Alvaro Lopez Ortega / 2026-06-23 Briefing

Created Tue, 23 Jun 2026 14:10:01 +0000 Modified Tue, 23 Jun 2026 14:12:07 +0000
3521 Words

A Russian hacker used Claude-integrated AI to breach hotel platforms, exposing millions of guest records. Cybersecurity threats rose with the discovery of the ‘PixelSmash’ FFmpeg flaw, allowing remote code execution via video. Meanwhile, Microsoft reported North Korean hackers targeting AI software supply chains. Finally, Europe’s Parliament backed a digital euro to reduce reliance on U.S. payment networks.

🤖 Artificial Intelligence

ByteDance unveils Seedance 2.5, generating 30s AI videos from up to 50 reference materials.

Large companies are reducing their AI expenditures by switching from expensive flagship models to more affordable alternatives from providers like OpenAI and Anthropic. For instance, Ensemble Health Partners has successfully transitioned to a significantly cheaper OpenAI model to manage its large-scale AI spending.

Nvidia VP Kimberly Powell on how AI can ease doctor workloads and improve patient care

Nvidia’s VP of Healthcare, Kimberly Powell, argues that artificial intelligence can address global medical staff shortages and reduce clinicians’ administrative workloads. By automating tasks such as medical reporting and transcription, AI has the potential to improve patient care and accelerate advancements in drug discovery.

Real estate pros increasingly use AI virtual staging to create misleading property listings.

Real estate professionals are increasingly using AI-powered virtual staging tools to enhance property photos, which can lead to misleading online listings. While these tools help clients visualize potential upgrades more affordably, they often create significant discrepancies between digital images and the actual condition or size of an apartment.

The Low-Tech AI Of Elden Ring

Elden Ring utilizes a Pushdown Automaton system rather than a traditional Finite State Machine to manage NPC decision-making. This architecture employs a stack of “Goals” implemented via Havok Script, allowing characters to execute complex behaviors by pushing and popping sub-goals based on their success or failure.

Unlimited OCR: One-Shot Long-Horizon Parsing

Baidu Inc. has introduced Unlimited OCR Works, a model designed for one-shot long-horizon parsing of single images and multi-page documents such as PDFs. Building upon Deepseek-OCR, the project is available on platforms including GitHub, Hugging Face, and arXiv.

The Reversal Curse: LLMs trained on “A is B” fail to learn “B is A”

Research has identified a phenomenon called the “Reversal Curse,” where large language models fail to automatically generalize knowledge from statements of the form “A is B” to their reverse, “B is A.” This limitation persists across various model families and sizes, though models can deduce reversed relationships when the information is provided within the prompt’s context.

The Coming Loop

The article explores the shift toward “agentic engineering,” where autonomous loops manage coding agents instead of direct human prompting. The author argues that this hands-off approach can degrade software quality by producing overly complex, defensive, and unmanageable code.

Neural Particle Automata

Neural Particle Automata evolve neural cellular automata by replacing fixed grids with agentic particles that move freely in space. By following simple shared rules, these particles can form complex morphologies and exhibit emergent behaviors such as self-regeneration.

AI Built a Nuke and Still Lost

An experiment using Civilization VI as a simulator revealed that frontier AI models struggle with the complex, adaptive decision-making required for effective governance. While these models can achieve specific strategic goals, they fail to adjust their strategies when faced with unforeseen systemic changes and cascading consequences.

Ultralytics YOLO26: Unified Real-Time End-to-End Vision Models

Ultralytics YOLO26 is a new family of unified real-time vision models that utilizes an NMS-free dual-head design and advanced training strategies like MuSGD to enhance efficiency and small object detection. The framework supports multiple tasks, including detection, segmentation, and pose estimation, while significantly improving the accuracy-latency trade-off on the COCO dataset.

VibeThinker: 3B param model that beats Opus 4.5 on reasoning with novel SFT+GRPO

VibeThinker-3B is a 3-billion parameter language model that utilizes a “Spectrum-to-Signal” post-training paradigm to enhance verifiable reasoning. The model achieves performance on mathematical and coding benchmarks comparable to much larger flagship systems, supporting the hypothesis that complex reasoning capabilities can be effectively compressed into compact architectures.

An Introduction to YOLO26

Released in January 2026, YOLO26 is a multi-task model family supporting tasks such as object detection, instance segmentation, and pose estimation across five size variants. The architecture is optimized for edge deployment by removing Non-Maximum Suppression and Distribution Focal Loss to reduce latency and improve hardware compatibility.

💻 Software & Development

Microsoft confirms Recycle Bin bug in Windows 10 and 11: a patch is on the way

Microsoft has confirmed a bug in Windows 10 and 11 that causes cryptic internal codes to appear instead of original filenames in Recycle Bin deletion dialogs. The company is currently developing a patch to resolve the issue, which follows a recent security update that has also triggered reports of OneDrive glitches and system instability.

Record type inference for dummies

This introductory post provides a foundation in type theory by defining anonymous records across various programming languages, such as JavaScript and Python. It serves as a precursor to an advanced discussion on implementing effective type inference for these records within statically typed languages.

Implementing a Custom Query Language with Python and Apache Spark

This article demonstrates how to implement Entity History Query Language (EHQL), a custom tool designed for analysts to query vehicle maintenance data. The implementation utilizes Python and the Lark parsing library to transform EHQL commands into executable Apache Spark queries.

FEXPRs vs. vtable: how LispE interpreter works

LispE addresses the traditional trade-off between interpreters and compilers by acting as an interpreter for compiled C++ objects. By representing each language instruction as a subclass with a uniform eval method, the system creates an executable Abstract Syntax Tree that leverages C++’s full optimization capabilities.

Please keep code descriptions simple

The author advocates for concise commit messages and merge request descriptions that focus on the rationale behind changes rather than redundant details. To improve accessibility for neurodivergent reviewers, they also recommend using atomic commits and avoiding an over-reliance on LLM-generated text.

Matt’s Script Archive: The Scripts That Reshaped The Web

In the mid-1990s, Matt Wright’s “Matt’s Script Archive” provided popular but highly insecure web tools like WWWboard that prioritized ease of use over security. This historical example illustrates the long-standing tension between non-technical users seeking simple software solutions and developers advocating for robust coding practices.

Blobly

The application Blobly is currently experiencing connectivity issues. Users are being prompted to attempt to reconnect to the service.

The Traditional Vi

Originally developed by Bill Joy in 1976, the vi editor was released under a BSD-style license in 2002 following decades of proprietary restrictions. This specific port preserves the original lightweight functionality while adding support for international character sets and minor POSIX.2 enhancements.

The new HTTP QUERY method explained

The newly introduced HTTP QUERY method, defined in RFC 10008, allows for complex data filtering using a request body to overcome the length and encoding limitations of GET parameters. Designed to be safe, idempotent, and cacheable, this method provides a standard alternative to using POST for large-scale or nested queries.

Matrix and Quaternion FAQ

The Matrix and Quaternion FAQ is a technical resource detailing the mathematical concepts and programming implementations of matrices and quaternions. The document includes a comprehensive history of collaborative updates and contributions from numerous developers dating back to 1997.

A pure ARM64 Assembly web server, now on Linux with CGI for no reason

The ARM64 Assembly web server ymawky has been ported from macOS to Linux and now includes CGI scripting support. This update enables the processing of dynamic content, query strings, and POST requests for CGI resources.

In praise of memcached

The author argues that Memcached is a superior choice for caching because its architecture prevents it from being mistakenly used as a permanent database. Unlike Redis, which can create operational risks due to its persistence features, Memcached simplifies cluster management and handling server downtime.

Microsoft Access finally breaks free of its 22-inch form limit

Microsoft Access has removed its 34-year-old 22-inch form limit, and AWS has introduced Lambda MicroVMs for long-running tasks. The news also highlights several cybersecurity threats, including Russian-led phishing campaigns and SharePoint zero-day vulnerabilities.

Blast from the past as GIMP 0.54 is revived in Flatpak form

This technology news digest covers updates on cloud computing advancements, emerging security threats, and open-source software developments. Key highlights include AWS’s new Lambda MicroVMs, phishing attacks targeting Signal users, and the revival of GIMP 0.54 through Flatpak.

🔒 Security & Privacy

Hacker employs Claude to breach booking firms, leaves millions of records publicly accessible

A Russian hacker used HexStrike AI integrated with Anthropic’s Claude to breach multiple hotel booking platforms by disguising malicious activity as legitimate penetration testing. The attack resulted in the theft of over 2 million email addresses along with sensitive guest information, including names, phone numbers, and reservation details.

Critical FFmpeg flaw discovered: just watching a video can fully compromise your system

A critical vulnerability known as “PixelSmash” (CVE-2026-8461) in FFmpeg’s MagicYUV decoder allows attackers to execute arbitrary code via malicious video files without any user interaction. Due to the widespread use of FFmpeg in various applications and IoT devices, users are urged to immediately update to version 8.1.2 or disable the affected decoder.

North Korean hackers infiltrated software used to build AI apps, Microsoft says

Microsoft has identified the North Korean hacking group Sapphire Sleet as responsible for a supply chain attack targeting 141 Mastra npm packages used to build AI applications. By compromising a trusted contributor account, the attackers distributed malware designed to steal cryptocurrency wallet data, browser extensions, and login credentials from developers’ devices.

The European Commission’s plan to replace disruptive cookie banners with an automated consent signal has been abandoned following opposition from Google and several EU member states. This setback means the tracking industry can continue using manipulative “dark patterns” to harvest user data through frustrating banner interactions.

🏗️ Hardware & Infrastructure

Windows hibernation may contribute to SSD wear as storage costs rise

Windows hibernation can accelerate SSD wear by writing RAM contents to the disk, potentially shortening the lifespan of drives with low endurance or limited capacity. This issue is drawing increased attention as sharply rising storage prices make hardware replacement much more expensive than in recent years.

China’s Arm-based LineShine beats US El Capitan by 20%+ to become world’s fastest supercomputer

China’s LineShine supercomputer in Shenzhen has surpassed the U.S. El Capitan to become the world’s fastest, marking China’s first top ranking since 2017. Notably, the system achieves its high performance using only standard microprocessors instead of the graphics processing units (GPUs) typically used in high-end computing.

Data center boom exposes US power grid weaknesses and needed infrastructure upgrades

The rapid expansion of AI data centers is creating unprecedented electricity demands that threaten to bottleneck technological growth. Although the U.S. has sufficient energy production, outdated grid interconnection processes and significant backlogs are preventing new infrastructure from accessing necessary power.

Eli Lilly plans $7.3B biotech “App Store” using 1,016 Blackwell chips in new data center.

Eli Lilly plans to utilize its $7.3 billion cash reserve to create an “App Store” for biotech scientists aimed at accelerating drug discovery through AI. The company also intends to launch a new data center featuring 1,016 Blackwell chips in 2025 to support these collaborative efforts with smaller biotech firms.

AWS Launches New Lambda MicroVMs

AWS has introduced Lambda MicroVMs, a new serverless compute primitive built on Firecracker technology that provides VM-level isolation for securely executing user and AI-generated code. This update supports runtimes of up to eight hours and features near-instant launch/resume speeds with state preservation, making it ideal for long-running tasks such as managing AI agents.

Datacenters dip a toe back into waterborne computing despite obvious challenges

This news digest highlights recent advancements in cloud computing, such as AWS’s new Lambda MicroVMs and the exploration of waterborne datacenters. It also covers critical cybersecurity threats, including phishing and zero-day exploits, alongside updates in open-source software and global technology policy.

Digital indigestion: Fizzy Coca-Cola display chokes on full storage

This technology news digest covers recent advancements in cloud infrastructure, such as AWS’s new Lambda MicroVMs, and Google’s shift toward AI-integrated databases. It also highlights various cybersecurity threats and significant updates within the open-source community, including the release of OpenBSD 7.9.

💰 Business & Finance

Digital euro gains support as Europe seeks payment independence from the US

A European Parliament committee has backed legislation for a digital euro aimed at reducing Europe’s reliance on U.S.-based payment networks like Visa and Mastercard. If approved, the European Central Bank plans to launch a pilot program in 2027, with a potential full rollout by 2029.

Crypto analytics startup Allium raises $40M Series B led by Amplify Partners

New York-based crypto analytics startup Allium has raised $40 million in a Series B funding round led by Amplify Partners, with participation from Kleiner Perkins and Theory Ventures. The company specializes in making complex blockchain data legible and usable for institutional players such as Coinbase and Visa.

EU to impose €3 duty on items under €150 from July 1 to curb Shein and Temu imports

Starting July 1, the European Union will impose a €3 duty on online purchases valued under €150. The measure aims to curb the influx of low-priced merchandise from non-EU retailers such as Shein and Temu.

Tencent negotiating exits from Japanese game studios as it reassesses global portfolio

Tencent Holdings Ltd. is negotiating exits from several minority investments in Japanese game studios, including Marvelous Inc., as part of a global portfolio reassessment. The company may sell these stakes back to the original management teams, even if such moves result in financial losses.

Apple is going to raise device prices, but when?

Apple may soon increase product prices due to a global shortage and rising costs for RAM and SSD components. While some analysts suggest these price hikes are imminent, others believe they may be timed to coincide with new hardware releases this fall.

Crypto in 2026: Oh, This Is the Bad Place

The article presents a dystopian vision of 2026 where cryptocurrency is deeply integrated into political influence, military prediction markets, and shadow finance. It argues that these developments transform crypto-based instruments into self-referential systems that lack the traditional market function of price discovery for real-world value.

Ideate a trading strategy with an Ex-Citadel Trader

A former Citadel Securities options market maker is offering a consultancy service to help traders develop robust and actionable trading strategies. Using a professional market-making perspective, the service evaluates trade ideas by assessing edge legitimacy, determining optimal position sizing, and analyzing execution feasibility.

Brits still reckon Big Tech isn’t paying enough tax

This technology news digest covers British support for taxing Silicon Valley companies alongside recent developments in cloud computing and cybersecurity. It also highlights significant software updates, such as the release of OpenBSD 7.9 and new AWS Lambda features.

🌐 Internet & Society

UFO skeptic Michael Shermer apologizes to David Grusch

UFO skeptic Michael Shermer has apologized to whistleblower David Grusch for previously disparaging his claims regarding recovered non-human technology. The apology follows Shermer’s recent appointment to Avi Loeb’s new UAP Science Advisory Panel.

Man tries to make a sale on Facebook Marketplace, gets scammed out of $300 via Zelle

A man lost $300 after falling for a Facebook Marketplace scam involving a fraudulent Zelle email that claimed funds needed to be “unlocked” with an additional payment. While he shared his experience on TikTok as a warning, many commenters expressed little sympathy, noting the tactic is a well-known fraud method.

Tensions brew over use of Palantir software in Germany

The Green Party in North Rhine-Westphalia is pushing to cancel state contracts with Palantir due to concerns over mass surveillance and human rights violations. This move threatens to create tension within the ruling coalition and reflects a growing movement in Germany toward digital sovereignty through the use of European or open-source alternatives.

NLRB judge rules Amazon broke law by refusing to recognize SF Teamsters; must bargain.

A US NLRB judge has ruled that Amazon violated federal law by failing to recognize the Teamsters union at a San Francisco warehouse. As a result, the company is now required to engage in collective bargaining with the workers.

It’s Only When You Look Back

Marking his website’s 25th anniversary, an author reflects on four decades of technological evolution by comparing early 8-bit microcomputers to modern high-speed computing. He notes that the profound scale of such advancements is often most visible through a retrospective lens.

Wikipedia cofounder Larry Sanger blocked from editing Wikipedia

Wikipedia co-founder Larry Sanger has announced his intention to return to the platform in fall 2025 to lead various reform efforts. His proposed initiatives include establishing the WikiProject Intellectual Diversity and presenting the “Nine Theses on Wikipedia” to foster a more inclusive contributor base.

Will It Mythos?

A new benchmark suite has been developed using real-world security vulnerabilities discovered by Mythos to evaluate whether other AI models can replicate its specialized bug-finding capabilities. By testing models against bugs that occur after their training data cutoffs, the study aims to determine if Mythos possesses unique power or if its performance is merely hype.

1,700 free online courses from top universities

Over 1,700 free online courses from prestigious institutions such as Yale, MIT, and Harvard are available in subjects including archaeology, architecture, and art history. To access the course content without cost, learners must select the “audit” or “no certificate” options on platforms like Coursera and edX to avoid paying for credentials.