Rahim Amir

Researchers at the University of Massachusetts Amherst demonstrated that certain expired Visa contactless cards can potentially be used for real purchases by altering the expiration date with an intermediary setup involving two smartphones. The vulnerability arises because the expiration date read by the terminal is not cryptographically bound to the value checked during online authorization, while related card certificates may remain valid beyond the printed expiration date. Testing found that Mastercard, Discover, and American Express cards rejected the altered dates. Although EMV’s optional Relay Resistance Protocol could detect and block the attack, it was not enabled on any tested card or terminal. Visa and relevant banks were notified in 2025, but no confirmed fix had been announced at publication; consumers are advised to destroy the chip and cut through card numbers when disposing of expired cards.

Infoblox Threat Intel reports that roughly 65,000 expired domains are re-registered each day, allowing buyers to exploit established domain authority, inbound links and search visibility. The company identified Sable Squirrel, an operation controlling more than 10,000 domains and estimated to have spent over $7 million acquiring expired names, primarily for a Vietnamese-language sports-piracy and gambling network. Infoblox linked 405 of those domains to malware command-and-control infrastructure supporting threats including Quasar RAT, AsyncRAT, DCRat, NanoCore, Remcos and njRAT. Vietnamese authorities froze several sites, charged 30 suspects and seized assets, but the operation reportedly survived and continued expanding.

Michael Dell posted a ten-second, AI-generated video on X that recreates Dell's famous early-2000s advertising slogan, replacing the desktop computer with a crane-delivered AI server rack. Although there is no evidence that Dell plans to revive the campaign, the clip reflects the company's shift from consumer PCs toward data-center infrastructure. In its latest reported quarter, AI-optimized server revenue rose 757% to $16.1 billion, surpassing the PC division's $14.6 billion in revenue. The use of a synthetic Ben Curtis lookalike, despite Curtis being available as a Dell ambassador, adds an ironic AI-related element to the stunt.

Nine PBS is suing over access to roughly 50 terabytes of archival material stored in an Iron Mountain facility after its cloud provider, Open Source Storage, went out of business and cut off access when its contract expired. Iron Mountain says it only rented physical infrastructure to OSS and did not control the data or the client-owned hardware. Courts have ruled that Nine PBS owns the material and ordered cooperation with its retrieval within 30 days, provided the station identifies a qualified third party and pays outstanding storage fees. The archive includes decades of programming and coverage of major events including the history of East St. Louis, the COVID-19 pandemic, and the 1993 Great Flood.

Alibaba Cloud has begun commercially offering its Lingjun Zhenwu M890 supernode, a rentable 64-card AI computing unit built with Alibaba-designed silicon. The system uses T-Head’s 144GB Zhenwu M890 processors and ICNSwitch 1.0 interconnect technology, reportedly supporting models of up to 10 trillion parameters and already powering Kimi K3 and Qwen3.8-Max. Availability is currently restricted to customers in Ulanqab, Inner Mongolia, and pricing, power consumption, and independent benchmarks have not been disclosed. The launch reflects China’s push for domestic AI infrastructure as access to leading Nvidia and AMD chips becomes more constrained, while Alibaba plans substantially more powerful Zhenwu systems for 2027 and 2028.

Nvidia has released NeMo Switchyard, an open-source router that selects the least expensive language model capable of handling each request. Nvidia claims the approach can reduce costs by 74% compared with using a frontier model exclusively, with a 6% reduction in accuracy. However, routing adds judge-model expenses, approximately 700 milliseconds of latency, and substantial variability in per-task costs, making it less suitable for short or latency-sensitive workloads. Switchyard enters a growing market that includes RouteLLM, LiteLLM, OpenRouter, and internal efforts by major companies. The tool may also encourage enterprises to run smaller open models on Nvidia hardware, aligning cost savings with Nvidia’s accelerator business.

Researchers from the University of Birmingham and Fuzzware found that the Proactive SIM feature’s RUN AT command was exposed on nine of 26 tested devices, including six of eight cellular modules used in IoT equipment. They used SIM-issued commands to achieve code execution on a commercial Autel EV charger, showing that vulnerable chargers, industrial routers and vehicle telematics systems could potentially be taken over. The attack requires an attacker to control and physically access the SIM, limiting its usefulness against smartphones, but unattended machine-to-machine equipment remains more exposed. Qualcomm has developed a hardened configuration that disables the interface by default, although vendors had not yet published public advisories.

Schenker has opened worldwide orders for the 16-inch Element 16, priced from €1,499 and expected to ship in late August 2026. Its headline feature is a potential 72GB of DDR5 RAM, achieved through 8GB of soldered memory plus a 64GB SODIMM, although that maximum configuration is not currently available and 40GB modules are the practical option. The laptop uses Intel Panther Lake processors, integrated graphics, a 120Hz 1600p display, PCIe 5.0 storage, Wi-Fi 7, and a 72Wh battery in a 1.9kg chassis. A screwless bottom panel and Schenker's commitment to a future drop-in mainboard make it a potential Framework rival, but the long-term cost and viability of the upgrade path remain uncertain.

GlobalSign revoked thousands of TLS certificates serving Russian domains after certificate-authority rules began requiring screening against U.S. and European sanctions lists. The revocations reportedly affected banks, state services, email systems and internal infrastructure, leaving some sites dependent on manually installed Russian trust roots that mainstream browsers do not recognize. Russian organizations have sought replacement certificates from HARICA and TrustAsia, while the government promotes its own root certificate and Yandex Browser. Ukrainian officials and security researchers warn that the transition could increase exposure to outages, interception and cyberattacks, with infrastructure reconfiguration potentially costing large companies millions of rubles and taking months.

OVHcloud plans to raise dedicated-server prices from September 2026, with its latest-generation Game servers increasing by as much as 87%, current-generation hardware by 51%, and older 2024-generation machines by about 28%. CEO Octave Klaba attributed the increases to an AI-driven surge in memory and storage costs, saying memory prices have risen sixfold and could potentially double again by early 2027. The increases could significantly affect private and commercially operated gaming servers. OVHcloud’s vertically integrated manufacturing model has not insulated it from component-cost pressures, and industry observers expect AWS and Microsoft Azure to face similar pressures, though their larger supply agreements and balance sheets may make any increases slower and less transparent.

The U.S. Department of War has expanded a restricted counter-drone marketplace operated by Joint Interagency Task Force 401 and built by Kaizen Laboratories under a $15 million prototype contract. Approved military, government, allied, and defense buyers can access vetted products from vendors including Anduril, DroneShield, AeroVironment, and SMARTSHOOTER, with access limited according to classification and export-control status. Marketplace purchases rose from $13 million in April to $21 million by early August, while the platform remains in a controlled rollout. A planned AI-assisted layer will recommend counter-drone configurations based on threat profiles, budgets, readiness, and operating environments, although the marketplace remains inaccessible to the general public and its broader effectiveness is still unproven.

Linux kernel release candidate 7.2-rc7 remained unusually large late in the development cycle, following an equally substantial rc6. Linus Torvalds attributed much of the increased activity to AI-powered code-review and analysis tools, which identify bugs while human developers write and submit the fixes. More than 400 fixes from over 230 contributors affected drivers, filesystems, networking and architecture code. Although Torvalds said the release did not appear especially risky, the sustained flow of AI-assisted findings could challenge the Linux kernel’s traditional maintainer review process and make oversized updates the new normal.

Russia's MyOffice office-software business, developed by New Cloud Technologies and majority-owned by Kaspersky Lab, has reportedly closed development centers in St. Petersburg and Innopolis, transferred remaining staff to Moscow and laid off employees. Restated 2025 accounts increased its net loss to 8.82 billion rubles, or about $107 million, while revenue fell to roughly 1 billion rubles. Several products had their development suspended, and Kaspersky CEO Eugene Kaspersky said the business had entered “slow-burn mode.” MyOffice's domestic market share also declined from 5% to 3% despite Microsoft's withdrawal from Russia and government mandates favoring domestic software.

Democratic Congressman Ro Khanna has introduced H. Res. 1471, a nonbinding proposal that would establish a 2,500-foot buffer between data centers and homes, schools, childcare facilities, hospitals, and nursing homes. It would also protect local veto power, require disclosures on electricity and water use and community impacts, and allow subsidies to be clawed back when conditions are not met. Although the resolution has no legal force and is unlikely to advance in the current Republican-controlled Congress, it offers a blueprint for cities and states confronting rapid AI infrastructure expansion. Critics may view the buffer as excessively broad, while stronger proposals from Bernie Sanders and Alexandria Ocasio-Cortez seek a moratorium on large AI data centers. State-level actions in Oregon, Ohio, North Carolina, and Virginia indicate that regulation is advancing even as political divisions and local opposition grow.

Nashville’s Metro Council voted 27–5 to authorize acquiring the 23.49-acre site where DC BLOX plans to build a $700 million data center beside the Nashville Zoo. The measure permits negotiation or condemnation but does not yet commit the city to purchase the property; an appraisal, a three-week waiting period and possible court approval are required before condemnation could proceed. Zoo officials say continuous tonal noise and vibration could disturb nearby clouded leopards and that industrial runoff could worsen existing stormwater problems. The property was purchased by DC BLOX for $23 million shortly before the vote, potentially increasing the city’s financial exposure. A data-center moratorium, new regulations and pending zoning challenges could also prevent or delay construction, making the case a test of whether a city can afford to use eminent domain against a private data-center project.

Scammers are disguising Lumma Stealer malware as pirated downloads of The Odyssey, using movie-release-style filenames and VLC icons to hide executable files. Bitdefender says the malware steals browser passwords, authentication cookies, payment data, cryptocurrency wallets, autofill information and remote-desktop credentials. The campaign appears less technically sophisticated than earlier samples because it lacks droppers and persistence, but stolen session cookies can still let attackers bypass renewed multi-factor authentication. Users are advised to avoid unofficial movie downloads and enable file-extension visibility in Windows Explorer.

TSMC is reportedly holding about $1 billion worth of finished A20 Pro wafers because the LPDDR5X DRAM required for wafer-level packaging has not arrived. Apple’s shift from conventional InFO packaging to WMCM was intended to improve memory capacity and latency for on-device AI, but it prevents finished processor dies from being stockpiled separately from memory. Apple and its assembly partners reportedly expect to meet initial iPhone 18 demand without delaying launch, though constrained processor deliveries could produce thin inventory, rapid preorder sellouts and extended shipping estimates. Memory shortages are linked to demand from AI data centers, while efforts to broaden Apple’s supplier base have reportedly failed.

A paid Google ad led a Mac user to a fraudulent Claude Code installation guide hosted on Anthropic’s legitimate claude.ai share infrastructure and falsely attributed to “Apple Support.” Following its Terminal command installed MacSync, a six-stage macOS malware chain involving information theft, remote access, permission abuse and Trojanized Ledger and Trezor applications. The tampered wallet apps displayed fake recovery prompts that harvested seed phrases and enabled cryptocurrency theft. Huntress says the campaign exploited trusted domains and platform features rather than relying on obvious impersonation, echoing earlier attacks using poisoned AI conversations and fake installers.

Nvidia argues that AI performance will increasingly depend on storage efficiency, particularly as inference workloads generate huge volumes of sub-kilobyte reads for embeddings and KV caches. It has open-sourced its cuFile APIs and underlying storage stack and launched Storage-Next with more than 40 vendors, including Intel, Google and Meta. Its SCADA framework moves the storage control path from the CPU to the GPU, complementing cuFile’s direct data transfers and targeting the latency and read amplification limits of conventional 4KB-tuned enterprise SSDs. The initiative aims to establish GPU-initiated storage as an industry standard, with partner systems expected in 2026 and next-generation SSDs targeting 100 million IOPS.

Developer slvDev ran a 28.9-million-parameter TinyStories-style language model fully offline on a $10 ESP32-S3 microcontroller at 9.88 tokens per second. Four-bit quantization and per-layer embeddings, inspired by Google’s Gemma 3n, place most parameters in flash storage while keeping the actively used model components in RAM. The resulting system produces short, mostly coherent stories but cannot answer questions, follow instructions, write code, or retrieve facts, demonstrating an impressive memory-management technique with limited practical capability in its current form.

China has revised its integrated-circuit layout-design regulations for the first time since 2001, removing the phrase “semiconductor integrated circuit” from the definitions and extending protection to photonic and quantum chip layouts from October 2026. The change is less evidence of current manufacturing superiority than a signal that China wants its intellectual-property system to cover future chip technologies. The broader revision strengthens damages provisions, tightens registration requirements, and allows third parties to seek revocation of registrations. China is advancing in photonic research and has companies such as Huawei, Alibaba, Cambricon, and Loongson producing competitive components, but the United States and Taiwan still control key parts of the supply chain needed to turn research breakthroughs into commercial products.

SK hynix and SanDisk have published the first High Bandwidth Flash specification through the Open Compute Project. HBF supports stacks of up to 512GB, far exceeding HBM4 capacity, with bandwidth grades ranging from about 0.4TB/s to 3.0TB/s and UCIe connectivity for use with processors from different vendors. Google and Tenstorrent have joined the consortium, but Nvidia and other major memory makers remain absent. Memory samples are planned for the second half of 2026 and early AI inference-device samples for 2027, while broad commercialization is not expected until late 2027 or 2028, with consumer availability likely much later.

HP, Asus, and Acer have qualified and begun shipping small volumes of CXMT DRAM in selected notebooks sold outside the United States. The move is driven less by US policy than by the severe memory shortage and the need to maintain supply-chain alternatives while Samsung, SK Hynix, and Micron control more than 90% of the DRAM market. CXMT currently offers neither lower prices nor dependable long-term availability, but the qualification process gives the PC makers an option if supply conditions or geopolitical restrictions change. AI infrastructure demand has redirected wafer capacity toward high-bandwidth memory, driving sharp DRAM price increases, higher PC prices, and weaker shipment forecasts. Although the incumbent suppliers face little immediate commercial threat, CXMT’s growing capacity and completed certifications could make it a more significant competitor once supply constraints ease.

Michael Dell reposted PC's Limited's July 31, 1984 quarterly financial statement, which showed approximately $1 million in sales and $134,762.75 in net income from $1,000 of starting capital. The 19-year-old used the document to convince his parents to let him leave the University of Texas and pursue the business full time. The article argues that his decision was supported by concrete financial evidence rather than being a conventional leap of faith, ultimately leading to Dell's rise into a multibillion-dollar technology company now benefiting from demand for AI infrastructure.

A Reuters investigation reviewed by TechRadar found that Chinese military-linked researchers have reportedly used outputs from OpenAI and Anthropic models to train smaller domestic AI systems for defense applications. The technique, known as model distillation, transfers capabilities and reasoning patterns from large models to systems that can run locally with less computing power, potentially bypassing U.S. controls focused on advanced chips. The findings highlight an asymmetry in which model-generated text can cross borders without being regulated like hardware. The report also notes the apparent contradiction between China’s restrictions on Western AI and its military researchers’ use of American model outputs, while concluding that Washington currently lacks a practical way to control this form of technology transfer.

LinkedIn plans to keep its GPU investment, compute capacity and storage footprint roughly flat over the next fiscal year rather than expanding its AI data centers. The company says it has approximately doubled the efficiency of its existing GPUs in six months through better utilization, workload allocation and model distillation, allowing it to deploy more compute-intensive features without buying substantial new hardware. This contrasts sharply with Microsoft’s broader AI spending and the massive capital expenditure planned by major US hyperscalers. LinkedIn attributes its flexibility partly to operating its own data centers after abandoning a migration to Azure, while acknowledging that the efficiency drive was prompted by rising query costs and rapidly growing stored data. The strategy could challenge the assumption that AI product expansion requires proportional hardware growth, but its success will depend on whether the efficiency gains can support LinkedIn’s future product demands.

Samsung expects the global memory shortage to persist through 2028, with conditions becoming more severe in 2027 as AI infrastructure demand outpaces the industry's ability to add capacity. New fabs require more than three years to become productive, delaying the impact of current investment. Samsung reported record quarterly revenue and operating profit as DRAM and NAND prices surged, and plans to place 60–70% of its capacity under five-year supply agreements with price floors and advance payments. The shortage is expected to raise PC prices and reduce shipments, while companies such as Micron increasingly prioritize enterprise AI customers over consumer products.