Shamil Khan

Meta is accelerating plans to replace human content reviewers with AI, targeting 90% automation by year-end. A recent Instagram breach that compromised 20,000 accounts highlights the security risks of this rapid shift.

China's 360 and Japan's Sakana AI have launched Tulongfeng and Fugu, directly challenging Anthropic's Mythos and Fable 5. The US export ban on these models is accelerating regional AI development and threatening American dominance.

OpenAI launched GPT-5.6 Sol, Terra, and Luna, but a U.S. government request has indefinitely delayed public access. Only pre-approved customers can use the models, sparking debate over AI regulation and innovation.

A Chinese startup’s AI-powered laser mosquito zapper raised $2.7 million on Indiegogo but now faces production delays due to sensor calibration and safety certifications. The device uses lidar and AI vision to kill mosquitoes mid-flight.

The Trump administration is nearing a decision to let Anthropic restore its Fable 5 AI model after a sudden export control ban. The move follows a partial reversal for the related Mythos 5 model, but questions remain about new restrictions and national security implications.

Hong Kong's Correctional Services Department removed an AI-generated K-pop video meant to warn youth about drugs after it was criticized for glamorizing substance use. The incident raises questions about using generative AI for sensitive public messaging.

Google published a practical guide detailing 11 ways Gemini can streamline work inside Workspace. The tactics range from prompt engineering rules to AI-generated presentations and meeting summaries, aiming to cut repetitive tasks and sharpen collaboration.

OpenAI postpones GPT‑5.6 public rollout under U.S. pressure, restricting initial access to vetted partners. The move reflects growing Washington concern over frontier AI risks and a new executive order framework.

Anthropic restores access to its Claude Mythos 5 model for over 100 approved US organizations after the government eased a national security suspension. The opaque vetting process draws fire from critics, while Fable 5’s return remains uncertain.

Washington partially reversed its suspension of Anthropic's Claude Mythos 5, granting access to over 100 trusted US firms. The move follows a two-week ban over fears the AI could aid cyberattacks by China and Russia.

A new technical walkthrough shows how to transform a locally hosted LLM into a tool-using research agent. The setup combines Gemma 4, Ollama, OpenAI Agents SDK, and Tavily MCP for web search, enabling grounded answers with citations.

OpenAI has postponed the public launch of GPT‑5.6 after the U.S. government requested early access to assess national security risks. The model will first be available only to a select group of vetted partners, marking a new phase in AI oversight.

AWS’s Jeff Barr says developers are moving from writing code to orchestrating AI agents, a shift showcased by Indian winners of the 10,000 AIdeas Competition. Their production-ready apps highlight a new era of spec-driven development.

China has released its first national standard for AI agent connectivity, introducing mandatory digital ID cards for autonomous agents. The framework aims to slash development costs and secure cross-domain interactions, but raises questions about global interoperability and centralized control.

Ukraine is building its own AI computing infrastructure with Kyivstar to secure military and civilian data. The project, backed by VEON, starts with 3-5 MW and aims to reduce reliance on foreign tech.

OpenAI plans a limited ChatGPT 5.6 rollout with access restricted to US government-approved customers, according to an internal memo. The move follows a federal executive order encouraging voluntary AI review, but recent actions by Anthropic suggest oversight may become more active.

Zhipu AI’s GLM-5.2 is being called a new ‘DeepSeek moment’ as the first open-weight model reliable enough for daily coding. Its cost-effectiveness and direct performance are winning praise from top tech leaders, challenging proprietary giants like GPT-5.5.

A controversial video of Punjab CM Bhagwant Mann has ignited a debate on AI misuse in politics. As parties deploy deepfakes, the line between real and fake blurs, threatening electoral integrity ahead of the 2027 Assembly elections.

At ISTE 2026, Google launched a Classroom app in Gemini, teacher-led AI activities, and Study notebooks to personalize learning. The move aims to save time and boost equity, but details on privacy and research are scarce.

Italy's Domyn will release a fully open-source AI model with over 400 billion parameters within a year. Backed by EU funding and EuroHPC access, it aims to reduce Europe's reliance on foreign AI systems.

China must develop its own version of Anthropic's Mythos AI to counter cybersecurity risks, says 360 founder Zhou Hongyi. He likens the US model to a 'cyber nuclear weapon' and calls for reciprocal deterrence.

A new retrieval pattern lets a single LLM call rank document candidates with explicit reasons, replacing score fusion. The arbiter preserves detection signals, handles conflicts, and outputs a typed JSON for generation—all while building a defensible audit trail.

Fysics AI, a Shanghai start-up, has launched Fysiverse, a world model that embeds physical laws directly into its code. The move challenges US data-driven paradigms and could reshape robotics and autonomous driving simulation.

Chinese AI startup Z.ai released its GLM-5.2 model, nearly matching Anthropic and OpenAI on benchmarks, just after Anthropic disabled global access. The open-source model excels in coding and agent tasks at a fraction of the cost, fueling a 2,000% stock surge and dual-listing plans.

Facebook reintroduces Creator Studio as an AI-powered companion app offering personalized recommendations and a dedicated workspace. The platform also splits its Professional Dashboard into Creator and Business dashboards for tailored experiences.

MIT and Microsoft researchers built Murakkab, a platform that lets developers describe AI tasks in plain language and automatically optimizes the entire agentic workflow. It cuts computational needs to 35%, energy to 27%, and costs to under 25% versus traditional methods.

OpenAI and Broadcom have launched Jalapeño, a custom AI chip designed from scratch for LLM inference. Co-developed in just nine months, it promises substantial performance-per-watt gains and will power gigawatt-scale data centers starting in 2026, marking a major step in OpenAI's full-stack infrastructure strategy.

Anthropic's Mythos AI model reportedly found vulnerabilities in nearly all classified US systems in hours during Project Glasswing. The revelation fuels debate over AI's role in national security and government restrictions.

ByteDance revealed Seedance 2.5 at Volcano Engine's FORCE conference. The AI video model generates seamless 30-second clips and accepts up to 50 inputs, launching in July with competitive pricing.

Google is rolling out AI improvements for Nest cameras that recognize people without clear facial data and add audio context to event summaries. The updates also bring HVAC health alerts and wider Matter support.