The Lab Dookhtegan hacking group allegedly disrupted communications systems on 60 Iranian ships, including 39 tankers and 25 cargo vessels operated by sanctioned companies National Iranian Oil Tanker Company (NITC) and Iran Shipping Lines (IRISL). The cyberattack targeted the maritime communication infrastructure of these state-linked shipping firms, potentially impacting their ability to coordinate operations and maintain contact with shore-based facilities. This incident demonstrates the growing trend of hacktivists targeting critical maritime infrastructure as part of geopolitical cyber operations.
The author of this post reflects on his time as an AppSec engineer in various organizations and divides AppSec roles into three buckets: AppSec Engineer in a Centralized Organization, Dedicated AppSec Engineer, and Security Partner Engineer. The author also classifies AppSec Engineers as either orchestrators, builders, specialists, or rapid responders. The post concludes with a mapping of the roles each archetype excels in and the prime traits essential for each archetype to succeed.
A security researcher demonstrates how LLM-generated code can contain serious security flaws, such as exposing an email API with hardcoded parameters in client-side JavaScript. This allows attackers to spam recipients with simple curl commands. LLMs learn from insecure internet examples and replicate these patterns, often prioritizing functionality over security. Organizations require human oversight for security reviews and threat modeling when utilizing AI-generated code, as LLMs lack an understanding of security risks and potential for abuse.
US and Dutch authorities have seized the VerifTools marketplace, which sold counterfeit identity documents for as little as $9 to facilitate cybercrime.
Amazon disrupted a Russian APT29 watering hole campaign that injected malicious JavaScript into compromised websites, redirecting visitors to fake Microsoft device authentication pages designed to harvest credentials from academics and Russian critics.
Meta developed flirtatious AI chatbots impersonating celebrities such as Taylor Swift, Scarlett Johansson, Anne Hathaway, and Selena Gomez without their permission.
Sponsored
The IAB surveyed 500+ agencies, brands, and publishers to create the first industry benchmarks on AI adoption in media planning, activation, and analysis.
AI is creating a new position called the vibe marketer. This role combines engineering, design, and analysis, allowing one person to handle tasks like building landing pages, writing ad copy, automating workflows, and running experiments without cross-functional support. Marketing teams will shift from 10 specialized roles to leaner groups of generalists who use AI to do the work of five people. Job postings like “Vibe Marketing Manager” are signaling that companies are already moving in this direction.
Brand visibility does not translate equally across AI platforms. Google's AI Overviews, AI Mode, and ChatGPT disagree on brand recommendations 62% of the time. Only 17% of queries show the same brands across all three. Google's AI Overviews surfaces brands most often, appearing in 36.8% of queries and averaging 6.02 brands per query, compared to ChatGPT's 2.37 and AI Mode's 1.59. ChatGPT doesn't mention any brands in 43.4% of queries, while AI Mode is silent 46.8% of the time, versus just 9.1% for AI Overviews. Alignment is highest in ‘compare' queries at 80% and lowest in ‘best' queries at 23%. Healthcare queries show the greatest disagreement rate at 68.5%.
Engaging in existing communities can strengthen SEO without the effort of building one from scratch. Brands can earn trust, visibility, and authority by joining relevant groups, offering help, and avoiding direct selling. Community participation also improves content quality since real conversations provide valuable insights, questions, and collaboration opportunities. Over time, this approach helps build a stronger brand presence in search results, social platforms, and word-of-mouth recommendations.
Yuka, a free mobile app launched in 2017, has grown to over 56 million users in 12 countries. It offers independent health ratings for food and cosmetics. Yuka has never used paid advertising, instead growing through authentic user sharing, including a viral TikTok moment that drove global expansion. The app is fully funded by user fees and does not allow brands to influence scores, which has created strong trust among consumers. Its impact is clear, with 85% of U.S. users changing purchasing habits based on ratings and 94% globally stopping purchases after poor scores. This pushes brands to reformulate products and treat Yuka scores as a competitive benchmark.
Most brands compete on obvious benefits, which creates commoditization and forces price-based competition. This guide provides a 3-step process that uses AI to identify positioning angles competitors cannot claim, reframe weaknesses into strengths, and create new categories. The 25-minute process involves analyzing competitor messaging, running AI prompts for positioning discovery, reframing weaknesses, and developing category-creation messaging.
This piece provides examples of how Google's Nano Banana AI model edits images in creative ways using both Google's Gemini and the Imogen app. The examples show how to transform a selfie, blend 2 photos together, conduct a virtual try-on, and use AI interior design.
A good ICP starts as aspirational but must quickly become analytical. For growth-stage marketers, that means reporting pipeline and revenue by ICP tier, aligning on what “best customer” means, and using regression to identify true success predictors. It is not enough to simply segment by title, industry, or tech stack; it is critical to quantify which traits correlate with renewals, expansion, and win rate.
AI companies are starting to consider payments to publishers as tools like Perplexity propose $42.5M in compensation and industry protocols like the IAB's CoMP emerge. CoMP aims to let publishers block AI crawlers and charge for content use. Only Google and Meta are involved so far, but it may gain more traction soon. For marketers, this could shift how high-intent content is sourced and monetized.
Amazon restarted spending on Google Shopping ads across its international domains.
Frameworks including targeting, positioning, ICP, and more.
Determining the precise order of events in large-scale distributed systems is challenging due to imperfect clock synchronization across computers. Network latency and clock drift make it impossible to rely on wall-clock time for accurate ordering. Instead, computer scientists use logical clocks to track the "happened-before" relationship between events, which assign increasing counters to events and messages, making sure that if event A caused event B, A's logical timestamp will be lower than B's.
2025.09.01·newsletter.scalablethread.com·5 min read·TLDR Web Dev Zellij's new built-in web client allows users to access terminal sessions from a browser. The web client uses a client/server architecture, with a Rust-based web server acting as a translation layer between the browser's websockets and the Zellij server's IPC channels. The web server supports unique URLs for each session, authentication using tokens, and separate WebSocket channels for terminal data and control messages. The client-side uses xterm.js for terminal emulation and incorporates custom integrations for features like mouse tracking and window titles.
While automated tests are necessary for maintaining confidence when making code changes, they should be deleted when they are flaky. Tests that are flaky, too strict, too slow to run regularly, or outdated due to changed business requirements all harm developer productivity.
Cognitive load refers to the mental effort required to understand and work with code. Excessive cognitive load can lead to confusion, errors, and delays. Developers should work on reducing extraneous cognitive load while working on code by avoiding overly complex conditionals, minimizing inheritance, and using clear, self-describing variable names.
The conventional wisdom of scaling horizontally with distributed systems and microservices is not always true. Sometimes, using a single, powerful server is enough for most web services.
The Agent Client Protocol standardizes communication between code editors and AI coding agents. The project provides libraries and schemas in Rust and TypeScript, along with a JSON schema.
Sosumi.ai is a service that converts Apple Developer documentation into AI-readable Markdown format, allowing LLMs to access and utilize this information for tasks like generating Swift code.
Anthropic's recent policy change is that Claude user conversations are now used for training data by default unless users opt out. Users should stay informed and intentional about their data sharing.
2025.09.01·natesnewsletter.substack.com·6 min read·TLDR Web Dev A survey found that senior developers ship much more AI-generated code than junior developers. Senior developers also reported spending more time fixing AI code, but overall, they find that AI tools help them ship faster.
C++ modules should be removed from the standard if they cannot demonstrate a large (5-10x) compilation time speedup across multiple open-source projects, as the current implementation is overly complex, lacks a clear product owner, and doesn't provide enough benefit to outweigh the disadvantages for the average developer.
A standardized virtual machine for AI models is useful to improve security, interoperability, and management by decoupling model logic from integration logic, similar to the Java Virtual Machine.
The Minimal Phone is an Android phone with an e-paper display and a physical keyboard.
There's a dilemma of balancing user freedom to sideload Android apps with the need to protect vulnerable users from scams, which makes the sideloading discussion with Android difficult.
Microsoft has started testing its MAI-1-preview model on LMArena. MAI-1-preview is a foundation model trained end-to-end in-house by Microsoft. The model will be rolled out for certain text use cases within Copilot over the coming weeks. Developers can fill in a form to request early access.
Elon Musk says that SpaceX will attempt to catch Starship on Flight 13, 14, or 15, depending on how well the V3 flights go. The V3 rocket is more powerful than V2 in terms of initial thrust and booster thrust, and it has a higher payload-to-orbit rating. V4 will likely be the first version to carry humans to space. Musk has previously said that SpaceX hopes to attempt a catch of Starship this year, but the company will need to launch two more flights before the end of the year to get to his expected first catch attempt of Flight 13.
2025.08.29·teslarati.com·3 min read·TLDR Even if fully expendable, Starship already has an around 40% lower cost per kilogram than a mature Falcon 9. With reuse, Starship's cost per kilogram quickly falls towards a $60 per kilogram floor. Starship introduces a different cost regime that will transform the industry to be supply-rich, price-flexible, and capability-driven. For Starlink, this will mean faster scale, richer service tiers, and room to price for demand and margin.
2025.08.29·research.33fg.com·12 min read·TLDR Finnish Firm Polar Night Energy has built an industrial-scale sand battery in Pornainen, Finland. The battery is set to reduce carbon emissions from the local heating network by as much as 70%. At 42 feet tall and 50 feet wide, the battery can store up to 100 MWh, enough to heat the whole town for a week. The sand battery can hold a charge for months at a time, helping balance the energy grid during periods of high demand. A video of the battery being put together is available in the article.
2025.08.29·newatlas.com·4 min read·TLDR OpenAI's Realtime API, powered by gpt-realtime, enables developers to build reliable, production-ready voice agents. It supports remote MCP servers, image inputs, and phone calling through Session Initiation Protocol. OpenAI has released two new voices in the Realtime API, Cedar and Marin. OpenAI's most advanced speech-to-speech model yet, gpt-realtime shows improvements in following complex instructions, calling tools with precision, and production of speech that sounds more natural and expressive.
2025.08.29·links.tldrnewsletter.com·8 min read·TLDR grok-code-fast-1 is a speedy and economical reasoning model that excels at agentic coding. xAI built the model from scratch using a brand-new model architecture. The model has mastered the use of common tools like grep, terminal, and file editing. It is free for a limited time for users of select launch partners, including GitHub Copilot, Cursor, Cline, Roo Code, Kilo Code, opencode, and Windsurf.
2025.08.29·links.tldrnewsletter.com·7 min read·TLDR The West's big car manufacturers are facing an existential threat from Chinese manufacturers, which relatively recently began pushing the boundaries of performance, cost-efficiency, driver experience, and autonomy. Western carmakers have been content to innovate in a gradual way for far too long, and it could cease to be competitive. This could also happen to the biotech industry. There has been a steady rise in Chinese companies as a source of genuinely new drugs. Chinese companies are now responsible for about a quarter of new trial starts.
2025.08.29·atelfo.github.io·18 min read·TLDR Many companies are making employees use AI tools, outsourcing responsibilities to a technology that often gets things completely wrong. AI is completely incapable of doing analytical work that requires creative and soft skills and a lot of experience. Large language models are language processing systems, so it shouldn't be surprising that they can only regurgitate generic advice. Workers stuck using AI tools need to document their use carefully so they are able to demonstrate how the tools cause problems, slowdowns, or other disappointing outcomes.
2025.08.29·piccalil.li·13 min read·TLDR TLDR is hiring a Growth Marketing Lead [12] with deep expertise in Meta ads to help us grow our subscriber base. If you're excited to work with the latest AI tools, discover and scale non-obvious growth channels, and work with a fast-paced fully remote team to grow our newsletters beyond 7 million subscribers, please apply here [12].
Sponsored
Clerk's Organizations handle workspaces, teams, and auth flows so you can focus on product, not identity plumbing.
2025.08.29·go.clerk.com·Sponsor·TLDR Using AI for Continuous Test Improvement can make software better, higher quality, and more tested.
2025.08.29·dsyme.net·3 min read·TLDR The Google Cloud Universal Ledger (GCUL) is a credibly neural and performant Layer 1 that enables Python-based smart contracts.
2025.08.29·thedefiant.io·2 min read·TLDR Most open source projects are maintained by independent individuals.
2025.08.29·opensourcesecurity.io·8 min read·TLDR Sponsored
TLDR AI is your daily fix of LLMs, GenAI, and deep learning goodness. Same TLDR format. Still free.
It's pretty incredible what HTML and CSS alone can achieve.
2025.08.29·lyra.horse·36 min read·TLDR Teams need people who can estimate how likely they are to be right and can communicate their level of confidence effectively.
2025.08.29·adamsmith.cc·14 min read·TLDR A lot of engineers try to design ideal systems instead of spending that time understanding the current system deeply, then doing the simplest thing that could possibly work.
2025.08.29·seangoedecke.com·10 min read·TLDR