directly with global customers, they may keep more of the margin that previously went to intermediaries.

Also Read: Indonesia’s AI momentum: Big investments, bigger questions

For startups, this opens space for tools that combine AI translation with payments, logistics, compliance, product photography, storefront creation and customer relationship management. The opportunity is not simply to build another chatbot, but to help local businesses cross borders without losing their voice.

Culture is harder than vocabulary

Localisation is often described as a language problem. In Southeast Asia, it is also a cultural one.

In Malaysia, the report highlights marketing coordinators using Gemini in Malay to brainstorm visual concepts that capture the jiwa, or soul, of local culture. Terms such as lepak, referring loosely to the relaxed act of hanging out, or the familiar glow of kopitiam lighting, carry emotional weight that does not survive cleanly in literal English translation.

This distinction matters for the region’s creative economy. Southeast Asian brands increasingly want to participate in global digital culture without flattening their identity into generic international English. AI tools that understand local nuance could help agencies, content creators and small brands produce work that feels specific rather than templated.

It also matters for inclusion. In Thailand, the report notes that users over 54 are the most multimodal age group, using voice and image prompts in Thai to navigate daily tasks. That hints at another frontier for AI adoption: people who may not type comfortably, may not speak English, or may prefer to show the AI something rather than describe it.

In markets where ageing populations, rural connectivity gaps and uneven digital literacy remain real constraints, voice and image-based AI in local languages could be more transformative than text-only productivity tools aimed at office workers.

Rivals are racing for the same multilingual future

Gemini’s regional language performance puts Google in a strong position, but it is far from alone. OpenAI’s ChatGPT remains widely used across Southeast Asia, especially among English-speaking professionals, students and developers. Anthropic’s Claude has gained traction for writing and analysis-heavy workflows, while Meta’s open-source Llama models are attractive to developers and enterprises that want more control over deployment. Singapore’s AI Singapore has also developed SEA-LION, a family of language models focused on Southeast Asian contexts.

The contest will not be won only by benchmark scores. Distribution, pricing, developer tools, enterprise trust, government relationships and data governance will all matter. In Southeast Asia, one additional factor may prove decisive: whether the model can handle the region’s messy linguistic reality, where users mix English, local languages, dialects, slang and visual cues in the same conversation.

Why this matters for Southeast Asian startups

The Philippines remains an outlier in the Gemini report, with 90 per cent of prompts currently in English, the highest share in the region. That reflects the country’s strong English-language education base and its role in outsourcing, customer support and global services. But the broader direction is clear: the regional internet is becoming more multilingual, not less.

For startups, this changes product assumptions. Interfaces built only for English-speaking urban users will miss large segments of the market. Customer support bots will need to handle code-switching. Education apps will need to explain concepts in the language students use at home. Commerce platforms will need product descriptions that work across borders but begin in local speech.

Also Read: How Vietnam is emerging as a leading AI builder ecosystem in Southeast Asia

The bigger point is that AI adoption in Southeast Asia may not follow the same path as in the US or Europe. Here, the breakthrough use case may be less about replacing white-collar workflows and more about removing the language barriers that have kept millions of people from fully participating in the digital economy.

If generative AI can speak the language of the user’s home, it may become not just a productivity tool, but infrastructure for a more inclusive regional internet.

The post Southeast Asia’s AI future is being written in Vietnamese, Thai, Indonesian appeared first on e27.

Southeast Asia‘s AI moment is arriving, but not in English. Across Vietnam, Thailand, and Indonesia, a quiet but consequential shift is under way: researchers, startups, and governments are building large language models trained on local languages, dialects, and cultural contexts, rather than waiting for Silicon Valley to localise its tools.

The stakes are significant. English-centric models systematically underperform in Bahasa Indonesia, Thai, and Vietnamese — languages spoken by more than 400 million people. That gap is now attracting serious capital and policy attention. In Vietnam, homegrown LLM efforts are being backed by state-linked entities and private labs alike. In Indonesia, local language AI has become a national priority tied directly to digital sovereignty. Thailand is investing in Thai-language models to serve its healthcare, legal, and public services sectors.

What’s emerging is not a pale imitation of Western AI stacks, but a distinct regional layer, one that reflects local grammar structures, cultural nuance, and governance priorities. For Southeast Asia’s founders and investors, this is both an infrastructure play and a market-creation opportunity. The developers who own the language layer may ultimately own the region’s AI future.

REGIONAL

SEA’s AI future is being built in local languages: Researchers and startups across Vietnam, Thailand, and Indonesia are developing local-language LLMs, challenging English-centric models that underperform across SEA’s 400 million-plus non-English speakers.

SBI buys majority stake in Coinhako for Singapore digital assets push: Japanese financial giant SBI Group has acquired a majority stake in Singapore-based crypto exchange Coinhako, deepening its foothold in the city-state’s regulated digital asset market.

KWAP moves to contain damage after eFishery fraud shock: Malaysian pension fund KWAP is managing exposure from the eFishery accounting fraud, as the fallout from one of SEA’s most high-profile startup collapses continues to ripple through institutional investors.

eFishery ex-CEOs get sentences cut on appeal: A Jakarta court reduced prison terms for eFishery’s former co-founders, following their conviction over the US$600M fraud that shook Southeast Asia’s agritech sector.

Ant International closes US$1.2B series A with Ant Group and Alibaba: Singapore-based Ant International raised US$1.2B in a series A led by its parent entities, signalling continued commitment to cross-border payments and financial services expansion across SEA.

CMBI and SMBC back Whale’s US$40M series C extension: The enterprise AI firm secured a US$40M series C extension co-led by CMBI and SMBC, funding its push to scale AI-driven solutions across Asian enterprises.

SEA IPO fundraising surges 85% in H1 2026: Capital raised through Southeast Asian IPOs jumped 85% year-on-year in the first half of 2026, according to EY data, pointing to renewed investor confidence in public market listings across the region.

VinFast opens 20 Indonesia e-motorcycle dealerships: Vietnamese EV maker VinFast has launched 20 electric motorcycle dealerships across Indonesia, accelerating its regional two-wheeler rollout in SEA’s largest automotive market.

Vietnam emerges as SEA’s next data centre contender: Vietnam is positioning itself as a major data centre destination as regional capacity shifts and hyperscaler demand grows, according to a BMI analysis.

Malaysia poised to become regional data centre hub: A new report identifies Malaysia as a leading candidate for Southeast Asia’s data centre expansion, driven by land availability, energy policy, and growing hyperscaler interest.

WBBA convenes APAC’s first broadband summit in Bangkok: The World Broadband Association held Asia-Pacific’s inaugural broadband development summit in Bangkok and launched a new AI-Net certification standard for regional connectivity infrastructure.

Maybank foresees visible growth ahead for JustCo: Malaysia’s largest bank says Singapore co-working operator JustCo is on track for meaningful expansion, backed by recovering office demand and new markets.

OpenAI capacity strained by surging Asian startup demand: Asian startups are pushing OpenAI’s infrastructure to its limits, with surging API demand from the region testing the company’s ability to scale supply to fast-growing markets.

---

INTERVIEWS & FEATURES

Inside SEA’s AI gold rush: the 20 biggest cheque writers: A deep-dive into the 20 investors deploying the most capital into SEA’s AI sector, mapping who is leading deals, at what stages, and in which verticals.

How Vietnam evolved from execution hub to product powerhouse: A feature tracing Vietnam’s technology industry’s shift from outsourcing centre to homegrown product builder, driven by a new generation of founders with global ambitions.

How Gen Z pioneers are leading ASEAN’s new AI revolution: A profile of young founders across ASEAN who are building AI-native companies that go well beyond chatbot applications, reshaping sectors from logistics to healthcare.

What SEA’s digital economy needs beyond growth: An examination of the structural gaps — governance, trust, and inclusion — holding back Southeast Asia’s digital economy despite strong headline expansion numbers.

GenAI tops SEA finance professionals’ skills wishlist: A new ACCA survey finds generative AI is the top skill Southeast Asian finance professionals want to acquire, though access to quality training remains a persistent gap.

Skills built faster than workplaces redesigned to use them: A Singapore Polytechnic analysis finds that workers are upskilling in AI faster than organisations are restructuring roles and workflows to deploy those skills effectively.

Singapore is not a small market, it is a compressed one: A reframing of Singapore’s market dynamics, arguing that density, purchasing power, and institutional access make it uniquely valuable for startups testing high-value propositions.

---

INTERNATIONAL

Anthropic’s US$1.5B copyright settlement approved: A US court approved Anthropic’s landmark US$1.5B settlement with publishers over training data use, setting a significant precedent for how AI firms compensate content creators globally.

Netflix paid US$587M for Ben Affleck’s AI filmmaking startup: The streaming giant acquired Make It, Affleck’s AI-assisted film production company, in one of the largest deals yet at the intersection of Hollywood and generative AI.

CuspAI raises US$450M, hits US$2.6B valuation: Temasek-backed CuspAI closed a US$450M round and launched an AI-powered materials discovery network, positioning itself at the frontier of AI-driven scientific research.

OpenAI fears open-weight models, and it should: An analysis of why OpenAI’s internal anxiety over open-weight competitors reflects a broader strategic vulnerability for US AI dominance as capable open models proliferate globally.

Bitcoin reclaims key technical levels as Ethereum leads gains: Bitcoin recovered critical support levels while Ethereum outperformed the broader crypto market, driven by renewed institutional appetite and improving macro sentiment.

AI’s most important protocol gets easier to use: The Model Context Protocol (MCP), the standard enabling AI agents to interact with external tools, has been updated to reduce integration friction for developers building agentic applications.

---

CYBERSECURITY

GoSerpent exposes the quiet cyber war on SEA governments: A newly identified threat actor, GoSerpent, has been conducting sustained cyberespionage campaigns against Southeast Asian government agencies, using stealthy malware strains to exfiltrate sensitive data.

Your offshore vendor’s AI is running on your code: Enterprises using offshore development partners face growing IP and security risks as vendors adopt AI coding tools trained on proprietary client codebases without disclosure.

---

SEMICONDUCTOR

Powertech and Broadcom to build US$400M Singapore chip JV: Taiwan’s Powertech Technology will invest US$400M in a Singapore-based advanced chip packaging joint venture with Broadcom, reinforcing Singapore’s role in the global semiconductor supply chain.

Google develops new AI chip to boost Gemini efficiency: Google is building a dedicated chip designed to lower inference costs for its Gemini model family, intensifying the in-house silicon race among US AI hyperscalers.

---

AI

Singapore turns AI scrutiny to chatbots, personal data, and digital twins: Singapore’s Personal Data Protection Commission is intensifying oversight of AI deployments involving chatbots and digital twins, signalling stricter data governance expectations for AI developers.

Singapore strengthens privacy tools for AI adoption: The government has released updated privacy-enhancing technology guidelines to help enterprises adopt AI responsibly while complying with personal data protection obligations.

AI is about to blow a hole in ASEAN’s climate targets: Soaring AI energy consumption threatens to derail ASEAN’s net-zero commitments unless the region accelerates the transition to clean baseload power for data centres.

AI is answering customers before they click and may never mention you: Businesses risk being bypassed entirely as AI-powered search and answer engines resolve customer queries without surfacing brand websites or product pages.

Should AI investors diversify into Chinese tech stocks?: An analysis of whether allocating to Chinese AI firms offers a meaningful hedge against rising competition from Chinese models challenging US incumbents.

AI will not fix hiring if it only works for employers: A critique arguing that AI-driven recruitment tools entrench existing power imbalances by optimising for employer efficiency while stripping candidates of transparency and recourse.

---

THOUGHT LEADERSHIP

Why boards are getting CEO succession wrong: Poor succession planning is costing companies far more than most boards acknowledge, with gaps in structured process and internal talent development creating avoidable leadership crises.

Global expansion is now about reducing trust costs, not information costs: The real barrier to cross-border growth for startups is no longer market data access but building credibility and institutional trust in unfamiliar ecosystems.

It’s time to rebuild the enterprise house with AI: Enterprises must rearchitect their core operations around AI rather than layering tools onto legacy structures, argues this analysis of AI-driven business transformation.

Finance doesn’t have a math problem, it has an ego problem: The financial services industry’s resistance to AI adoption stems less from technical barriers than from cultural rigidity and leadership identity tied to conventional expertise.

Why DNA is becoming the next platform layer in digital health: Genomic data is emerging as foundational infrastructure for personalised healthcare, enabling AI-driven diagnostics and treatment models that move beyond population-level averages.

SEA’s sustainability opportunity is infrastructure: The path to sustainable growth in Southeast Asia runs through physical and digital infrastructure, not just policy commitments, requiring coordinated capital deployment at scale.

Two years to agentic: comply or grow: Organisations have a two-year window to prepare for agentic AI deployment before it reshapes competitive dynamics and those that only focus on compliance will be outpaced by those that build for growth.

The extinction events in product evolution: A framework examining which product categories face existential pressure from AI-native competitors and what survival looks like for incumbents unwilling to cannibalise themselves.

From tools to partners: the Socratic dream of two butterflies: A philosophical exploration of AI as a thinking partner rather than a productivity instrument, drawing on classical dialogue and Eastern metaphor to reframe human-AI collaboration.

The job you’re studying for might not exist: An urgent challenge to education systems and students to rethink career preparation in an era when AI is eliminating roles faster than institutions can update curricula.

The psychology of believing AI: certainty vs truth: Explores how humans are cognitively primed to over-trust confident AI outputs, and why distinguishing between an AI’s expressed certainty and actual accuracy is a critical skill for practitioners.

The post Ecosystem Roundup: SEA’s AI future is being coded in its own languages appeared first on e27.

Anthropic may look stronger than OpenAI on the usual pre-IPO scoreboard.

It has reported stronger private-market momentum. It appears closer to near-term operating profit. Its gross margin is reported above OpenAI’s. It has deep enterprise relationships, a strong reputation with developers, and Claude Code has become one of the clearest examples of an AI product that customers already pay for at scale.

On the surface, that looks like the cleaner business.

But the more useful question is not which company looks better today. It is which company gets stronger as AI does what everyone expects it to do: improve and get cheaper.

On that question, OpenAI and Anthropic are not the same kind of company.

Anthropic mainly sells access to frontier intelligence. OpenAI sells that too, but it also controls a mass consumer interface used by hundreds of millions of people. That difference matters because falling AI costs do not affect both businesses in the same way. For a frontier model seller, cheaper AI erodes the price of the thing being sold. For a consumer platform, cheaper AI lowers the cost of serving users whose attention can be monetised through advertising, commerce, subscriptions, and referrals.

The same cost curve can damage one business model and strengthen another.

That is the central divergence.

The price of AI work is falling extremely fast. Depending on the benchmark and the task, equivalent-quality AI has been getting cheaper by orders of magnitude. One widely cited estimate puts constant-capability inference cost decline at roughly tenfold per year. Other measurements show even sharper falls for some tasks. Work that once cost tens of dollars per million tokens has moved toward cents.

This is not a normal software pricing cycle. It is the economics of a manufactured input. As models improve, hardware scales, inference systems are optimised, and competition intensifies, the unit cost of producing cognitive work falls. The unusual part is that the manufactured product is not a phone, chip, battery, or solar panel. It is intelligence delivered through computation.

That distinction matters for valuation.

If a company sells a scarce software product with durable pricing power, investors can imagine software margins. But if a company sells a manufactured input on a steep cost curve, the better analogy is not classic enterprise software. It is a commodity producer with a premium tier on top. The premium may be valuable, but it is constantly under attack from the next cheaper substitute.

This is the problem Anthropic has to solve.

Also Read: Your offshore vendor’s AI is running on your code: Do you know which one?

At the moment, frontier models are not good enough for many complete workflows. They still make errors. They need supervision. They lose context. They often complete pieces of work rather than whole jobs. Because of that, each capability improvement increases demand for the newest model. Buyers want the best system because the current one is still not quite enough.

That creates a temporary premium market.

But the demand for capability is bounded by the task. Once a model can perform a defined job to a competent standard, a better model adds less value. If an AI system can take a set of accounting records, identify what matters, apply the rules, produce the filing, flag the judgment calls, and explain the output, then the buyer no longer needs the newest frontier model for that task. The job is done.

At that point, the buyer has a different question: what is the cheapest model that clears the bar?

That is the good-enough threshold. Once a task crosses it, the task leaves the premium market. It falls into the commodity market, where open-weight models, older frontier models, and cheaper specialist systems compete for the work. The frontier model may still be better in a general sense, but better no longer matters enough to command a large price premium for that specific job.

This is the structural trap for a business built around selling frontier access.

Improving the model conquers more tasks. But conquering a task means that task eventually stops needing the frontier. Over time, the frontier-only market does not automatically expand. It may narrow unless new categories of work open faster than old ones commoditise.

Anthropic is not blind to this. Claude Code is important because it moves the company higher up the stack. It is not merely selling tokens. It is selling a specific outcome inside a valuable workflow. That is the right direction. Application-layer products are less exposed than raw model access because customers are paying for the completed job, integration, and workflow value rather than just the intelligence underneath.

But the valuation question remains. How much of Anthropic’s future value comes from durable application products, and how much still depends on frontier access retaining premium pricing?

OpenAI has a different problem and a different opportunity.

It also sells model access. It also faces inference costs. It also competes in the frontier race. But it owns something Anthropic does not: a consumer destination at enormous scale.

That changes the effect of falling AI costs.

Also Read: AI is answering your customers before they ever click, and it may never mention you

For OpenAI, cheaper inference reduces the cost of serving free and low-paying users. If those users can be monetised through advertising, commerce, referrals, subscriptions, enterprise conversion, or in-chat purchasing, then falling AI costs widen the gap between the cost of serving attention and the value of monetising it.

That is not the economics of a pure model vendor. It is the economics of a platform.

This is why OpenAI’s experiments with advertising and in-chat commerce matter. They are not just incremental monetisation ideas. They are attempts to shift the company away from selling tokens and toward monetising the interface where users already spend time, search for answers, compare products, and make decisions.

If ChatGPT becomes a meaningful consumer gateway, then cheaper AI helps OpenAI twice. It lowers the cost of each interaction, and it increases the number of interactions that can be economically served. More usage is no longer only a cost burden. It becomes monetisable surface area.

That is the flywheel OpenAI is trying to build.

The contrast is simple. A tenfold annual fall in inference cost lowers the price of the thing Anthropic mainly sells. The same tenfold fall lowers the cost of the thing OpenAI can give away to attract and monetise users.

One business sells the deflating input. The other may use the deflating input to build a larger platform.

That does not make OpenAI’s outcome guaranteed. Advertising inside an AI assistant could damage user trust. Commerce may not convert at scale. Referral economics may be weaker than expected. Regulators may limit parts of the model. Users may resist a shift from neutral assistant to monetised shopping interface. The consumer flywheel is still a thesis, not a proven revenue engine.

OpenAI also carries its own financial pressure. Serving hundreds of millions of users is expensive, even when unit costs are falling. Infrastructure commitments are large. The company still has to prove that mass usage converts into durable economics rather than just enormous demand for subsidised computation.

But structurally, OpenAI has more ways to benefit from AI deflation.

Also Read: Beyond the chatbot: How Gen Z pioneers are leading ASEAN’s new AI revolution

Anthropic’s strongest counterargument is that frontier capability may stay scarce. If the best models remain meaningfully better, if enterprise customers are deeply locked into workflows, if safety, reliability, compliance, and data integration matter more than raw token price, then premium pricing can survive longer than the simple commodity story suggests. In that version of the market, Anthropic’s enterprise concentration is not a weakness. It is evidence of pricing power.

The second counterargument is that new frontier-only work may open faster than existing work commoditises. If each generation of models enables qualitatively new tasks — autonomous agents, longer-horizon reasoning, richer multimodal work, or entirely new software workflows — then demand for the newest model may keep expanding.

That is the key uncertainty.

The question is whether frontier intelligence remains a scarce product, or whether it becomes a rapidly cheapening input.

If it remains scarce, Anthropic’s model is stronger than the deflation argument implies. If it becomes a cheap input, value moves elsewhere: to compute and energy capacity, to distribution, to consumer interfaces, and to applications that convert cheap intelligence into specific outcomes.

That is why the IPO filings matter. The headline valuation will get the attention. The more important signals will be beneath it.

The first signal is revenue quality. Is reported run-rate revenue gross or net of reseller and partner pass-through? A large gap between gross and net would make the top line look stronger than the underlying economics.

The second signal is gross margin. If margins rise sharply while inference prices keep falling, that supports the view that frontier labs can retain pricing power. If margins remain compressed, the commodity interpretation gains strength.

The third signal is revenue mix. How much comes from raw model access, and how much comes from higher-stack products? For Anthropic, Claude Code and similar workflow products matter because they reduce dependence on frontier access alone. For OpenAI, advertising, commerce, subscriptions, and platform monetisation matter because they show whether consumer distribution can become a real economic engine.

The fourth signal is customer behaviour. If enterprises keep paying for the newest model even after cheaper alternatives become good enough for many tasks, lock-in is stronger than expected. If customers shift workloads aggressively to lower-cost models once capability thresholds are crossed, the frontier premium decays.

The useful conclusion is not that OpenAI is certain to beat Anthropic.

It is that the normal scoreboard may be measuring the wrong thing. Revenue growth, private valuation, filing sequence, and near-term profitability describe the present. They do not answer the more important question: what happens as the product gets cheaper?

Anthropic may be ahead on today’s visible metrics. But if most of its value remains tied to selling frontier access, then it is exposed to the falling price of its own output. OpenAI may look less clean financially today, but if it turns cheaper AI into cheaper user acquisition, cheaper user service, and more monetisable attention, then the same deflation becomes an advantage.

The AI market is usually described as a race to build the best model.

That may be the wrong race to watch.

The durable value may not sit with whoever produces the frontier model at any given moment. It may sit with whoever owns the interface, the workflow, the distribution, and the customer relationship once intelligence itself becomes cheap.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post The real difference between OpenAI and Anthropic is what happens when AI gets cheaper appeared first on e27.