Live wire Humanoids · Autonomous vehicles · Drones · Industrial robotics · Embodied AI research

Blog

  • The Chatbot That Foretold Why People Share Secrets With ChatGPT

    The Chatbot That Foretold Why People Share Secrets With ChatGPT

    In the 1960s an MIT professor named Joseph Weizenbaum created a chatbot called ELIZA. The conversations people had with it set precedents for the chatbots to come.

  • Synthetic Sincerity review – Marc Isaacs’ AI interrogation grapples with identity and existence

    Synthetic Sincerity review – Marc Isaacs’ AI interrogation grapples with identity and existence

    A combination of fact and fiction leaves the celebrated documentarian’s puzzling project about software training wanting for depth

    Marc Isaacs’ new film is a curious, intriguing, semi-sincere affair that I couldn’t make friends with. It is an odd, shallow piece of work about artificial intelligence that is itself exasperatingly artificial, a self-aware docudrama hybrid. Isaacs is, or rather pretends to be, licensing the vivid characters from his previous, acclaimed documentaries to a fictional AI research lab called Synthetic Sincerity at the fictional University of Southern England, so that the lab’s software can be “trained” in the creation of AI human figures on screen.

    The lab’s research staff are played by actors, or at any rate people acting; these include Lebanese independent film-maker Lynn El Safah. Isaacs has amusing scripted conversations about this project with a disapproving AI avatar on screen, like Max Headroom of old, whose face is digitally modelled on Romanian actor Ilinca Manolache, from Radu Jude’s Do Not Expect Too Much from the End of the World. The film, however, does not show the process by which Manolache was approached and her face transformed into an AI figure.

    Continue reading…

  • Ant Group Advances AI Agent Security

    Ant Group Advances AI Agent Security

    Ant Group’s AI Security Lab today announced the open-source release of SingGuard-NSFA, a specialized security guardrail framework designed specifically for autonomous AI agents. The framework secures agentic AI systems against operational threats like prompt injection, addressing critical vulnerabilities as AI transitions from passive content generation to active, autonomous execution.

    As AI agents rapidly move from research labs to business scenarios, the security landscape has fundamentally shifted. The explosive global adoption of open-source agent frameworks like OpenClaw, celebrated for their “one-click deployment” and “full-stack autonomy”, has simultaneously exposed significant operational risks, including permission escalation and prompt injection.

    Industry frameworks, including the OWASP (Open Web Application Security Project) Top 10 for Agentic Applications 2026, have formally categorized threats such as goal hijacking, tool misuse, malicious code execution, and identity and privilege abuse as critical vulnerabilities. These behavioral threats represent blind spots that traditional security technologies struggle to address.

    SingGuard-NSFA is designed specifically to tackle these gaps. It provides a robust security layer that intercepts malicious requests and validates responses before autonomous actions are executed, ensuring safe deployment in complex operational environments.

    The framework introduces a systematic defense mechanism grounded in industry-standard security principles. It categorizes agent-specific risks into a comprehensive taxonomy encompassing 185 distinct operational threat scenarios spanning 7 categories. To facilitate validation and continuous iteration, Ant Group’s AI Security Lab constructed a rigorous benchmark suite covering 133 languages with nearly 100,000 test samples.

    In benchmark evaluations, SingGuard-NSFA demonstrates state-of-the-art performance. The compact 0.8B parameter model rivals the capabilities of competing 8B models, while the 9B variant achieves real-time detection latency of approximately 50 milliseconds, delivering both speed and accuracy for production environments.

    This open-source release builds upon Ant Group’s commitment to AI security research. The AI Security Lab has been actively contributing to the safety of the open-source agent ecosystem, having conducted specialized security audits on the OpenClaw framework and, in April 2026, collaborating with Tsinghua University to open-source ClawAegis, a security plugin covering the entire lifecycle of OpenClaw agents.

    Rooted in over two decades of expertise in payment safety, data security, and privacy protection, these advancements are already deployed in Ant Group’s products such as Alipay AI Pay and AI healthcare app AQ. Furthermore, Ant Group is actively participating in global AI governance by supporting ITU international standards and advancing industry protocols for secure agent interoperability, driving AI security from technological innovation to industrial practice.

    In addition to the agent-focused SingGuard-NSFA, Ant Group’s AI Security Lab has also open-sourced SingGuard, a policy-adaptive multimodal guardrail model family. Designed to secure complex interactions across both image and text generation, SingGuard serves as a comprehensive safeguarding tool that provides robust protection for user privacy and intellectual property, while preventing the generation or dissemination of illicit material such as child sexual abuse material (CSAM), self-harm content, and other illegal outputs.

    The post Ant Group Advances AI Agent Security first appeared on AI-Tech Park.

  • DOGE Used AI for Housing Policy. The Government Won’t Say How

    DOGE Used AI for Housing Policy. The Government Won’t Say How

    In response to a public records request, HUD has withheld documents about DOGE’s use of AI—in part by citing a privilege that doesn’t exist.

  • Glint Opens Public Beta of AI-Native Git Workspace

    Glint Opens Public Beta of AI-Native Git Workspace

    Native desktop application brings multi-repo Git management, terminals, worktrees, and AI-assisted development into one repository-aware workspace

     Glint Software today announced the public beta of Glint (glint.dev), a native desktop workspace that brings multi-repo Git management, native terminals, and agentic AI coding into a single application. Instead of stitching together a code editor, multiple chat windows, and separate terminal tabs, developers get one repository-aware environment built for the way AI-assisted development actually happens.

    As teams increasingly use AI agents to refactor code and ship features, the surrounding tools have not kept pace. Editor extensions often bolt AI onto interfaces designed primarily for single-file editing, lose context when developers switch branches, and treat multiple repositories as an afterthought.

    Glint takes a different approach: a purpose-built desktop application where Git worktrees, issues, pull requests, terminals, and an AI agent called Glint Assist share the same workspace and repository context. Glint also supports multiple providers, working across GitHub, GitLab, and Bitbucket rather than requiring developers to use a single repository host.

    “Developers shouldn’t have to bounce between chat windows, terminal tabs, and editor extensions just to run an AI-driven refactor,” said Jim Bourke, Founder of Glint Software. “We built Glint as a native desktop app because repository awareness, responsiveness, and developer control matter. We’re opening the beta now because the core workflow is solid, and we want real multi-repo workloads shaping where it goes next.”

    What Glint offers

    Agentic AI in context. Glint Assist works directly with repositories, worktrees, files, and terminals, allowing it to read files, propose edits, and run commands with awareness of the active project. Developers can also run CLI agents such as Claude Code, Codex, and Gemini in built-in terminal tabs.

    Multi-repo and worktree-native. Glint is designed around multiple repositories and Git worktrees, making parallel branches, related projects, and simultaneous development workflows easier to manage.

    One workspace, many tools. File browsing and editing, commit history, branches, pull requests, issues, and terminals are brought together in a tabbed, splittable interface, allowing more day-to-day development work to remain in one place.

    Works with major repository hosts. Glint supports GitHub, GitLab, and Bitbucket, so teams are not forced to adopt a single provider to gain an integrated development experience.

    A native desktop application. Built natively rather than as a browser-based wrapper, Glint is designed for developers working across large, complex, and multi-repository codebases.

    Pricing and availability

    Glint enters public beta on Monday, July 13, 2026, and is available to download at glint.dev.

    Free plan. Glint is free to download and use with GitHub for public repositories, with no subscription required.

    Try Glint Assist at no cost. Registered users receive a promotional trial of Glint Assist, allowing them to use the AI agent with their own repositories before choosing a paid plan.

    Beta discount. Paid plans start at $7 per month. Eligible beta users can apply code GLINTBETA to receive up to $10 off per month on select plans for their first three months.

    As a public beta, Glint will continue to evolve, and early users will directly help shape what comes next. Downloads, documentation, plan details, and information about submitting feedback are available at glint.dev.

    The post Glint Opens Public Beta of AI-Native Git Workspace first appeared on AI-Tech Park.

  • Superhuman Announced the Launch of Superhuman Docs

    Superhuman Announced the Launch of Superhuman Docs

    New capabilities move AI beyond the chat window and into the documents teams rely on, enabling them to build with AI together and create custom tools connected to their data

    Superhuman Databases launches in beta as a dedicated home for enterprise-scale data management

    Superhuman, the productivity platform bringing AI to wherever people work, today announced the launch of Superhuman Docs, marking the evolution of Coda into an AI-native product within the Superhuman suite. Docs pairs a flexible document surface with a new generation of AI capabilities that turn simple prompts into drafted content, structured tables, and custom, interactive views, so teams can bring AI into how they already work together.

    What’s new:

    • Docs AI: A completely rebuilt AI assistant that can read, plan, write, track, and build inside any doc. Unlike traditional documents, Superhuman Docs are interactive, with building blocks like tables, formulas, automations, and live connections that power the way teams work, from project trackers and content calendars to product roadmaps and team hubs. With Docs AI, users describe what they need in plain language, and it produces content, builds formatted tables, resolves comments, and pulls information across documents. Docs AI can also leave inline comments with feedback, pull in information from the web, and suggest what to build next based on the doc.
    • AI views (private beta): Describe a custom interface, app, or tool, and AI builds it on top of a table’s data, no developer experience needed. AI views run on live data, with drag-and-drop interactivity and AI-powered actions, and automatically update as the underlying data changes. Users can also build AI views from data pulled from connected tools like Jira or Salesforce.
    • AI blocks and columns: Describe what a column should do, and AI columns auto-populate results across every row. AI blocks generate live summaries, actions, and key insights that refresh as data changes.
    • Superhuman Databases (private beta): A new product for managing structured data at enterprise scale, with support for up to 1 million rows per database. Teams can centralize shared datasets, build trackers and operational systems that scale with their work, and connect that data directly to Docs so the same information is available across surfaces.
    • MCP: Connect ChatGPT, Claude, Cursor, and other AI clients to Superhuman Docs with one click, allowing those tools to read and write to docs and tables in real time.
    • New desktop and mobile apps: A new native macOS app with tab navigation and native notifications, and redesigned iOS and Android apps that are more responsive and simpler to use.
    • Table pages: A fullscreen page type built for structured data, giving teams room to work on project trackers, data workflows, and bulk editing.

    AI built for team collaboration

    Despite the rapid growth of AI tools today, the impact has been personal, not collective: 65% of workers who use AI say it has had a positive impact on their individual productivity, but only 12% agree it has transformed how their team works. Most AI tools require careful prompting and produce results that need to be manually moved back into documents or trackers. The work that AI helps create stays with the person who prompted it, while the rest of the team is left to rebuild the context around it.

    Superhuman Docs closes that gap. Users describe what they need in plain language, whether that’s a project tracker, a brief, or a weekly status update from meeting notes, and AI builds it directly in the doc. From there, teammates can edit and build on the work together, leave comments for the AI to address alongside their own, or ask Docs AI to research a topic, draft feedback, or extend the work further.

    “We’ve spent over 10 years making Coda the most flexible surface for teams to write, track, and build together,” said Lane Shackleton, Head of Superhuman Docs. “Superhuman Docs is the next chapter. When AI lives inside the same surface where teams already collaborate, it stops being a private tool one person uses in a side window and becomes a true team multiplier. Anyone can describe what they need, our AI uses context from the doc and builds it immediately, and the whole team can edit, refine, and improve on it together.”

    “What Superhuman Docs does for an organization is make things simpler for the team,” said Pam Larson, an education technology Professional Learning and Development Leader. “It helps overcome the frustration of working in several different programs, makes teams more efficient, and adds to the professionalism of what they’re creating. I’m excited to have AI sitting inside the tools we use every day.”

    Superhuman Docs in action

    Superhuman Docs supports teams, from planning to execution. It can draft documents from a single prompt, surface answers across a team’s docs, build tables and charts from your data, and write formula logic, with all context already built in rather than pasted into a separate AI window.

    A few examples of what teams can do with Docs:

    • Marketing teams run campaigns from a single connected hub. A campaign manager prompts Docs AI with “build a product launch hub with sections for project vision, team tasks, launch status, and exec summary with charts,” and AI builds it directly in the doc, drawing on the team’s existing messaging and recent campaign data. A teammate jumps in to refine the messaging section, then asks Docs AI to research positioning for comparable launches and pull the findings back into the doc. A follow-up like “create a content calendar with columns for channel, format, status, and owner” produces a structured tracker with formulas that auto-calculate publish dates. Docs can also pull and push live data from tools like Jira, Salesforce, and Asana, so the hub is accurate without anyone copying data between tools.
    • Sales teams build account plans that stay current as deals evolve. Preparing for a quarterly business review, a sales lead connects a live view of the team’s Salesforce pipeline into the doc, then prompts Docs AI with, “draft a QBR narrative with the top five at-risk accounts and recommended actions.” It produces a structured first draft with charts that update as the pipeline does, so leadership reads the story and the data side by side with no stale numbers or tabs to switch between.
    • Product teams plan together in one place. After a planning sprint, a product manager prompts Docs AI with “design a project dashboard pulling from my active tracker that shows progress, risks, and owners.” AI builds it directly in the doc, reading from the tracker the team already maintains. When a designer and an engineer need different views of the same data, Docs AI generates a roadmap for one and a Kanban board for the other, both pulling from the same source.
    • Executive teams turn an event guest list into a draggable floor plan of the venue. Ahead of a board dinner, a chief of staff keeps the guest list in a table with each person’s name, role, and a note on which teammate is the best fit to sit beside them, then prompts Docs AI with “lay this out on the venue’s floor plan and use the notes to seat everyone in the optimal seat.” AI builds an interactive map that mirrors the real room, drops each guest into a chair based on the notes, and lets the team drag people between seats, with every move updating the table underneath.
    • Use Superhuman Docs from AI clients with the new Superhuman Docs MCP. Teams can use AI tools like ChatGPT, Claude, or Cursor to create docs and tables, turn rough notes into a polished write-up, or gather information from other connected tools, all without leaving their chat conversation. The Docs MCP is also available today in Anthropic’s Connectors Directory and the OpenAI App Directory.

    Learn more:

    • Access to Superhuman Docs will start rolling out today for all existing Coda customers and new Superhuman suite customers. Existing Coda customers will see their docs automatically become Superhuman Docs, with access to Docs AI and the updated capabilities.
    • Read our blog post: Coda is now Superhuman Docs — the best place for teams and AI to work together
    • Learn more about Superhuman Databases in beta and request access here.
    • AI views is in private beta, with early access available through a waitlist.

    The post Superhuman Announced the Launch of Superhuman Docs first appeared on AI-Tech Park.

  • Exercise doesn’t just strengthen the heart. It rewires it

    Exercise doesn’t just strengthen the heart. It rewires it

    Exercise doesn’t just make the heart stronger. It also rewires the nerves that regulate it, a discovery that could pave the way for more personalized treatments for common heart conditions such as arrhythmias and angina.

  • Yale scientists found a hidden network inside the eye

    Yale scientists found a hidden network inside the eye

    Researchers have discovered that the retina uses an unexpected communication network that lets separate visual pathways cooperate instead of working alone. A newly identified “commander” cell appears to coordinate this system, helping the eye detect faint details that might otherwise be missed.

  • New dark matter theory could solve multiple cosmic mysteries at once

    New dark matter theory could solve multiple cosmic mysteries at once

    Dark matter may be far more complicated than scientists once believed. A new study suggests it could consist of at least two different kinds of particles that slowly separate over time, with heavier particles sinking toward the centers of galaxies and lighter ones drifting outward. This simple idea could explain several puzzling cosmic observations that have frustrated astronomers for years, from unusually diffuse dwarf galaxies to surprisingly dense dark matter clumps that bend light through gravitational lensing.

  • Alan Turing’s biggest AI assumption may have been wrong

    Alan Turing’s biggest AI assumption may have been wrong

    A new book claims AI has been built on a flawed assumption dating back to Alan Turing’s famous 1950 paper. Peter J. Denning argues that the most important parts of human intelligence, including common sense, intuition, culture, and practical know-how, cannot be encoded into computers. He believes this makes true human-level AI impossible, regardless of how large language models become.