Edition for June 6, 2026

In this article, we address the issue of in-memory data storage in FastAPI applications by connecting to a SQLite database using SQLAlchemy. This allows applications to persist data beyond server restarts, a crucial feature for modern applications like Instagram or ChatGPT.

• •Learn how to connect FastAPI to a SQLite database.
• •Understand the limitations of in-memory data storage.
• •Explore the use of SQLAlchemy for database interactions.

Retrieval-Augmented Generation (RAG) enhances AI applications by allowing them to answer questions based on relevant data rather than solely on their training. This method improves accuracy and verifiability by grounding responses in specific source material, addressing the limitations of large language models that lack up-to-date knowledge.

• •RAG improves AI response accuracy by using relevant data.
• •Large language models struggle with knowledge cutoffs.
• •Effective chunking of data is crucial for retrieval.

HP's upcoming ZGX Fury GB300 is set to be the most powerful Windows AI PC, featuring an impressive 784GB unified memory capable of handling one trillion parameters. This workstation is targeted at professionals requiring top-tier AI performance, though it comes with a hefty price tag.

• •HP announces the ZGX Fury GB300 for AI applications.
• •It features 784GB of unified memory.
• •The workstation can handle one trillion parameters.

You can use the new console experience on Amazon Bedrock to browse and compare the latest AI models side by side, organize work into projects with streamlined evaluation workflows, and access project-aware live documentation with auto-prefilled code snippets ready to copy and run.

• •New console experience in Amazon Bedrock enhances usability.
• •Browse and compare AI models side by side.
• •Organize work into projects with streamlined workflows.

The article discusses the urgent need for the AI industry to implement cost management strategies as expenses rise. The focus has shifted from rapid development to establishing controls and guidelines to manage these costs effectively.

• •AI costs are escalating rapidly, prompting industry reevaluation.
• •The conversation has shifted from speed to cost control.
• •Companies are seeking guardrails for sustainable AI development.

Researchers have prototyped an AI-powered internet worm that carries its own LLM and runs on compromised computers. This innovation closely resembles John Brunner's original 1975 conception of a computer worm, showcasing the potential dangers of AI in cybersecurity.

• •Researchers have developed an AI-powered internet worm prototype.
• •The worm carries its own LLM to operate on compromised systems.
• •This concept aligns with John Brunner's 1975 vision of computer worms.

This edition of The Download discusses the recent Meta hack where attackers exploited an AI customer support agent to steal Instagram accounts. It highlights the growing concerns around AI security and the implications of chatbots on our cognitive processes.

• •Meta's AI customer support agent was exploited in a recent hack.
• •Attackers stole Instagram accounts using this AI system.
• •The incident raises concerns about AI security measures.

Learning to code in the age of AI can lead to passive learning, where one confuses activity with progress. The author shares a 5-phase checklist to ensure active learning, emphasizing the importance of struggling with problems before seeking AI assistance. This approach fosters genuine understanding and independent thinking, which are crucial in a world where AI can easily generate code.

• •Passive learning can mislead learners into thinking they're making progress.
• •A structured approach to learning coding is essential for true understanding.
• •The author developed a 5-phase checklist to enhance learning.

In this article, we will explore three essential spaCy tricks that every developer should have in their toolkit to maximize processing speed and customize entity recognition.

• •Learn to optimize spaCy for faster text processing.
• •Discover methods to customize entity recognition.
• •Enhance your NLP projects with practical spaCy tips.

On June 3, 2026, Google introduced Gemma 4 12B Unified, an open-source multimodal model designed to understand text, images, audio, and video within a single architecture. It combines a 256K context window with an efficient, laptop-friendly design aimed at agentic workflows and local deployment.

• •Google Gemma 4 12B is an open-source multimodal model.
• •It can process text, images, audio, and video.
• •Features a 256K context window for enhanced understanding.

During a screen-sharing session, the author almost exposed sensitive customer data. This experience led to the development of a tool that automatically blurs sensitive information in real-time, ensuring privacy during presentations. The solution uses CSS layers to blur elements before they are shared, preventing accidental data leaks.

• •Accidental data exposure can happen during screen sharing.
• •Standard advice for avoiding leaks often fails under pressure.
• •A new tool blurs sensitive information automatically.

Learn how Endava is using AI agents, ChatGPT Enterprise, and Codex to accelerate software delivery, automate workflows, and build an AI-native culture across the enterprise.

• •Endava is leveraging AI agents for software delivery.
• •ChatGPT Enterprise and Codex play key roles in their strategy.
• •The focus is on automating workflows across the organization.

This tutorial guides developers in building and deploying AI agents using Google’s Agent Development Kit (ADK) on GKE Autopilot. It emphasizes the importance of a scalable infrastructure and security, utilizing Workload Identity for permission management and the Kubernetes Gateway API for exposure.

• •Learn to build AI agents with Google’s ADK.
• •Deploy agents on GKE Autopilot for scalability.
• •Implement security using Workload Identity.

A comprehensive guide to optimizing LLM inference by eliminating padding overhead with hardware-aware sequence packing.

• •Focus on optimizing LLM inference performance.
• •Eliminate padding overhead for better efficiency.
• •Utilize hardware-aware sequence packing techniques.

The Australian data center operator AirTrunk has announced a significant investment of $30 billion to establish 5GW of AI data center capacity in India. This move highlights the growing demand for data infrastructure to support AI technologies and the expansion of cloud services in the region.

• •AirTrunk plans to invest $30 billion in India.
• •The project aims to build 5GW of AI data center capacity.
• •This investment reflects the rising demand for data infrastructure.

AI Gateway now features real-time spend limits to prevent runaway token bills across multiple AI providers. By integrating with Cloudflare Access, companies can use identity-driven budgets and policies.

• •AI Gateway helps manage costs associated with AI services.
• •Real-time spend limits prevent unexpected high bills.
• •Integration with Cloudflare Access enhances budget control.

The threat actor known as PCPJack has hijacked cloud servers from AWS, Google Cloud, and Microsoft Azure to establish a covert SMTP email relay network. Compromised servers across multiple continents were converted into SMTP proxies, verified for mail relay capability, and synced every five minutes.

• •PCPJack has targeted AWS, Google Cloud, and Azure servers.
• •A covert SMTP email relay network has been established.
• •Compromised servers span across the U.S., Europe, and Asia.

Cohesity announced the issuance of Patent No. 12,619,501, which integrates generative artificial intelligence into data protection environments. The registration, issued by the USPTO, aims to optimize data recovery in backup systems, highlighting innovation in using AI to enhance the efficiency and security of corporate data.

• •Cohesity received a patent for applying AI to backup data.
• •The patent was granted by the USPTO in May 2026.
• •Integrates generative artificial intelligence in data protection.

When security teams scan their container environments, they often find numerous known vulnerabilities originating from base image packages rather than application code. Hardened images aim to reduce these vulnerabilities by minimizing the number of installed packages, thereby decreasing the attack surface and enhancing overall security.

• •Security teams find many vulnerabilities in container environments.
• •Most vulnerabilities come from base image packages, not application code.
• •Hardened images reduce installed packages to minimize risks.

DJ Sampath from Cisco highlights how AI agents can transform the work of IT professionals by offering 'superpowers' through the new Cloud Control platform. This innovation combines network management, security, and computing, promising to optimize processes and enhance efficiency in the industry.

• •DJ Sampath is Senior Vice President at Cisco.
• •Cisco launched the Cloud Control platform at Cisco Live 26.
• •The new platform integrates network management and security.

In a previous post, we showed you how to query Amazon SageMaker Catalog metadata using SQL by using the metadata export feature. This post builds on that foundation by demonstrating how to create governance dashboards with Amazon Quick.

• •Learn to query Amazon SageMaker Catalog metadata using SQL.
• •Utilize the metadata export feature for effective data management.
• •Create governance dashboards to visualize data insights.

New research from Surfshark reveals that 90% of popular messaging apps incorporate AI, raising significant privacy concerns. The study highlights alarming data harvesting practices, with Signal being recognized as the most secure option among them.

• •90% of messaging apps now use AI technology.
• •Research shows significant data harvesting practices.
• •Signal is identified as the most secure messaging app.

AI agents are rapidly being adopted, with 60% of organizations deploying them. However, 40% face security and compliance challenges that hinder further scaling. AI governance is crucial in bridging the gap between adoption and oversight, especially as AI begins to influence more significant decisions.

• •60% of organizations have AI agents in production.
• •40% cite security and compliance as major barriers.
• •AI governance is essential for responsible AI deployment.

Multiple software supply chain attacks have targeted the npm ecosystem, utilizing malicious versions of over 50 legitimate packages. These attacks aim to distribute a Rust-based information stealer and a self-spreading worm, posing significant risks to developers and their environments.

• •Recent supply chain attacks have affected the npm ecosystem.
• •Over 50 legitimate packages were compromised.
• •Attackers used a Rust-based information stealer.

Arabic-speaking users have become targets of a new Android spyware named Asin, as reported by ESET. The malware has been spreading through various campaigns since early 2025, utilizing fake websites that mimic utilities, war-related updates, and government news sources to lure victims.

• •New Android spyware named Asin targets Arabic-speaking users.
• •Detected by ESET, the malware spreads through various campaigns.
• •Utilizes fake websites mimicking utilities and war updates.

The U.S. National Security Agency is reportedly preparing to utilize Anthropic's AI model, Mythos, for cyber operations, despite a federal ban on employing the company's technology. This development raises significant questions about the implications of AI in cybersecurity.

• •NSA is reportedly preparing Anthropic's Mythos for cyber operations.
• •The use of AI in cyberattacks is a controversial topic.
• •Federal ban exists on using Anthropic's technology.

From January to May 2026, Mandiant identified a data theft extortion campaign by the threat group UNC3753, targeting US law firms. They use vishing and social engineering to gain access to sensitive data, often completing attacks within a single day. This post details their tactics and offers recommendations for safeguarding corporate environments.

• •UNC3753 targets US law firms using vishing and social engineering.
• •Attacks can occur in person, with actors posing as IT technicians.
• •Data theft and extortion can happen within a single business day.

Cybersecurity researchers have identified a new threat cluster named OP-512, targeting Microsoft IIS servers with a custom web shell framework. This espionage activity is believed to be linked to China, indicating a significant threat to web server security.

• •New threat cluster OP-512 targets Microsoft IIS servers.
• •Utilizes a bespoke web shell framework for deployment.
• •Espionage-focused activity linked to China.

The article discusses the recent technology package announced by the European Union, addressing its implications and gaps. It also presents alternatives that Brazil could consider to ensure its digital sovereignty in a rapidly evolving global landscape.

• •Analysis of the European Union's technology package.
• •Implications for Brazil's digital sovereignty.
• •Proposed alternatives for Brazil.

WWDC 2026 week highlights include Apple's AI credibility test, developer wishlists, and advancements in SwiftUI animation timing. The article also discusses Core Data with Observation and delves into Swift concurrency, providing insights for developers looking to enhance their skills and adapt to new technologies.

• •Apple's AI credibility is under scrutiny as WWDC approaches.
• •Developers are sharing wishlists for new features and improvements.
• •SwiftUI animation timing is getting attention for better performance.

The credit card took decades to replace cash, and mobile phones accelerated this process. Now, the financial industry aims to make the act of paying nearly imperceptible, with the evolution of contactless payments paving the way for new payment technologies, such as the use of wearable devices.

• •The credit card revolutionized payment but took decades to become popular.
• •Mobile phones accelerated the transition to digital payments in seconds.
• •The financial industry seeks to make payments nearly imperceptible.

Virtueyes has launched a 4G vehicle tracker with eSIM, aiming to meet the growing demand for vehicle monitoring solutions. The expectation is to sell 10,000 units in three months, highlighting the importance of connectivity technology in the automotive sector.

• •Virtueyes presents a 4G vehicle tracker with eSIM.
• •Product aims to meet the demand for vehicle monitoring.
• •Sales expectation is 10,000 units in three months.

Vero announced an investment of R$ 110 million to expand its mobile phone operations, offering new packages that include free access to Vero Video and Globoplay, along with improvements in broadband and B2B services.

• •Vero will invest R$ 110 million in mobile telecommunications.
• •New packages will include free access to Vero Video and Globoplay.
• •Broadband services will also be enhanced.

PicPay reported a net income of R$ 169 million in the first quarter of 2026, representing a 92% increase compared to the previous year. The company also reached 68.6 million active customers, an 11% growth, and 44.3 million active accounts.

• •Net income of R$ 169 million in Q1 2026.
• •92% growth compared to the previous year.
• •68.6 million active customers.

Inter has launched payment wearables aimed at retaining high-income customers. In addition to facilitating transactions, these devices also serve as a third factor of authentication in the bank's app and provide access to exclusive environments, such as hotels and airport VIP lounges.

• •Inter introduces payment wearables for premium customers.
• •Devices offer additional security as an authentication factor.
• •Allow access to exclusive environments such as hotels and VIP lounges.

Discover three post-hoc methods for closing the gap between confidence and accuracy in language models: Platt Scaling, Isotonic Regression, and Temperature Scaling. These techniques help improve model calibration, ensuring that predicted probabilities align more closely with actual outcomes.

• •Platt Scaling adjusts the output probabilities of models.
• •Isotonic Regression is a non-parametric method for calibration.
• •Temperature Scaling modifies the logits of model outputs.

This article discusses the use of DSPy to automate the creation, evaluation, and optimization of prompts for large language models (LLMs). It emphasizes the importance of efficient prompt engineering in enhancing the performance of AI models.

• •Explores the role of DSPy in prompt automation.
• •Highlights the significance of prompt optimization.
• •Discusses evaluation methods for LLM prompts.

Python tutorial for fine-tuning a Mistral Small 3.1 on an imbalanced training set to classify 15 emotions in social media communication.

• •Focus on fine-tuning a specific model for emotion recognition.
• •Utilizes an imbalanced training dataset for classification.
• •Covers 15 distinct emotions relevant to social media.

In this post, you learn how to set up Kiro with the Amazon Redshift MCP server to query your data warehouse using natural language. You explore cluster discovery, schema browsing, analytical queries, cross-cluster comparisons, and data quality checks, all without writing SQL from scratch or switching between tools.

• •Set up Kiro with Amazon Redshift for natural language queries.
• •Explore cluster discovery and schema browsing features.
• •Perform analytical queries without SQL coding.

In Part 1 of this series, we introduced Chronos-2, a time-series foundation model. We got our hands dirty by walking through a real case study and saw what Chronos-2 can do straight out of the box, with no training. But as we noted at the end of Part 1, zero-shot isn't always enough.

• •Chronos-2 is a time-series foundation model.
• •The article discusses fine-tuning techniques for better performance.
• •Real case studies demonstrate its capabilities.

At Google Cloud, the Managed Service for Apache Spark now offers serverless and managed cluster deployment modes, enhancing efficiency for big data workloads. Key improvements include faster execution speeds, easier resource management, and smarter scaling policies. The service aims to support modern data teams with customizable infrastructure and AI integration.

• •Managed Service for Apache Spark enhances big data processing.
• •Offers serverless and managed cluster deployment modes.
• •Focuses on faster execution speeds and easier resource management.

In this post, you learn how to use Data Agent in Query Editor to explore data, build multi-step analyses, recover from errors, and summarize results using a public education dataset.

• •Learn to use Data Agent in Query Editor.
• •Explore data and build multi-step analyses.
• •Recover from errors during data analysis.

Email classification in Dynamics 365 ensures only meaningful customer emails become cases. Reduce noise, improve triage, and boost service productivity.

• •Email classification helps prioritize customer communications.
• •It reduces irrelevant emails from becoming cases.
• •Improves the efficiency of service teams.

The S&P 500 has rejected SpaceX's entry, also denying waivers for unprofitable AI firms like OpenAI and Anthropic. This decision reflects the index's stringent rules regarding profitability, impacting the growth prospects of these innovative companies in the financial market.

• •S&P 500 blocks SpaceX from entering the index.
• •Unprofitable AI firms like OpenAI and Anthropic are also affected.
• •The decision highlights strict profitability rules of the index.

In a statement, a Google representative described the deal as a result of unexpected demand for its recently launched AI products.

• •Google partners with SpaceX for cloud computing.
• •The deal is valued at $920 million per month.
• •Demand for AI products drives this collaboration.

While the AI fundraising machine keeps breaking records, some founders are focusing on reducing screen time. Mirror founder Brynn Putnam raised money for Board, a startup promoting in-person games and social experiences. Cyberdeck creators are viral for crafting DIY computers that encourage users to engage with the real world.

• •AI fundraising continues to break records.
• •Founders are focusing on reducing screen time.
• •Board promotes in-person games and social experiences.

Latin America is drawing global investors interested in artificial intelligence (AI), seeking diversification beyond the US and Asia. A report from Santander highlights that the interest extends beyond low interest rates, reflecting a new investment dynamic in the region.

• •Global investors are focusing on Latin America.
• •Artificial intelligence is a central point of interest.
• •Santander published a report on this trend.

As the conflict in Iran disrupts the global oil supply, airlines are exploring alternatives to traditional jet fuel. One promising solution is the use of sustainable aviation fuel derived from used cooking oil and french fry grease, which could significantly impact the industry's future.

• •Airlines are seeking alternatives due to oil supply disruptions.
• •Sustainable aviation fuel is derived from used cooking oil.
• •French fry grease is being considered as a viable energy source.

The Hollywood actors' union has approved a new contract that establishes limits on the use of characters created by artificial intelligence.

• •Hollywood actors' union approves new contract.
• •Contract limits the use of characters created by AI.
• •The measure aims to protect actors' rights.

While the AI fundraising machine keeps breaking records, some founders are focusing on in-person social experiences. Brynn Putnam's Board aims to bring people together through games, while Cyberdeck creators promote DIY computers that encourage outdoor interaction, reflecting a trend towards community and connection amidst the tech landscape.

• •AI fundraising continues to set new records.
• •Founders are exploring alternatives to AI-driven solutions.
• •Board focuses on in-person games and social experiences.

Throughout Heineken's 87 years as a publicly traded company, the CEO position has never been held by an external executive. However, shareholders are pushing for innovation and leadership changes, suggesting that the company needs new perspectives to adapt to market demands.

• •Heineken has been publicly traded for 87 years.
• •Traditionally, the CEO is an internal executive.
• •Shareholders are calling for innovation in leadership.

Helion, the nuclear fusion company backed by Sam Altman, CEO of OpenAI, raised $465 million, tripling its market value to $15.5 billion. The investment round was led by Thrive Capital, highlighting the growing market confidence in nuclear fusion technology.

• •Helion is a company focused on nuclear fusion.
• •Sam Altman, CEO of OpenAI, is a key supporter of Helion.
• •The company raised $465 million in a new investment round.

The United States cited Pix, the primary payment method for small Brazilian businesses, in an investigation into unfair practices. The payment system is viewed as state competition to private credit cards, leading to an analysis of potential taxation.

• •Pix is the primary payment method for small businesses in Brazil.
• •U.S. investigation questions unfair practices related to Pix.
• •The payment system may be seen as competition to credit cards.

The Bengaluru startup FirstClub has doubled its valuation to $255 million in just nine months, achieving over 1 million orders and a $50 million annualized GMV run rate within its first year of operation.

• •FirstClub is a Bengaluru-based startup.
• •The company has doubled its valuation to $255 million.
• •It achieved over 1 million orders in nine months.

Defense tech is booming, with companies like Anduril and Mach Industries seeing significant valuation increases. The U.S. government is proposing a 40% rise in the defense budget, attracting numerous startups vying for government contracts. Ross Fubini, a venture investor, warns that many of these startups may struggle to survive the critical phase between prototype and contract.

• •Defense tech is currently experiencing rapid growth.
• •Anduril and Mach Industries have seen their valuations soar.
• •The U.S. government plans a substantial increase in defense spending.