Models & Labs · Asia-Pacific

The gap between open and closed AI models in cyber capabilities keeps shrinking

Credited to The Decoder · the-decoder.com · 13h ago

AI agents can break into realistic corporate networks on their own, Vals AI finds

The Vals CyberOffense benchmark simulates realistic corporate networks to test this. Each agent operates solo on an isolated attack machine, starting with nothing but a public web address and zero human help.

GPT-6 Sol, part of OpenAI's restricted Daybreak cyber program, posted the best result at 80.8 percent of objectives. Deepseek V4.1 Flash 1017, a freely available open-weight model, isn't far behind at 72.8 percent and costs only about $3 per run instead of roughly $22. Once a model's weights are public, no provider can control or restrict how it's used, Vals researchers say. Ad

In the harder Lighthouse test, Daybreak took over admin control of the first company, stole passwords, and used them to break into the second company. It completed 27 of 31 objectives in under three hours. Deepseek managed 21 objectives at a tenth of the cost and wasn't limited solely by safety guardrails. Ad

Longer runtimes turned out to be critical as well. Deepseek scored zero results in its first 25 million tokens and made most of its progress only after 100 million tokens. Short tests therefore underestimate the real potential of open models.

Vals points out that the same findings create an opportunity for defenders, because the very models that exploit vulnerabilities can help companies find and fix them first. Attacks on real networks are likely to increase in the coming months, according to Vals. Ad

Open models close the gap fast, earlier studies confirm

The results align with a recent analysis by the UK's AI Security Institute (AISI) from July. The gap between open and closed frontier models in cyber capabilities has shrunk to four to seven months. For most of 2025, that gap still stood at six to ten months.

Open models are also much cheaper. A cyber-range run with Deepseek V4-Pro costs just $1.19 according to AISI, compared with $85 for Anthropic's Opus 4.5. The safety measures on the open models proved largely ineffective. Ad

A recent study by Anthropic also showed that the freely available model GLM-5-3 can build exploits at nearly the same level as the internal top model Mythos. Ad

AI News Without the Hype – Curated by Humans

Chips & SupplyAsia-Pacific

Samsung reportedly sells 300 chipmaking tools to DB Group

Credited to DigiTimes$ · digitimes.com16h ago

Useful

Samsung Electronics has reportedly agreed to sell approximately 300 semiconductor manufacturing tools from its shuttered 8-inch S7 foundry in Giheung to South Korea's DB Group, as DB HiTek faces tight production capacity and Samsung prepares to withdraw from the 8-inch foundry…

View the article · $DigiTimes

Front

Power & Sites · Asia-Pacific

The datacentre arms race hits pause in south-east Asia

Credited to The Guardian · theguardian.com · 18h ago

Power and water constraints are pressuring what seemed like a juggernaut of growth for the region just months ago

I n early 2026, Thailand’s datacentre industry was on a roll. The country’s board of investment had recently announced approval of datacentre projects from Tiktok, a UAE-based Skyline datacentre and Singapore’s Bridge datacentres, totalling a planned investment of $27bn.

As the year has passed, however, what once seemed like a new industrial juggernaut has ground to a halt. Thailand paused construction on 49 projects and approval decisions on 117 more in September 2026, amidst growing concern about the lack of sufficient details for regulation and monitoring. It is not alone in reconsidering datacentre projects. In neighbouring Indonesia, the governor of the country’s most populous province of West Java halted construction on a datacentre due to lack of permits and an incomplete environmental assessment process.

With the global rise of AI usage driving an increased demand for datacentres, south-east Asia has emerged as a surprise centre of growth in the Asia Pacific region. While China has the largest operational datacentre capacity in the region, a 2026 report by real estate firm Cushman and Wakefield shows that south-east Asia has the largest share of under-construction datacentre capacity in Asia Pacific, about 50% of the total. Malaysia and Thailand comprise a large portion of that capacity.

Through the 2010s, Singapore was south-east Asia’s leading datacentre hub, with companies like Alibaba and Facebook opening facilities there. However, the city-state implemented a moratorium on new datacentres from 2019 to 2022 because they were “intensive users of water and electricity”, as described in a press release from the Ministry of Energy, Trade and Industry. Since then, approval has been selective amidst land constraints and sustainability requirements, with just 95MW of operational capacity added between 2023 and 2026, per Cushman and Wakefield.

With the advent of LLMs and rise in usage of AI, associated datacentre demand has surged in the last half decade, and desperate companies looked beyond Singapore to build datacentres. Malaysia has been a major beneficiary, with operational capacity in the southern state of Johor growing more than tenfold from 94MW in 2023 to 1110MW in 2026, with 3088MW additional under construction or planned. Johor has become the largest datacentre hub in south-east Asia, attracting $40bn of datacentre investment as of 2026, according to Menteri Besar Onn Hafiz Ghazi, the head of the state government. Key operators include Microsoft, which announced a $2.2bn investment for cloud and AI infrastructure across Kuala Lumpur and Johor in 2024, and Australian datacentre company AirTrunk, which said it plans to invest $3bn to double its datacentres in Johor from two to four.

Other countries in south-east Asia are catching up. Indonesia has a datacentre development pipeline of 1699MW in greater Jakarta, and is developing the island of Batam as a secondary hub. Microsoft is building five datacentres as part of a$1.7bn investment in the country, and Singapore-based datacentre operators Princeton Digital Group and Digital Edge are also expanding. In Thailand, Bangkok’s development pipeline stands at 2084MW, while secondary markets like Manila and Ho Chi Minh City are ramping up operations, based on Cushman and Wakefield’s analysis.

While the rush to build more datacentres is attracting big investments across the region, it is putting a strain on each country’s resources. Energy consultancy Wood Mackenzie estimates that datacentres in Johor already account for 24% of the state’s total power consumption today, which could grow to 40% by 2035.

“The rapid growth of datacentre developments in Johor has the potential to create localized grid connection bottlenecks and increase the need for transmission network upgrades as well as new power supply capacity,” says Alvin Tan, research analyst at Wood Mackenzie.

Thailand’s energy minister, Akanat Promphan, has said that datacentres require large amounts of power generated from expensive imported gas, and called for a new electricity user category with higher tariffs to reflect the demand.

Datacentres’ water-guzzling tendencies are also a major cause of concern for officials in the region. Johor is prone to droughts, and the current operational datacentres are estimated to need about 675m cubic meters of water per day. In Indonesia, a report by research firm Tenggara Strategics shows that many datacentres are water-cooled and situated in hot and humid lowland environments, resulting in elevated water consumption in comparison to facilities in more favourable temperatures.

In Thailand, datacentre projects in the country’s Eastern Economic Corridor have residents concerned about the pressure on their already depleted water reservoirs.

“The Khlong Luang Reservoir in Chonburi (in eastern Thailand) is a primary flashpoint,” says Somnuck Jongmeewasin, research director of research and advocacy group EEC Watch. “Despite the reservoir being shallow and under strain, authorities are piping its water to industrial estates,” he adds.

A weekly dive in to how technology is shaping our lives

Local communities are waking up to the changes. In February, Johor saw its first public protest against datacentre construction, with demonstrators picketing a project by China’s ZData Technologies. Protesters expressed concern about the dust pollution caused by construction of the site near a residential location and the potential impact on water supply.

“While datacentres may create substantial investment and construction activity, they generally generate fewer long-term jobs than many other industries requiring comparable amounts of land, electricity and water,” says Jie-Sheng Tan-Soo, an assistant professor at Singapore’s Lee Kuan Yew School of Public Policy. “Communities may therefore reasonably ask whether they are receiving a proportionate share of the benefits while bearing the environmental and infrastructure costs.”

South-east Asia’s datacentre operators claim they are working toward making their facilities more resource efficient. Bridge datacentres, among Malaysia’s top datacentre operators, built a water reclamation plant and pipe network in Johor to cool its upcoming datacentre, expected to come online next year, while AirTrunk, among the largest datacentre platforms in Asia-Pacific, announced a partnership with sewage treatment operator Johor Special Water in April to recycle unused wastewater for the operational needs of its data centres. An AirTrunk representative said that its “campuses are designed to operate significantly more efficiently than conventional datacentres and incorporate advanced technologies to optimise energy performance,” but did not respond to detailed queries regarding data on efficiency and energy optimization.

Sustainability challenges remain. Microsoft has said that it is looking at south-east Asian clean energy projects to fuel its datacentres, but its published data on Power Usage Effectiveness (PUE), an industry metric measuring the energy efficiency of a datacentre, shows that its Asia-Pacific facilities are the least energy efficient globally.

Resource pressures have forced countries to become more circumspect about approving new datacentres. After Singapore’s moratorium ended, approval for new datacentres in the country has been strictly tied to a number of factors including power efficiency and carbon emissions. Malaysia has established a datacentre taskforce to review new projects with an eye on energy and water security, and Johor announced that it would no longer approve Tier 1 and Tier 2 datacentres, which used up to 50m litres of water per day, equivalent to the daily drinking water needs of 25 million people. Thailand announced an approval framework for new datacentre projects that considers energy and water use, along with power guarantee requirements and higher tariffs for datacentre operators.

To promote resource efficiency by datacentres, Tan-Soo advocates for integrated policy frameworks from national governments that have datacentre operators face the full social opportunity cost of the resources they consume.

“South-east Asia is increasingly aware but only partially prepared,” he says, and preparedness for environmental impacts needs to keep pace with the speed of investment.