GEN-1.5 teaches robots new tasks from a single demo

Share
GEN-1.5 teaches robots new tasks from a single demo

A robotics model that learns from one short demo, a confirmed breach at an AI-data giant, and new guardrails for autonomous agents — here's the AI news moving now.


Generalist AI's GEN-1.5 teaches robots a new task from a 3- to 12-second demonstration — no training required. The startup loads a short video of the task into the model's context window as what it calls a "physical prompt," and the robot then performs the action on its own, with no fine-tuning. In ten internal tests — opening a jar, pulling bills from a wallet — Generalist reports a 59 percent success rate with zero training, rising to 83 percent after just ten practice steps on five minutes of data. The model can also chain two demos into longer sequences, learn from simulation, and partially mimic human hand movements, abilities the company says emerged on their own during eight months of pretraining rather than being explicitly programmed. The catch: every result comes from Generalist itself, none have been independently verified, and the tasks shown are deliberately simple — so treat the "first across a wide range of tasks" claim as promising but unproven.


AI data platform Alation has confirmed a cyberattack after a Tuesday incident knocked some customer systems offline. The company, whose natural-language data-search tools serve more than 500 global enterprises including roughly half the Fortune 1000, said it found "unauthorized activity" in one system and is investigating; it disclosed no root cause, scope, or whether any data was taken. The admission follows a Tuesday status-page entry reporting "degraded availability" for some customers, resolved within an hour, with much of Alation's infrastructure hosted on AWS. A breach of a platform that indexes a company's most sensitive data is close to a worst-case scenario for its customers, and the silence on exfiltration will not calm the security teams at the firms that depend on it.


Zero Networks is extending its Palo Alto Networks integration to put identity-based guardrails around AI agents. The expansion moves the pairing from policy orchestration into enforcement: protected Linux workloads can now route traffic to Palo Alto firewalls for deep inspection, and a compromised asset gets cut off instead of taking down an entire network segment. The agent-focused piece ties Zero's AI Segmentation to Prisma AIRS, drawing deterministic boundaries around where each agent — carrying its valid credentials — is allowed to connect, and sending suspicious prompts and responses for inspection. As agents win broad access to databases and production systems, telling a hijacked agent apart from legitimate automation is the hard problem, and vendors are racing to make "contain the agent, not the network" the default posture.

What to watch: whether Generalist opens GEN-1.5 to independent benchmarking, and whether Alation confirms any customer data left its systems.

Should AI agents be sandboxed by default with identity-based limits, or is that friction too costly for the speed teams want? Write to us via our contact page.

Source: The Decoder

Get SiliconNoon in your inbox

One email a day with the AI stories that matter — what happened, the numbers, and why it matters. No spam, unsubscribe in one click.

Read more

Daxiao Robot unveils Kaiwu World Model 3.1 at WRC

Daxiao Robot unveils Kaiwu World Model 3.1 at WRC

Two threads in today's AI news: the embodied-AI race has a new open contender, and a top law school is drawing a hard line on classroom AI use. Daxiao Robot debuted the Kaiwu (Kairos) World Model 3.1 at the 2026 World Robot Conference in Beijing, pairing it with an open-source data-collection stack and three industry solutions for instant retail, hotels, and logistics. Kairos 3.1 fuses generative, physical, and cognitive intelligence in a single architecture, compressing visual feeds…

ShengShu lays out a five-level roadmap to general world models

ShengShu lays out a five-level roadmap to general world models

ShengShu founder Zhu Jun used a World Robot Conference keynote to put a name and structure on where embodied AI is heading: a five-level roadmap for a general world model he argues must unify generating the world, interacting with it, and acting inside it — not a stack of separate models bolted together. Zhu, the ACM/IEEE/AAAI fellow who built video model Vidu, described the general world model as a closed loop rather than a single generator or robot policy: a system that understands the…

XPeng ships VLA 630: on-device driving model up 3.5×, 3× faster

XPeng ships VLA 630: on-device driving model up 3.5×, 3× faster

Chinese electric-vehicle maker XPeng is pushing the self-driving model arms race from the datacenter down into the car, announcing a major upgrade to its second-generation visual-language-action (VLA) model that runs entirely on-vehicle. XPeng unveiled the first big update to its second-generation VLA, version 6.3.0 — dubbed the "630" — with the on-device model tripling its parameters and its end-to-end response getting three times faster.

Xiaomi's MiLM Plus ships PROVE, a benchmark for AI object removal

Xiaomi's MiLM Plus ships PROVE, a benchmark for AI object removal

The AI-editing arms race just got a proper measuring stick: Xiaomi's research team shipped an evaluation framework that says the metrics everyone uses to judge object-removal models are out of step with human perception. Meanwhile, the open-source fine-tuning favorite Unsloth put local model training on the desktop. Xiaomi's MiLM Plus research group released PROVE, an open evaluation framework for object removal in images and video that its authors say fixes what standard metrics get wrong.

Anu ships six releases to cut a robot station deploy to days

Anu ships six releases to cut a robot station deploy to days

Embodied AI's bragging rights are moving from leaderboards to factory floors. One Chengdu company just bundled a world model, a small on-site model and a deployment system into a single pitch — and published the unglamorous numbers that decide whether any of it is a business. Anu Intelligence (安努智能) has released six results at once, and the one that matters most is ActiWorld — a world model built to check what a robot's actions actually do rather than what the next frame looks like.

Motus2: a robot world model that grades its own actions

Motus2: a robot world model that grades its own actions

Shengshu Technology, the Chinese world-model lab, shipped the rare robotics release where the loop actually closed: one model that acts, imagines, and scores itself — then learns from the grade. Shengshu released Motus2, a "self-evolving" general world model for dexterous manipulation that puts three capabilities — acting, predicting, and evaluating — inside one shared-parameter video-action model, according to the project page and the paper on arXiv.

Five robot brains in eight days — and the money follows

Five robot brains in eight days — and the money follows

Two things happened to Chinese robotics in the same two weeks: the body makers shipped model after model, and the funding data caught up with why. Three Chinese robot makers released five embodied-AI foundation models in eight days this month — Unitree, AgiBot and Noetix — a pace that says the hardware race is over and the brain is the product. The sequence, per Red Star Capital Bureau: Noetix launched its Scalabot embodied-AI brand and the first HERON world model on September 8, then…