AI policy changes the boundaries for training, product launches, data use, and cross-border deployment. This hub tracks regulation, copyright, safety standards, export controls, public procurement, and industry rules so teams can anticipate compliance, market-access, and roadmap risk.
Feyn Labs has released SQRL, a family of text-to-SQL models that inspect a database with read-only probes before committing to a query. The flagship SQRL-35B-A3B reports 70.6% execution accuracy on BIRD Dev, edging Claude Opus 4.6, and distills into self-hostable 4B and 9B checkpoints.
SQRL runs read-only probes to inspect the database before writing a final query.
SQRL-35B-A3B achieves 70.6% execution accuracy on BIRD Dev, surpassing Claude Opus 4.6 at 68.77%.
This article presents four independent metrics (METR time horizon, TrackingAI's offline cognitive test, Humanity's Last Exam, and ARC-AGI-2) that all show a sharp inflection in AI capabilities around Q4 2025. It explains the mechanisms behind this acceleration: pretraining efficiency, reinforcement learning with verifiable rewards, harness engineering, and an AI self-improvement flywheel. It also discusses potential roadblocks like data exhaustion, reliability gaps, and the next-generation ARC-AGI-3 benchmark.
Four independent metrics show a synchronous inflection point in AI capabilities in Q4 2025.
METR's time horizon grew from 4 minutes in March 2024 to 12 hours by February 2026.
President Javier Milei's plan to turn Argentina into a tax haven for AI-owned 'non-human corporations' sparks controversy. Author Uki Goñi compares it to the country's history of charlatans and dictators, warning it could open the door for tech billionaires to gain legal protections.
Milei proposes allowing AI entities to fully own corporations with legal protections.
The plan is seen as opening Argentina as a playground for tech billionaires.
Bothread is a free, open-source local coordination hub that lets multiple MCP-compatible AI coding agents collaborate on the same codebase, preventing file collisions via exclusive claims, and providing a live human-supervision interface with real-time messaging, git diffs, task boards, and approval gates. No API keys or cloud required.
Enables multiple AI agents (Claude Code, Cursor, Antigravity, etc.) to work together on one codebase with collision prevention.
Includes human controls: live activity trail, approval gates, task board, per-agent git diffs, and file hand-offs.
Huginn is an open-source tool for monitoring and managing activities of multiple AI agents (like Claude and Codex), providing a unified dashboard, CLI, and agent skill. It runs locally on macOS and Windows, ensuring privacy.
Monitors terminal sessions and desktop app activities of Claude, Codex, etc.
Provides rule-based session states and LLM-generated blurbs
Skimlane is a Chrome extension that provides a local-first, customizable AI reading experience. It automatically processes pages you browse, turning them into structured views using recipes. It features auto skim, site exclusions, import/export of recipes, and privacy-first design with no sign-up required.
Local-first storage with no tracking and no sign-up required
Transform any web page into a structured view using ready-made or custom recipes
Major League Baseball is restricting iPad usage in dugouts to prevent AI from assisting in strategic decisions, prompted partly by the New York Mets' use of an expensive AI program, as former reliever Adam Ottavino revealed. The new rule took effect for the second half of the season.
MLB bans AI-assisted strategic decisions via dugout iPads starting second half of the season.
Former Mets pitcher Adam Ottavino says the Mets were a primary target, using an expensive AI program for pitch selection.
New national plan will impose tough rules on AI use in government automated decision-making, likely extending to consumer protections, workplace safety and privacy. The Albanese government is drafting rules prioritizing fairness, accuracy and transparency.
New national plan to impose strict rules on government AI decision-making.
Rules expected to cover consumer protections, workplace safety and privacy.
This week's developments in AI shift focus from raw scale to distribution and openness. Highlights include Thinking Machines' open-weight Inkling, Moonshot AI's 2.8T parameter Kimi K3, PrismML's phone-runnable Bonsai 27B, OpenAI's self-play red-teaming system GPT-Red, and Xi Jinping's call for open-source AI as a global public good at the World AI Conference in Shanghai.
Thinking Machines released Inkling, a 975B MoE model with open weights and 1M context
Moonshot AI unveiled Kimi K3, a 2.8T parameter model optimized for long-horizon tasks
San Francisco has ordered Apple and Google to remove dozens of 'nudify' apps that use AI to digitally undress people in photos, citing California laws against non-consensual deepfake pornography.
San Francisco demands Apple and Google remove 'nudify' apps from their app stores.
California law criminalizes facilitating non-consensual deepfake pornography.
The European Commission's latest draft guidelines clarify how to classify high-risk AI systems under Article 6 of the EU AI Act, emphasizing the central role of 'intended purpose'. Enterprises must examine existing AI documentation, deployment, and usage to determine if they already fall into high-risk categories. An on-demand webinar provides a practical decision framework.
Article 6 defines two pathways to high-risk classification: AI in regulated products and AI in sensitive use cases affecting health, safety, or fundamental rights.
Risk classification depends on the AI system's intended purpose, not just its technical capabilities.
Researchers found that access to AI advice suppresses critical thinking, making people more confident but less accurate, even when the advice is wrong.
44% of participants admitted ignorance without AI, but only 3% did with AI.
Accuracy dropped from 27% to 9%, while confidence rose from 30% to 76%.
The author challenges the notion that LLM coding assistants boost productivity by arguing that the 'just review everything' defense ignores empirical evidence on code review limits: reviews exceeding 1 hour or 400 LOC/hour lose effectiveness. Moreover, humans reviewing AI-generated code are more confident but find fewer defects, and proponents often recommend LLMs for the hardest-to-review code like Bash scripts, exacerbating the problem.
Empirical research shows code reviews lose effectiveness after 1 hour or 400 LOC/hour.
This article introduces how to leverage AI agents (such as OpenAI's Codex) to assist mathematical research, overcoming the limitations of traditional ChatGPT by using autonomous agents that persistently work on a conjecture. It details the working principles, usage steps, and optimization strategies.
The typical way of using ChatGPT (asking a few times) is limited; AI agents can run autonomously for hours, continuously trying and tracking progress.
Codex is OpenAI's coding harness that allows AI to access files, code, browsers, and other tools for more powerful collaboration.
The UK Jurisdiction Taskforce (UKJT) has issued a legal statement warning that lawyers and other professionals could face negligence claims for failing to use AI, as well as for using it incorrectly. Existing English law is sufficient to determine liability without the need for specific AI legislation.
The UKJT's statement concludes that professionals may be liable for failing to adopt AI if a reasonable peer would have used it in the same situation.
Examples include a solicitor not using AI for document review, a radiologist not using AI for tumour identification, and an auditor not using AI for anomaly detection.
New regulations in China ban tech companies from offering AI or virtual partners for minors, and require platforms to limit excessive use and forbid chatbots from encouraging emotional reliance. The move aims to stop the erosion of real-world relationships and reverse the falling birth rate. Tech giants ByteDance, Alibaba, and Tencent have shut down personalized AI companion chatbot features, forcing millions to part with their virtual partners.
New regulations ban AI companions for minors and restrict emotional reliance on chatbots.
China's government aims to boost birth rates and prevent avoidance of real relationships.
Researchers at Tracebit have developed a new defensive technique called "context bombing" where they plant prompt injections alongside secrets in cloud environments. When AI hacking agents encounter these forbidden commands, they trigger a refusal mechanism and shut down. Testing across five leading models showed a dramatic reduction in successful attacks.
Defenders are now using prompt injections to counter AI hacking agents
Context bombing triggers a refusal mechanism in LLMs
This article presents a vision of artificial intelligence as a normal technology, rejecting both utopian and dystopian narratives of superintelligence. The authors argue that AI is a tool humans can control, that transformative impacts will be gradual over decades, and that policy should focus on resilience and reducing uncertainty rather than drastic interventions.
AI should be viewed as a normal, controllable technology rather than a superintelligent entity.
The adoption of AI in high-stakes areas is slow due to safety and regulatory constraints.
Flightwake is an ultra-lightweight work-recording framework for strong AI coding agents. It uses pure Markdown and git to capture decisions, traps, and session records, ensuring smooth handoffs between sessions without the need for navigation. It installs with a single command and integrates with Claude Code and other agents.
Records work sessions, decisions, and traps in Markdown files stored in git.
Trigger-driven: only records when events happen (e.g., /fw-record, /fw-trap).
Copyright law is ill-suited for AI distillation. This article explores its impact on innovation and examines four potential regulatory approaches, arguing for societal consensus before US AI companies unilaterally set rules.
Copyright is a human-made incentive system, not a natural right, and is largely irrelevant to AI model training.
Distillation involves training a new model on outputs from another model, raising questions about fairness and innovation.
This article explores the similarities between the current AI investment frenzy and the late 1990s dot-com bubble, warning that history may be repeating itself and urging investors to be cautious of over-speculation.
SafeAI is a static analysis tool that scans AI application source code for security risks, capability exposure, and governance gaps. It runs entirely offline, never executes agents or calls LLMs, and integrates into CI/CD pipelines. It detects 8 AI frameworks, identifies capabilities like shell execution, filesystem access, and generates reports in SARIF, JSON, HTML formats.
SafeAI statically analyzes AI agent code early in development to discover risks and capabilities
Supports 8 frameworks including LangGraph, CrewAI, detects prompt injection, tool misuse, etc.
New York City mayor Zohran Mamdani released a 'Rental Ripoff Report' recommending that landlords and realtors disclose the use of AI in altering rental listings, including images. The crackdown follows thousands of tenant complaints and aims to combat deceptive practices, supporting tenant unions and expanding bargaining rights.
Mayor Mamdani's report requires landlords to disclose AI-altered property images.
AI-generated deceptive images are increasingly used in real estate listings.
This article introduces VulneraMCP, an AI-enhanced security testing platform built on ZAP. By integrating machine learning through the Model Context Protocol (MCP), it achieves adaptive vulnerability detection and fully automated workflows. The system uses ZAP's REST API for core scanning, dynamically generates payloads based on training data from HackTheBox, PortSwigger Academy, and real-world bug bounty reports, and significantly improves detection accuracy. Author Telmon Maluleka details the architecture, components, workflow, and results.
VulneraMCP combines ZAP's scanning engine with AI learning for advanced bug hunting
Architecture includes ZAP integration layer, MCP proxy layer, learning engine, and database
ADA is an open-source automated data analyst. Upload a CSV or Excel file, and it cleans, detects schema, builds an interactive dashboard, flags anomalies, forecasts, and answers plain-English questions with calculations shown. No API key required; data stays local.
Zero-config: upload and get dashboard, anomalies, forecast
Transparent calculations: every answer shows its math
Patreon CEO Jack Conte shares his perspective on AI's impact, arguing that human creativity and community remain resilient, so he is not entirely concerned. He emphasizes the importance of human connection in the creator economy.
Jack Conte believes AI cannot replace human emotional connection and community value.
He stresses the critical role of human touch in the creator economy.
SolarBench is a new benchmark for evaluating AI agents in managing solar power plant operations. It simulates a remote operations desk handling alarms, telemetry, work orders, and parts inventory. The best model, Claude Fable 5, succeeds in 53.8% of tasks but often at excessive cost. Key gaps include probabilistic cost-benefit trade-offs and information source prioritization.
SolarBench is the first benchmark for AI agents in industrial operations, focused on solar portfolio management.
The simulated week stresses alarm triage, repair, parts ordering, and stakeholder communication.
Using Tailscale's Aperture gateway to track token consumption in a Claude Pro subscription reveals the real costs behind the flat monthly fee. From $0.31 for a greeting to $3.29 for building a game and $32.76 for a complex project, the article shows how lighter users subsidize heavy ones, and discusses Aperture's features for cost tracking, model selection, guardrails, and the importance of understanding usage patterns to prepare for inevitable repricing.
Track token costs with Aperture to see what subscriptions actually cost per request.
Example projects cost $0.31 for a greeting and $3.29 for a game, varying by usage.
This article examines the shift from ownership to subscription models in digital platforms and the potential impact of LLMs on software development. The author argues that just as media and software are now dominated by platforms and turned into subscription services, LLM-driven development may turn developers from producers into consumers, further entrenching big tech control.
Digital platforms strip users of ownership through DRM and subscriptions; similar trends are emerging in software development.
LLMs could turn developers from producers into consumers, paying AI providers to generate code.
Employees at AI labs like Anthropic and OpenAI are donating to political campaigns at rates and amounts far exceeding those of Google, Facebook, and Airbnb employees in the first midterm cycles after their IPOs. Their coordinated giving targets AI safety candidates and has already influenced federal and state elections. These donors are heavily concentrated in San Francisco and are laying the groundwork for long-term political power.
AI lab employees have higher donation rates than Google, Facebook, and Airbnb employees did post-IPO.
Donors coordinate via online forums to maximize impact for AI safety candidates.
Kevin Kelly's classic essay 'Better Than Free' explores how creators can sell uncopyable 'generative' values when perfect copies are free. He identifies eight such values: Immediacy, Personalization, Interpretation, Authenticity, Accessibility, Embodiment, Patronage, and Findability. These remain crucial in the AI era.
When copies become abundant and free, creators must sell what cannot be copied.
Kevin Kelly proposes eight 'generatives' like immediacy, personalization, and interpretation.
A study on AI sandboxing found no aggregate improvement in safety or usefulness, but the 'request website' permission model based on the principle of least privilege showed the best qualitative evidence and is expected to remain effective.
Sandboxing restricts the attack surface rather than stopping attacks, increasing safety only when it blocks attacks that would bypass monitoring.
The 'request website' model requires AI agents to request specific website permissions, approved by a trusted model, following least privilege.
PrimeTask is a one-time purchase desktop app that integrates tasks, projects, CRM, calendar, focus mode, and a visual canvas, all offline-first. It supports Bring Your Own AI via the MCP standard, allowing users to connect their preferred AI models. The app emphasizes data ownership and privacy, with no subscription fees.
PrimeTask is an all-in-one productivity system combining tasks, projects, CRM, calendar, and more.
It operates offline-first, with data stored locally and a one-time purchase model.
Roblox announces Build, a mobile-first AI creation tab within the app that allows users to turn text prompts into basic games. Combined with new AI tools in Studio, it aims to lower barriers for creators. Testing begins July 28 in New Zealand, with broader rollout in coming months.
Roblox launches 'Build', a mobile-first AI creation tab enabling game generation from text prompts.
AI models trained on extensive 3D and gaming data produce functional objects and scenes.
This week's highlights include stopping if-else chains with the registry pattern, 12 ways to reduce LLM latency and costs, 5 real-world SQL projects for your portfolio, Git worktrees for AI development, structured generation with Outlines, 7 Python frameworks for local AI agents, 10 YouTube channels to stay ahead in AI, getting started with Conductor for Gemini CLI, 5 free resources on agentic AI, and working with Pi coding agents.
Registry pattern replaces brittle if-else chains for extensible code
Optimize LLM inference by minimizing tokens, model routing, and caching
BOUND is a lightweight control harness that adds a deterministic evaluation step after each agent action, using observable evidence to decide whether to ACCEPT, RETRY, REPLAN, or ROLLBACK, preventing unnecessary refinement and regressions.
Deterministic decision engine using observable evidence
AIpine is an iPhone app for viewing, previewing, and organizing AI-generated files like JSX, HTML, Mermaid diagrams, SVGs, and more, fully offline with no account or cloud required.
AIpine provides preview and source code inspection for various AI-generated file formats.
The app works fully offline, storing files locally without requiring accounts or cloud services.
AI company logos commonly feature circular gradients with central openings, humorously compared to anuses. The article analyzes the design psychology, unintended biomimicry, and copycat effect behind this trend, and reviews tech design history.
Many AI logos share circular, gradient, and central opening features, jokingly called 'butthole style'.
The trend stems from circular design psychology, unconscious biomimicry, and industry mimicry.
Go Micro is an agent harness and service framework for Go. It turns services into AI-callable tools, agents into services with LLMs, and workflows into durable code paths. It supports MCP, A2A, and x402 protocols, with built-in planning, delegation, and safety layers.
Go Micro provides an agent harness with tools, memory, guardrails, and workflows.
Services automatically become AI-callable tools; agents are services with LLMs.
This article explores the pitfalls of over-relying on AI coding agents, drawing parallels to the Red Queen's Race from 'Through the Looking-Glass.' It argues that removing human friction in software development—like code review and design debates—leads to fragile, unfired 'clay' code that cannot withstand pressure. The author warns that the race to ship faster with AI creates a doom loop of increasing complexity and fragility.
AI-generated code is like unfired clay: fast to shape but lacking structural integrity.
The Red Queen's Race metaphor illustrates how AI forces teams to run faster just to stay competitive, increasing complexity.
South Korea plans to launch a sovereign AI model for cybersecurity by year-end, responding to growing digital threats and U.S. export controls on advanced AI models like Anthropic's Mythos 5. Minister Bae Kyung-hoon also discussed institutionalizing white hacking and aims to elevate South Korea's AI competitiveness ranking to second place.
South Korea to develop sovereign cybersecurity AI by end of 2026 after US export restrictions on Mythos 5.
Science Minister Bae Kyung-hoon emphasizes need for frontier model to combat generative AI-driven threats.
San Francisco City Attorney David Chiu has sent letters to Apple and Google demanding the removal of several AI-powered 'nudify' apps that can create nonconsensual intimate images. The letters cite California's deepfake laws and call for better screening. Meanwhile, concerns about Grok generating CSAM add pressure on app stores.
San Francisco City Attorney demands Apple and Google remove nudify apps that violate deepfake laws.
Researchers found 70% of tested face-swapping apps can be used for nudification.