San Francisco City Attorney Orders Meta to Explain AI Child Abuse Ads

San Francisco's City Attorney's Office has issued a formal demand to Meta over harmful AI-generated child abuse advertisements that repeatedly appeared on Facebook and Instagram. Authorities are seeking an explanation from the company about how such ads were allowed to run multiple times on its platforms. Meta has pushed back, arguing that the matter falls outside the city's legal jurisdiction. The dispute highlights growing concerns over the adequacy of Meta's content moderation systems in detecting and removing exploitative material.

Schedule E Guide: Key Tax Expense Categories Every Landlord Should Track

Schedule E on Form 1040 is where landlords report rental income and expenses to the IRS each year, but many small landlords miss out on deductions due to poor record-keeping. Deductible categories include advertising, auto mileage, repairs, insurance, mortgage interest, property taxes, and depreciation, among others. A critical distinction exists between repairs, which are deducted in the year paid, and improvements, which must be depreciated over several years — misclassifying them is a common audit trigger. Depreciation is especially valuable, as a residential rental property depreciates over 27.5 years, potentially yielding thousands in annual deductions without any cash outlay. Experts recommend logging each transaction with a date, category, and property tag as it occurs, and keeping separate records for each rental property to ensure accurate year-end reporting.

nginx blocks new HTTP QUERY method while Caddy and Traefik pass it through

RFC 10008, which reached Proposed Standard in June 2025, introduces QUERY as a new HTTP method that is safe and idempotent like GET but can carry a request body like POST. A developer tested how major reverse proxies and frameworks handle QUERY requests by setting up a DigitalOcean droplet with FastAPI, Django, nginx 1.24.0, Caddy 2.11.4, and Traefik 3.7.10. Testing revealed that nginx silently rejects QUERY requests rather than forwarding them to the backend, while Caddy and Traefik passed the method through without issue. On the framework side, FastAPI handled QUERY correctly when explicitly configured, but Django's class-based views returned a 405 Method Not Allowed response despite following the same lowercase-dispatch pattern used for GET and POST. The findings highlight that adopting QUERY in production requires verifying compatibility at every layer of the stack, as the RFC itself does not specify how non-compliant intermediaries will behave.

Developer Documents Sysmon and Wazuh SIEM Detection Setup, Learns From Troubleshooting

A cybersecurity learner set out to explore what Microsoft's Sysmon tool can capture by tuning its configuration on a Windows endpoint and forwarding logs to the Wazuh SIEM platform. The project involved modifying an existing Sysmon XML configuration file on a Windows server by removing an exclusion for the conhost.exe process to increase process telemetry. After applying the change, the expected conhost.exe Process Create events (Event ID 1) did not appear as anticipated, prompting deeper investigation into how Sysmon records process creation. The experience highlighted that effective SIEM detection requires not just log collection but also a clear understanding of what to look for and how to validate test results accurately. The author noted that troubleshooting self-made errors proved to be a valuable learning experience in understanding the investigative side of cybersecurity.

NSE cuts IPO size, may raise 23.5k crore

How a Reckless 1958 Mass Card Drop in Fresno Seeded the Visa Network

In September 1958, Bank of America mailed 60,000 unsolicited, pre-activated credit cards to residents of Fresno, California, in a bid to force-start a payment network — an experiment that quickly spiraled into widespread fraud and over 20% delinquency, earning the internal nickname 'the Fresno Drop.' The stunt exposed a fundamental challenge of two-sided networks: cardholders won't adopt a card merchants won't accept, and merchants won't invest in infrastructure for a card customers don't carry. Diners Club had tackled a similar problem in 1950, when Frank McNamara personally persuaded New York restaurants and wealthy diners to join simultaneously, proving that payment intermediation between strangers could be a viable business. By the late 1960s, Bank of America's licensed BankAmericard network had expanded nationally but descended into chaos, with inconsistent fraud controls and chargeback rules threatening its survival. Bank manager Dee Hock resolved the crisis by designing a member-owned cooperative with universal, non-negotiable rules — a structure that became Visa in 1976.

Presibo Builds AI-Powered Continuous Care Loop to Bridge Gaps Between Hospital Visits

Presibo, a healthcare technology startup that began as a prescription management app during COVID-19, is developing an infrastructure designed to provide continuous patient monitoring rather than relying solely on periodic hospital appointments. The platform connects patients, smart medical devices such as blood pressure monitors and glucometers, an AI analysis layer, and healthcare professionals into what the company calls the Presibo Care Loop. Connected devices transmit readings directly into a patient's digital health record, allowing the system to track health trends over time instead of relying on isolated snapshots. An AI layer analyses this data to detect concerning patterns and routes alerts through an intelligent triage system, escalating cases to nurses, doctors, or emergency pathways based on severity. The company emphasises that the AI is intended to support healthcare professionals rather than replace them, keeping clinicians central to all care decisions.

Self-Hosting AI Models Rarely Beats Hosted Open-Weight APIs on Cost

When choosing between closed frontier APIs and self-hosted open-weight models, engineers often overlook a third option: hosted open-weight APIs that serve the same models without operational overhead. Closed frontier APIs cost roughly $2.50–$15 per million tokens, while hosted open-weight endpoints for models like Llama 4 run as low as $0.07–$0.90 per million tokens, and self-hosting on an H100 costs around $0.18 per million tokens — but only at 85% GPU utilization. In practice, production Kubernetes clusters average just 5% GPU utilization, which can inflate self-hosting costs up to 17 times compared to theoretical estimates. Additional operational work — version upgrades, monitoring, and capacity planning — adds $1,500–$4,000 per month in hidden costs not reflected in GPU bills. As a result, self-hosting only breaks even against hosted open-weight APIs at volumes exceeding 50 million tokens per day, a threshold most workloads never reach.

ChatGPT Surpasses 900 Million Weekly Users as OpenAI Closes $110B Investment Round

OpenAI announced in late February 2026 that ChatGPT has surpassed 900 million weekly active users, making it one of the most widely used AI products globally. The company also reported over 50 million consumer subscribers, more than 9 million paying business users, and 1.6 million weekly users of its coding tool Codex. Alongside the user metrics, OpenAI disclosed a $110 billion investment round backed by partners including Amazon, NVIDIA, and SoftBank to expand computing infrastructure and meet growing demand. OpenAI described ChatGPT as the primary entry point through which people first engage with AI, signaling its shift from a niche tool to a mainstream product. The figures suggest conversational AI has reached a scale where many employees, customers, and business stakeholders already have direct hands-on experience with it.

Naomi Campbell's 2009 McQueen look could sell for $1 million at auction

Turning Point USA Plans Major Midterm Spending Push in Key Swing States

Conservative organization Turning Point USA is preparing to invest tens of millions of dollars ahead of the upcoming midterm elections, according to sources familiar with the plans. The group intends to allocate funds toward advertising campaigns and voter turnout efforts. The spending strategy is reported to be concentrated in three battleground states: New Hampshire, Arizona, and Nevada. The initiative comes roughly a year after the death of Turning Point founder Charlie Kirk.

Debian 13 on Four Machines Reveals Sharp Gaps Between Bare Metal and Cloud Images

A hands-on comparison run on 2025-09-09 tested the same Debian 13 (Trixie) operating system across four machines — one Lenovo laptop and cloud instances on AWS, GCE, and Azure. The bare-metal laptop carried 1,923 installed packages, while the three cloud images ranged from just 328 to 350, reflecting fundamentally different assumptions about user needs. All four machines boot from NVMe storage, yet only two ship tools capable of reading NVMe health logs, and both AWS and Azure omit nvme-cli entirely. GCE stood out as the most complete cloud image, shipping both dmidecode and nvme-cli, and also auto-correcting a misaligned GPT header in the same boot it detected the warning — a fix AWS did not apply. Azure additionally reported a Retbleed CPU vulnerability that the other two cloud images did not flag, highlighting how platform-level choices, not just the shared kernel, drive meaningful differences in security visibility and disk management.

Debian 13 on Four Machines Reveals Sharp Gaps Between Bare Metal and Cloud Images

A hands-on comparison conducted on 2025-09-09 tested the same Debian 13 (Trixie) operating system across four machines — one Lenovo laptop and cloud instances on AWS, GCE, and Azure. The bare-metal laptop carried 1,923 installed packages, while the three cloud images ranged from just 328 to 350, reflecting fundamentally different assumptions about user needs. All four machines boot from NVMe storage, yet only two ship tools capable of reading NVMe health logs, and all three cloud images include zero firmware packages. Among the cloud providers, GCE stood out as the most complete, shipping both dmidecode and nvme-cli, while Azure flagged a Retbleed CPU vulnerability that AWS and GCE did not report. A GPT header mismatch warning appeared on both AWS and GCE, but GCE automatically corrected it during the same boot — highlighting how identical kernels and messages can produce opposite outcomes depending on the platform.

Apple CEO John Ternus Defends iPhone's Future in First Keynote Address

John Ternus has made his debut as Apple's new CEO, delivering his first keynote address to the public. In his opening remarks, Ternus firmly positioned the iPhone as a continued cornerstone of Apple's product strategy. His stance echoes an approach previously championed by Apple co-founder Steve Jobs roughly 25 years ago. The move signals that Apple's leadership transition will not bring a dramatic shift away from its flagship smartphone business.

£8.28M Grade II-Listed Cotswolds Manor Severely Damaged by Electrical Fire

A historic 17th-century mansion in the Cotswolds has been extensively damaged following a major electrical fire. Barton House, a Grade II-listed property located near Guiting Power, was valued at £8.28 million at the time of its sale. The manor had only recently changed hands, having been purchased by a private buyer in January 2025. Firefighters battled the blaze for nearly 12 hours before bringing it under control. The new owners described themselves as being in a state of shock following the devastating incident.

Four Threat Groups Exploited Same Chrome and Windows Zero-Day Kit

Four separate hacking groups were found using an identical exploit kit targeting vulnerabilities in Google Chrome and Microsoft Windows. The discovery highlights a concerning trend where multiple threat actors gain access to the same exploitation tools, amplifying the potential damage. Security researchers identified a 'patch gap' — the window between a vulnerability being exploited and a fix being deployed — as a key contributing factor. The accelerating pace of AI-assisted vulnerability discovery is also believed to have played a role in the wider spread of these exploits. The findings underscore growing risks as advanced hacking tools become more accessible to a broader range of malicious actors.

Emergency Slide Accidentally Deploys on Trump's Qatar-Gifted Air Force One

An emergency evacuation slide deployed unexpectedly aboard the Qatari-gifted aircraft designated as Air Force One for President Trump. The incident has renewed scrutiny over the safety and reliability of the plane. Safety concerns surrounding the aircraft have been raised repeatedly in recent months. The issue gained further attention after Trump reportedly switched to a different plane during a visit to Turkey due to undisclosed security concerns. The series of incidents has intensified questions about the airworthiness of the gifted jet.

Developer Launches ICU Studio, a Lightweight Desktop IDE for ICU Messages

A front-end developer has released ICU Studio, a standalone desktop application designed to simplify the creation and editing of ICU messages used in software internationalization. The developer built the tool after finding that existing editors like VS Code lacked proper syntax highlighting and structural recognition for ICU message formats. ICU Studio was built using React and Tauri, allowing it to launch in under two seconds while consuming minimal memory at idle. Currently at MVP stage, the app supports message parsing, syntax highlighting, formatting for improved readability, and multiple copy formats suited for both JSON files and component usage. The release addresses a gap in tooling for the ICU internationalization standard, which is widely used in multilingual web development.

Pixy Tool Records Visual Context So Coding Agents Implement UI Changes Accurately

A new developer tool called Pixy aims to reduce miscommunication between developers and AI coding agents during visual UI edits. When a developer makes a change on a live site, Pixy automatically captures the visual context — including the target element, CSS property, old and new values, viewport width, and scope of the change. This eliminates the need for developers to manually re-describe what they already did in a text prompt, which often leads to agent misinterpretation. The coding agent then uses this recorded context alongside the repository to determine the correct implementation, whether in Tailwind, a CSS module, or another stylesheet. The tool is designed for visual frontend work where the relevant context exists at the moment of change but is typically lost when converted to natural language.

Delhi CSE Student Vinayak Joins DEV to Document Real-World ML Debugging Lessons

Vinayak Garg, a computer science undergraduate from Delhi, has joined the DEV Community platform to write about his hands-on experience building and deploying machine learning projects. His completed work includes a network intrusion detection system with an F1 score of 0.991, a customer churn model, and a student performance predictor. During a two-month internship, he also developed a LangGraph-based WhatsApp campaign bot and a multi-tenant RAG chatbot using Qdrant and Groq. Vinayak says his writing will focus on the debugging and deployment challenges that standard tutorials typically leave out, such as Docker caching issues and silent exceptions in LangGraph nodes. He plans to cover topics including ML model deployment to AWS EC2 with GitHub Actions, sharing the specific mistakes he encountered along the way.