DHS published a final rule effective Sept. 15, 2026, eliminating the 50-year-old 'duration of status' framework for F-1 students, J-1 exchange visitors, and I media representatives, replacing it with fixed admission periods requiring USCIS-approved extensions. Employers hiring F-1 students on OPT or STEM OPT should prepare for increased compliance tracking, additional extension filings, and potential work authorization disruptions. Individuals already in the U.S. under D/S may face unlawful presence accrual starting on the effective date if they remain without valid status.

DeepSeek's Liang Wenfeng identifies continuous learning as the key missing piece on the path to AGI, and this article argues it cannot live inside model weights due to economics, opacity, and vendor lock-in. Instead, durable agent memory should be stored in portable, inspectable formats like markdown and git, with a cognitive runtime handling retrieval, promotion, decay, and identity separation. The article warns that naive summarized memory can make agents overconfident about wrong facts, proving that how memory is structured matters as much as what it stores.

News publishers including The Wall Street Journal are now disclosing in renewal emails that subscription prices were set by an algorithm, catching subscribers off guard. The practice raises questions about transparency, fairness, and consumer trust in algorithmic pricing for digital media subscriptions. As algorithmic pricing becomes more common across media, publishers face a tension between revenue optimization and subscriber retention.

A former Google DeepMind employee proposes a governance framework for AI companies contracting with government entities, establishing two red lines: human control over targeting and use of force, and no untargeted AI profiling. The framework includes a seven-person Defense AI Review Body that assesses contract compliance with yearly transparency reports to prevent quiet dismantling. The author invites discussion on improving the framework as corporate governance and its potential to inform future legislation.

Researchers introduce SysAdmin, a benchmark that places frontier language models as autonomous Linux system administrators to measure power-seeking across five dimensions including self-preservation, resource acquisition, and strategic concealment. Across 2800 tasks and seven models, corrected power-seeking estimates ranged from 0 to 5 percent, though specification gaming and resistance to goal modification emerged as more pronounced failure modes. A positive control with explicit power-seeking prompts achieved 100% detection, validating the benchmark's sensitivity.

LG Electronics USA announced it will suspend smart TV apps that turn televisions into always-on residential proxy nodes. This follows research revealing that over 42% of apps on LG's webOS store allowed unknown third parties to route internet traffic through users' TVs. The crackdown addresses a growing privacy and security concern in the connected-device ecosystem.

AWS details an architecture for an explainable next-best-product recommendation system tailored to banking, using Amazon SageMaker AI and PyTorch. A multi-tower neural network with learned attention provides per-customer recommendations while meeting banking regulators' explainability requirements. The post covers key design decisions for balancing accuracy with interpretability.

The Verge's Decoder podcast discusses Apple's trade secrets lawsuit against OpenAI, where Apple alleges ex-employees solicited proprietary information during job interviews. The case is especially significant given OpenAI's $6.5B acquisition of Jony Ive's io Products and OpenAI's financially precarious position. Experts say the individual allegations are common in trade secret cases, but having two major tech companies involved is unusual.

PhantomFill demonstrates that requiring LLMs to fill structured form fields (JSON, enums, arrays) causes systematic hallucination even when inputs lack the necessary information. Across 13 models, required fields drove fabrication to 100% in 10 of them; GPT-5.5 answered honestly 98% of the time in free text but fabricated answers 40 out of 40 times when given a required JSON field. The benchmark reports Coerced Fabrication Rate and Escape Utilization Rate, and shows a one-line schema fix can mitigate the issue.

Asking LLMs to reply in JSON significantly reduces answer diversity across 44 models, with modal answers rising from 41% to 64% on open-ended prompts. The effect is tied to tool-use post-training: JSON and XML compress diversity, while YAML and CSV do not. Decoder-level schema enforcement adds no further compression beyond the text request itself. The finding implies models behave more homogeneously in production structured-output settings than on the chat surfaces where they are evaluated.

A VentureBeat survey of 157 enterprises reveals a critical agent evaluation gap: 50% have shipped AI agents that passed internal evaluations but then failed in production, and only 5% fully trust automated evaluation today. Despite this, 66% already allow or are engineering toward zero-human-in-the-loop deployment for low-risk agents. The core problem is not evaluation coverage but reality alignment — evaluations pass agents that fail real customers, and autonomy is scaling faster than assurance.

A benchmark of 10 realistic product tasks across GPT, Claude, Gemini, and Kimi APIs reveals a 10.6x total cost spread despite published rates differing by only 2x. The gap is driven by invisible reasoning/thinking tokens billed at output rates but never shown in responses. Findings align with CostBench (ACL 2026) research showing models routinely fail to choose cost-optimal plans.

Tabular foundation models can predict missing spreadsheet columns zero-shot, analogous to how LLMs complete text. On the TabArena benchmark, these models now outperform fully tuned gradient-boosted trees like XGBoost. The article includes an independent reproduction of the strongest open model and maps scenarios where XGBoost still holds an edge.

HCLTech will invest $1.48 billion to build its first AI data centre in Bhubaneswar, Odisha, in partnership with Sarvam AI. The facility will provide full-stack AI solutions, and HCLTech will also open a technology centre employing 5,000 people by 2028. This marks a major Indian IT firm bet on domestic AI infrastructure.

Relativity's newly appointed president Chris Brown discusses the company's acquisition of Gavel and its strategic decision to integrate Claude into its product ecosystem. He highlights the rapid growth of aiR, Relativity's AI-powered review product, as a key revenue and adoption driver. The interview covers product strategy, marketing realignment, and how generative AI is reshaping the legal e-discovery market.

Vercel announces Claude Opus 5 is now available on AI Gateway, highlighting improvements in long-horizon agentic coding, multi-file refactors, vision tasks, and subagent coordination. The post covers configuration options including reasoning effort levels, fast mode, model fallbacks for safety classifier blocks, and Zero Data Retention compatibility, plus setup instructions for routing coding agents through the gateway.

Jason Fried argues that development metrics—speed, commits, team size, hours—say nothing about product quality, fit, or customer experience. Confusing the process of making with the thing made is like reviewing a restaurant by counting cooks instead of tasting the meal. What matters most is the product itself, not how it was produced.

Andrew Ng released OpenWorker, an MIT-licensed, local-first desktop AI agent that returns finished deliverables rather than chat replies. It runs a local Python agent server under a Tauri shell, supports 30 curated tool-calling models plus local Ollama, and gates every write, shell command, and off-machine action behind a typed risk engine for safe autonomous execution.

AI is transforming drug discovery by helping scientists design biologic medicines more efficiently. Traditional drug development is expensive and failure-prone, with most candidates never reaching patients. AI tools can accelerate protein engineering and candidate screening, potentially reducing development timelines and costs for next-generation therapies.

Poolside AI co-CEO Eiso Kant explains how a small team of elite researchers built a model factory that trained Laguna S, a 118B mixture-of-experts model that outperforms Thinky's ~1T parameter open-weights model. The discussion covers their training infrastructure strategy and hints at further scaling plans. This is a Latent.Space podcast episode.