RuntimeWire

Real-time startup intelligence for the AI economy. Funding, model launches, infra shifts, and founder moves — the signals that matter.

Latest stories — page 27

• Skywork ships SkyClaw v1.0, an agent model tuned for OpenClaw, Hermes, and Nanobot

  Free access is available via APIFree, plus a faster lite variant and an OpenAI-compatible agent chat API with thinking mode, tool calls, and long-context stability.

• SafeDep says Mini Shai-Hulud payload slipped into 317 npm packages

  Researchers tie a 22-minute mass publish from a compromised npm account to the same toolkit seen in the recent SAP-related incident, with CI and AI tool persistence.

• MIT Media Lab debuts Human Operator, an AI that briefly moves your muscles to teach tasks

  The prototype uses Claude-driven vision-language planning and electrical muscle stimulation to guide wrist and finger movement for training and assistance.

• Boston Dynamics posts Atlas carrying a mini fridge

  A short X post frames Atlas as doing whole-body manipulation, shifting emphasis from acrobatics to moving real, awkward objects.

• HeyGen demos Avatar V workflow that stitches scenes into continuous AI video

  The team showed cinematic avatar shots linked into longer sequences, pointing at narrative workflows instead of isolated clips.

• Marin debuts Delphi to forecast 25B pretraining runs with 0.2 percent error

  In an X thread, Will Held says small-run fits predicted a 25B-param, 600B-token run at ~1e23 FLOPs, matching Paloma macro loss within 0.2% error.

• PolicyLayer: prompts are not permissions for production agents

  In a post on X, the company argues agent behavior should be controlled by enforceable policies rather than instructions buried in a prompt.

• Meta's AIRA paper points to agents co-designing model architectures

  Meta's AIRA architecture discovery work signals that agents are starting to help design neural models.

• Startup Spotlight: CancelFlow - Stripe cancel flows in under 10 minutes

  CancelFlow's site now outlines an embeddable retention flow that reads from Stripe via your publishable key, applies outcomes in Stripe, and supports A/B testing and full customization, with a 14-day free trial.

• RLWRLD debuts RLDX-1, a dexterity-first robot hand foundation model

  A short X video frames the bet: focus on robot hands and the failure points in everyday tasks like pouring coffee and grasping objects.

• UniPat AI team debuts SaaS-Bench as agents finish under 4% of real SaaS workflows

  Built across 23 live SaaS apps and 106 long-horizon tasks, the open benchmark finds frontier agents stumble on planning, memory, cross-app context, and error recovery.

• Mentra Live is GA: open-source smart glasses platform and SDKs

  He says thousands of units have shipped and hundreds of companies are deploying, with SDKs live at MentraGlass.com and international expansion targeted later this year.

• Anthropic reportedly edges OpenAI in U.S. paid business accounts. Marketing, product, or OpenAI fatigue?

  An X post cites 34.4% vs 32.3% U.S. enterprise share for Anthropic with no disclosed methodology. The gap is slim and could reflect marketing, product fit, or buyers seeking an OpenAI alternative, even as some developers say OpenAI's code tools beat Claude Code.

• Colossus pegs Cognition AI at $445M run rate on Devin; US Army, Goldman, Mercedes named as customers

  A Colossus magazine Q&A and profile cites Devin at a $445M annualized run rate with usage doubling every eight weeks, and names the US Army, Goldman Sachs, and Mercedes-Benz as early customers.

• OpenAI launches Deployment Company, agrees to acquire Tomoro

  Majority-owned unit will embed Forward Deployed Engineers; backed by 19 partners and starting with 150 specialists and $4 billion, per OpenAI and Greg Brockman.

• CSTA picks 15 teachers for second Responsible AI Fellowship

  The six-month program begins this summer, selected from 124 applicants and backed by Google and GenerationAI.

• Kexin Huang introduces BiomniBench to grade AI agents on how they think, not just what they answer

  In an X post amplified by Peter Jansen, Huang calls BiomniBench the first benchmark to score the process behind an agent’s output, not only the final result.

• Physician-founded Kin Health raises $9M to put an AI notetaker in the patient’s pocket

  Led by Maveron, the seed backs a free app that records clinic visits, turns them into clear summaries with next steps, and keeps family in the loop.

• Vita Dynamics raises about $70M in pre-A as China’s humanoid race shifts to production

  The sizable pre-A round signals a push from R&D to manufacturing as Chinese humanoid startups aim to scale ahead of Western peers.

• Files.md takes a swing at Obsidian with an open-source Markdown notebook

  Built in Go with a server and web UI, the project is seeing near-daily commits as it positions itself as a hackable Markdown note app on GitHub.

• Ciridae raises $20M seed from Accel and Andreessen Horowitz to push AI into the physical economy

  Accel and Andreessen Horowitz are backing the early-stage startup as AI automation shifts focus from software workflows to real-world operations.

• MIT's GenCAD turns images into editable CAD programs

  MIT researchers Md Ferdous Alam and Faez Ahmed release a demo, paper and code for an image-conditioned model that outputs full parametric CAD command histories.

• Researchers unveil Fabricked, a software attack that misroutes Infinity Fabric to break AMD SEV-SNP

  The team shows a malicious hypervisor can subvert SEV-SNP init by dropping PSP writes via Infinity Fabric misrouting; they report 100% success and confirm on Zen 5 EPYC.

• Mistral CEO Arthur Mensch tells French lawmakers Europe has two years to build its own AI stack

  The 33-year-old cofounder warned control of chips, energy, and data centers will decide AI power, urging a rapid sovereignty push in a National Assembly hearing.