Capabilities · Concept
Claude for Physical AI (UST Deployment)
Anthropic Model Hardware Standard research preview for AI agents to safely operate physical devices in labs and manufacturing.
Forecast horizon
Shared MHS spec may become interop layer for lab and factory agents.
Connections
Connections · 7
How this node ties into the rest of the map, and the evidence behind each link.
from · enables 7/10
Anthropic Anthropic case study covers UST bringing Claude to physical AI.
+7 growth from · enables 4/10
Model Hardware Standard (MHS) for Physical AI Agents Shared hardware interface specs are intended to let Claude-class agents safely control lab and manufacturing devices.
+4 growth from · constrains 4/10
GuardianBench: Same-Scene Instruction-Contrastive Embodied Safety Benchmark Latent instruction-scene hazards benchmark the safety needs of embodied and physical AI agents.
+4 growth to · applies to 3/10
Claude Fable 5 / Claude Mythos 5 / Claude Opus 4.8 (Anthropic) UST is applying Claude models to physical AI use cases, extending Anthropic's model family into robotics/industrial contexts.
+3 growth from · implemented by 3/10
Anthropic Anthropic newsroom highlights UST bringing Claude into physical AI deployments.
+3 growth from · standardizes 3/10
Model Hardware Standard (MHS) for Physical AI Agents MHS is intended as a shared specification layer for agents safely operating physical hardware.
+3 growth from · standardizes 3/10
Model Hardware Standard (MHS) for AI Agents on Physical Devices MHS aims to standardize safe agent operation of physical devices for labs and advanced manufacturers.
+3 growthSignal sources
Signal sources
Dated facts from primary sources in this direction.
Task horizon doubling Mar 2025
The length of software tasks AI agents can do autonomously at 50% reliability has doubled about every 7 months — and since 2024 closer to every ~3 months.
METR → Benchmarks saturating Apr 2025
In one year scores rose by 18.8, 48.9 and 67.3 points on MMMU, GPQA and SWE-bench; real-world software solve rate jumped from 4.4% to 71.7%.
Stanford HAI — AI Index 2025 → Autonomous coding 2025–2026
On SWE-bench Verified (500 real GitHub issues), autonomous coding agents reached ~80–86% by late 2025, up from under 50% in early 2025.
Epoch AI →