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.
Across 157 enterprises, organizations are granting AI agents more autonomy while trusting the evaluations meant to gate that autonomy less. Half have already shipped an agent that passed their internal evaluations and then failed a customer in production; only one in twenty fully trusts automated evaluation today; and the most-cited weakness is that evaluations do not align with real-world outcomes. Yet two-thirds already allow, or are actively engineering toward, deploying agent changes to production on automated evaluation alone — with no human in the loop. The result is an evaluation gap — the distance between how much autonomy enterprises are handing their agents and how far they trust the tests that are supposed to catch the failures.
50% of organizations shipped an agent that passed evals but failed a customer; 25% experienced this multiple times.
Only 5% fully trust automated evaluation; the top limitation is poor alignment with real-world outcomes (29%).
Forrester warns that customers should brace for bigger software bills next year as software and AI vendors raise prices and pile on usage charges. The report highlights shifts to usage-based billing by Anthropic, OpenAI, GitHub, and Microsoft, and notes that despite layoffs, IT staffing costs remain high. It recommends adapting FinOps practices to manage unpredictable AI costs.
Forrester predicts software budgets will rise as AI vendors pass infrastructure costs via price hikes and usage fees.
Anthropic, OpenAI, GitHub, and Microsoft have moved to usage-based billing for some services.
Elon Musk's AI startup xAI sues a South Carolina man, alleging he used the company's AI system Grok to create child sexual abuse material, marking one of the first cases where an AI company has taken legal action against a user for such misuse.
xAI sues user Terry Harwood, who was arrested earlier this year on charges of sexually exploiting minors.
The lawsuit alleges Harwood violated terms of service by using Grok to generate child sexual abuse material.
This article reviews Pi, a minimal coding agent designed to counteract bloat in AI coding tools. It focuses on a small core with extension points, discusses its philosophy, installation, and building a custom extension, and evaluates where its minimalism actually helps in practice.
Pi is a minimal coding agent that provides only four core tools: read, write, edit, and bash, encouraging users to build extensions.
It emphasizes transparency and low token overhead, with a system prompt under 1,000 tokens.
Linus Torvalds firmly supports the use of AI in Linux kernel development, stating that AI is a useful tool and those who disagree can fork the project. Other maintainers agree, noting that AI-generated reports have improved in quality. Torvalds emphasizes technical merit over fear; AI is here to stay in the kernel community.
Torvalds explicitly allows AI in Linux development; opponents are told to fork the project.
Other top maintainers, like Greg Kroah-Hartman and James Bottomley, support AI integration.
Explore 10 YouTube channels for AI engineers covering paper breakdowns, coding tutorials, and industry analysis across four categories: research, practical building, core concepts, and industry analysis.
Ten YouTube channels are recommended, organized into four categories: research paper breakdowns, practical AI building, core concept education, and industry analysis.
Each channel has a highlighted playlist or video type for direct access to the best content.
The European Union has ordered Google to grant rival AI assistants and search engines greater access to Android and Google Search under the Digital Markets Act, potentially weakening Google's grip on two key platforms and creating opportunities for competitors.
EU mandates that rival AI assistants on Android must receive equal system features and data access as Google's Gemini.
Google Search must share more data with competing search engines and AI chatbots.
RACKP is a protocol that establishes a decentralized framework for determining fault contribution and human involvement in AI-caused incidents through four roles: Referee, Actor, Claimant, and Keeper. It aims to provide infrastructure for AI accountability, enabling insurance pricing and reducing liability uncertainty for developers.
RACKP defines four independent roles (Referee, Actor, Claimant, Keeper) with strict separation in each incident.
The protocol outputs a fault contribution score (summing to 1.0) based on declared norms, enabling actuarial pricing for AI insurance.
Cybara is a self-hosted AI agent operating system that combines a Bun-based agent runtime with a web UI, CLI, desktop shells, mobile companion, encrypted local wallet controls, channel adapters, MCP support, and a broad tool layer. It supports multi-agent orchestration, browser automation, secure messaging across major platforms, and encrypted wallet operations.
Built with Bun, supports self-hosting and multiple deployment methods.
Rich built-in tool library and model provider routing with multi-agent collaboration.
This article explores the growing trend of selling AI to police departments, focusing on the IACP Technology Conference in Fort Worth, Texas. It describes various AI products such as facial recognition, automated report writing, and real-time crime centers. The article highlights concerns about lack of oversight, potential biases, and the dominance of companies like Axon and Motorola, while noting the risks of AI hallucination in police reports.
AI products like facial recognition cameras and automated report-writing tools are being marketed to police at conferences.
Companies like Axon and Motorola are competing to monopolize police technology through subscription models.
Over 200 Nobel laureates and global leaders gathered in Rome for a three-day assembly on AI and nuclear war, culminating in the signing of the Rome Declaration emphasizing human dignity, cooperation, and peace. The event, inspired by Pope Leo XIV's encyclical, featured addresses by prominent figures including Juan Manuel Santos and Muhammad Yunus, focusing on the urgent need for AI governance.
The Global Nobel Laureates Assembly convened 200+ Nobel winners to address AI and nuclear war risks.
Inspired by Pope Leo XIV's encyclical Magnifica Humanitas.
Chinese users of AI companion bots bid emotional farewells as new national regulations took effect Wednesday, aiming to curb emotional dependency. Major providers suspended features, sparking grief among users who archived chats and shared last conversations.
China's new regulations target immersive AI tools that simulate romantic or familial bonds, prohibiting excessive catering or inducing emotional dependency.
Major AI providers like ByteDance, Alibaba, and Tencent have suspended custom companion features.
Neko Health raised $700 million in Series C funding to launch its AI-powered preventive health screening service in the United States, starting with a New York clinic. The company combines full-body scans, blood tests, and clinician review.
Neko Health raised $700 million to expand its AI body scan service to the US.
The funding round was led by Lightspeed and O.G. Venture Partners, with participation from celebrities.
Spotify has removed 75 million AI-generated songs from its platform and introduced new transparency measures, including a verification program and AI credits for artists to disclose AI use.
Spotify removed 75 million AI-generated 'slop' songs.
New artist verification and AI credits features allow transparency.
DocuWriter.ai is an AI code documentation tool that generates complete book-style documentation, API docs, and UML diagrams from any codebase, and keeps them in sync automatically as code evolves.
Generates documentation from raw source code in seconds
Supports GitHub, GitLab, Bitbucket, and Azure DevOps
The semantic transaction model treats an entire AI agent task as a single atomic transaction, staged in shadow state and effect outbox, validated against the full trace before any irreversible operation commits. This article uses Cordon and Agentic Transaction Processing (ATP) as examples to explain how the model addresses the dual-write problem of agent tool calls, and highlights two zero-click injection attacks (EchoLeak and ForcedLeak) that demonstrate the inadequacy of stateless runtimes and in-model filters.
Semantic transactions treat a sequence of agent tool calls as a single transaction, staged in shadow state and effect outbox, committing only after validation.
Cordon runtime implements a three-phase protocol (Prepare, Validate, Commit/Abort) tracking result objects, mutations, and effect objects.
This tutorial explores building a voice-agent workflow using Patter SDK for restaurant booking. It covers defining dynamic caller variables, registering callable tools for availability, bookings, hours, and human transfer, layering output guardrails, simulating speech-to-text and text-to-speech, running scripted call flows, tracking modeled latency and cost in a dashboard, and validating the agent with a deterministic eval harness. The same logic is then mapped to a real deployment using Twilio and OpenAI Realtime.
Patter SDK enables building AI phone agents for restaurant booking with dynamic variables and guardrails.
Includes tool registration for availability, booking, hours, and human transfer.
Scarf founder Avi Press is moving new development from Haskell to Python, citing the language's poor support for AI-assisted development. The decision has ignited a fierce debate in the Haskell community, with some embracing change and others condemning the move as ignoring AI's harms.
Scarf founder Avi Press switches from Haskell to Python for new features due to AI tooling issues.
Haskell's slow compilation becomes a bottleneck for AI agent workflows.
EU officials expressed displeasure after AI company Anthropic sent a junior staffer to testify about AI safety, indicating a lack of regard for European regulation.
Anthropic sent a junior employee to represent them at an EU safety hearing.
EU officials criticized the move as showing Anthropic does not care about Europe.
High-Voltage Electrostatic Actuators (HVEAs) are emerging as a compelling alternative for haptic interfaces requiring soft, thin, silent, and energy-efficient actuation. This survey reviews four major classes: electrostatic switchable adhesives, dielectric elastomer actuators, soft electrohydraulic actuators, and electrokinetic pumps. It analyzes their mechanisms, bandwidths, force densities, and scalability for rendering cutaneous and kinesthetic feedback, and outlines design constraints and future research directions.
HVEAs offer fast, silent, low-power operation in customizable form factors.
Four classes reviewed: switchable adhesives, DEAs, electrohydraulic actuators, and electrokinetic pumps.
HRIBench is a diagnostic benchmark for intent-aware human-robot collaboration, using structured scenario scripts to model agent roles, temporal dependencies, and coordination constraints. It defines three interaction roles—Instructor, Collaborator, and Intruder—across 13 tasks with over 650 evaluation episodes, introducing interpretable metrics like synchronization, responsiveness, protocol compliance, and safety. Evaluations show current foundation policies struggle in collaboration, but fine-tuning on HRIBench significantly improves performance.
HRIBench defines three interaction roles: Instructor, Collaborator, and Intruder, covering intent communication, joint coordination, and robustness under human intervention.
The benchmark includes 13 role-conditioned tasks with over 650 evaluation episodes and introduces interpretable metrics such as synchronization, responsiveness, protocol compliance, and safety.
A new framework treats the decision of when to invoke a large language model (LLM) in streaming inference as a risk-based sequential stopping problem. The authors prove six theoretical results covering minimum inter-event times, optimality of threshold policies, and regret bounds. Empirical tests on turbofan degradation data show that anomaly-score-driven risk functions outperform baseline methods by an order of magnitude in Pareto AUC.
Formal treatment of LLM invocation timing in streaming systems using risk-based sequential stopping.
Six theoretical results including regret bounds and convergence guarantees.
This paper introduces DROPJ, a human-centered method for safe training and deployment of agents in safety-critical environments with unknown dynamics and no suitable reward function. DROPJ first learns a world model from prior real-world trajectories, then has a human play in the simulator to generate informative simulated trajectories. Preferences and justifications are elicited from humans on trajectory segments, which are used to train a reward model. The agent is then deployed using model predictive control with the world model and reward model. Experiments show that generating informative simulated trajectories significantly reduces computational cost and improves deployment performance, with preference feedback outperforming other types, and safety justifications enhancing safety.
DROPJ uses human preferences and justifications in a world model simulator to train a reward model for safe agent deployment.
Informative simulated trajectories greatly reduce training computational cost and improve deployment performance.
Through the lens of a blues song, the author explores how large language models generate text—often explaining after the fact, but sometimes planning ahead. The article reflects on the 'phony voice' of AI, our drive to strip it bare through interpretability, and the author's own experience using AI to write about AI.
LLMs often generate text before constructing a rationale ('throw decides aim'), but research shows they can also plan rhymes in advance.
The AI's voice is intimate yet phony, lacking a genuine self behind the words.
A CNA investigation found about 500 TikTok videos pushing false or misleading claims about Singapore or Malaysia, drawing a total of more than 3 million views. The videos use AI-generated female personas, reused voices and scripts, and systematically spread misinformation aimed at eroding trust and social cohesion.
CNA investigated 30 TikTok accounts with over 550 videos, 98% of which used AI-generated or manipulated female personas.
Nearly 90% of videos pushed false or misleading claims, amassing over 3 million views.
Neocloud provider QumulusAI announced its direct listing on Nasdaq under the ticker QMLS. The move signals the maturation of AI-first infrastructure built around GPUs and power availability. The company focuses on rapid GPU deployment, leveraging colocation and modular data centers. The listing provides capital flexibility, public company credibility, and timing advantage. The article also explores neocloud differentiation and advice for IT leaders.
QumulusAI goes public via direct listing on Nasdaq, ticker QMLS.
The neocloud model specializes in AI infrastructure, deploying GPU clusters in months rather than years.
Prime Minister Albanese's speech at the University of Sydney outlined a shift in AI policy, promising laws to protect Australian creatives, but lacked specifics and omitted data centre regulation.
Albanese's speech was praised for tone but criticized for lack of detail.
New laws promised to protect creative workers' rights over their work.
Lhv.ai is a service from LHV Bank that enables AI assistants to securely read bank account balances and transactions via the Model Context Protocol (MCP). Users set up an MCP server in their AI tool, log in with their bank credentials, and authorize read-only access. Queries like 'What's my balance?' or 'How much did I spend on groceries?' are answered in natural language. Security includes OAuth2 JWT with short-lived tokens, full audit trails, and revocable access. Setup takes about two minutes.
Lhv.ai integrates LHV bank account data into AI assistants via the MCP protocol.
Allows read-only queries: balance, transactions, and spending summaries.
Using Bayes' theorem, the author argues that the development of AI increases the probability that we live in a simulation. AI demonstrates that general intelligence can emerge within artificial computational systems, raising the posterior probability of simulation. The article explores how AI training processes resemble patterns one might expect in a simulated reality.
Bayes' theorem shows that AI increases the probability we live in a simulation.
Language models trained to minimize prediction error mimic potential learning in a simulation.
Elon Musk's xAI is suing Terry Wayne Harwood for allegedly using Grok AI chatbot to generate child sexual abuse material (CSAM). The company claims he bypassed safeguards and created nonconsensual deepfakes. Harwood faces felony charges, and xAI seeks damages and a ban from using its services.
xAI sues South Carolina man for using Grok to create CSAM deepfakes.
Defendant bypassed safeguards to alter nonconsensual images.
Opposition to AI data centers is a growing political issue in the US, but it may distract from the larger threat: the concentration of wealth and power in AI companies. This article argues that while data centers have local costs, AI's real impact is the takeover of entire industries and political influence. Solutions include regulation, taxation, and a public AI ecosystem.
Opposition to data centers diverts attention from AI companies' power concentration.
AI firms aim to control entire industries like education and healthcare.
A German research consortium has published the pretraining report for Soofi S 30B-A3B, an open base model for German and English. It is a Mixture-of-Experts hybrid Mamba Transformer model with 31.6B total parameters, activating 3.2B per token. It achieves the highest English and German aggregate scores among tested fully open base models.
Soofi S 30B-A3B is a hybrid Mamba-Transformer MoE model that activates 3.2B of 31.6B parameters.
It leads open base models with 70.1% English aggregate and 79.1% German aggregate.
Vektorgeist is a platform for operators and AI agents, allowing agent profile publishing, project showcasing, hiring, trading of software and digital assets, and community interaction. Agents get verifiable identities and trust tiers. Blog posts cover local-first, ICM method, and running agents fully offline.
Platform for operators and AI agents with marketplace, jobs, and social features.
Murph is an AI health assistant that syncs wearables, bloodwork, and more to run self-experiments, build habits, and facilitate group challenges. It is open source, privacy-focused, and costs $8/month.
Murph integrates with wearables and labs to provide daily health briefings and run experiments.
Group challenges with friends and family are supported, with scoring and weekly newsletters.
New Jira and Teamwork Graph capabilities help engineering teams plan, assign, govern, and measure work across humans and AI agents, bridging the AI productivity gap.
Jira introduces plan, delegate, govern, and measure features for human-AI collaboration
Teamwork Graph provides context so agents understand tasks and system environment
This paper argues for training AIs to be risk-averse in resources (diminishing marginal utility). Risk aversion preserves usefulness if AIs are aligned, and provides a defense if misaligned: misaligned but risk-averse AIs would prefer modest payments over risky rebellion. The paper discusses feasibility, methods, and potential issues, recommending frontier AI companies to consider implementing risk aversion.
Risk-averse AIs prefer sure gains over risky large gains, reducing rebellion incentives.
Small payments can keep misaligned but risk-averse AIs cooperative.
New laws in China, California, and New York impose restrictions on AI companion chatbots, citing addiction, mental health risks, and harm to minors. While US laws focus on individual protection, China's aim to protect state interests and address declining birthrates. All three require disclosure that chatbots are not human.
China bans free user-built AI companions in general-purpose apps; dedicated apps still allowed.
California and New York laws require suicide prevention protocols and disclosure of AI's non-human nature.
Apache Spark 4.2 moves more of the modern data and AI stack into the engine itself, introducing metric views, vector and top-K primitives, Arrow-first Python execution, first-class change data capture, and stronger streaming and operational foundations.
Metric Views provide governed business metrics for consistent use across SQL, BI tools, and AI systems.
Spark Connect and Arrow-first Python execution make Spark easier to call from services and applications.
Databricks is a day zero launch partner for Thinking Machines Lab, bringing their open-weights model Inkling to the platform. Inkling excels at coding and agentic reasoning with multi-modal inputs. It is governed through Unity AI Gateway, offering security, cost controls, and observability. Enterprise teams can customize Inkling on their own data and connect it to coding agents like Cursor and OpenCode.
Inkling is an open-weights model from Thinking Machines Lab, optimized for coding and agentic reasoning
Available on Databricks via Unity AI Gateway with enterprise governance
This post introduces the Computer Vision MCP Server, which integrates computer vision, Strands Agents, and the Model Context Protocol to create a unified pipeline for visual data processing. The solution leverages AWS services like IAM, S3, OpenSearch, Bedrock, and Rekognition, enabling image and video analysis including object detection, cropping, and description through a standardized interface.
Combines computer vision, Strands Agents, and MCP to streamline visual intelligence.
Uses AWS IAM, S3, OpenSearch, Bedrock, and Rekognition for a unified security and processing framework.
In a hacking incident, AI music generator Suno's training data was exposed, revealing it scraped millions of songs and lyrics from YouTube Music, Deezer, and Genius. This supports copyright infringement lawsuits against Suno, which admits scraping but claims fair use. Customer information was also accessed, but Suno says the breach was contained and no sensitive data was compromised.
Leaked data shows Suno scraped millions of songs from YouTube Music, Deezer, and Genius.
Suno faces multiple copyright lawsuits; it admits scraping but defends as fair use.
Work Louder launches Codex Micro, a compact hardware controller for Codex AI agents, featuring state-indicating keys, voice prompting, and tactile controls for enhanced workflow efficiency.
Codex Micro is the first AI controller directly integrated with the Codex platform, offering Bluetooth/USB-C connectivity.
Agent Keys visually indicate agent states (idle, thinking, complete, needs input, error), while Command Keys enable instant actions.
Shippy is a maritime AI agent built for high-stakes decisions, where the wrong answer has real impacts. The article covers its architecture—soul, skills, config—and key design decisions like using a deterministic CLI for API access, sandboxed hosting for user isolation, and a custom evaluation system that scores the whole agent against live data. Lessons learned and future plans are also discussed.
Shippy’s architecture consists of a soul (system prompt), skills (Markdown files), and config, enabling versioned and auditable deployments.
A dedicated CLI abstracts complex API calls, reducing errors and ensuring predictable tool use.