Meta Connect 2026 expands Muse agent to six life domains with Charm keychain wearable

At Meta Connect 2026, Mark Zuckerberg unveiled an expanded Muse personal AI agent capable of connecting to six life domains: email, calendar, payments, health, shopping, and smart home. A new keychain accessory called Muse Charm carries the full voice-and-avatar stack independently of a smartphone. The updated agent integrates with Meta AI Ray-Ban glasses and supports Shopify’s full product catalog, Shop Pay, and PayPal for in-conversation purchases. Muse runs inside a dedicated virtual machine overseen by Sentinel, a companion security agent that logs every action and maintains an audit trail accessible to the user.

Anthropic releases Claude Opus 5.5 at 1,846 GDPval Elo, outpacing Fable 5.1 at lower cost

Anthropic released Claude Opus 5.5 on September 22, 2026, scoring 1,846 Elo on the GDPval-AA v2.1 benchmark — 111 points ahead of Fable 5.1 at 1,735. Priced at $4 per million input tokens and $20 per million output tokens, the model is 40 percent cheaper than its predecessor and runs more than 30 percent faster. In coding evaluations, Opus 5.5 reached 66.4 percent on Terminal Bench 4.0, 54.4 percent on FrontierCode v1.1, and 57.8 percent on CursorBench 4.0. In a timed trial, the model completed a HAProxy C-to-Rust migration in 9.5 hours, 51 percent faster than Fable 5.1.

Arm proposes Robotics Capability Framework to standardize terms for AI-integrated robots

Arm proposed a Robotics Capability Framework on September 23, 2026, calling on standards bodies, manufacturers, and regulators to adopt consistent definitions for five terms commonly applied to AI-integrated robots: autonomous, intelligent, adaptive, collaborative, and self-improving. The chip designer said divergent usage of the terms creates confusion among enterprise buyers and complicates product certification and liability determination. Arm positioned the framework as a precondition for regulatory confidence as humanoid and collaborative robots expand into industrial and commercial environments. The company described itself as a platform-neutral coordinator across the robotics supply chain and said it would submit the proposal to relevant standards organizations.

China tops global physical AI readiness index with score of 82.9, ahead of Japan and US

China ranked first in physical AI readiness with a score of 82.9, ahead of Japan at 69.6 and the United States at 69.0, according to a report released by OpenMind on September 23, 2026. The index evaluates manufacturing capacity, government policy, talent pipelines, and infrastructure. China accounted for approximately 90 percent of global humanoid robot shipments in 2025 and around 97 percent in the first half of 2026, the report said. Japan led in industrial robotics heritage, while the United States scored highest for software and AI research ecosystems but ranked lower on manufacturing scale and policy coordination.

RLWRLD outlines deployment-first strategy for industrial humanoids at Humanoids Summit Seoul

Physical AI company RLWRLD presented a deployment-first strategy for industrial humanoids at the Humanoids Summit Seoul 2026, held September 22-23. Rather than conditioning robots in simulation before live deployment, RLWRLD places humanoids directly into active industrial environments and uses the resulting operational data to refine its AI models iteratively. The company said synthetic training cannot replicate the mechanical variance, unexpected safety interruptions, and human-collaboration patterns encountered on real production lines. RLWRLD described early deployments as generating proprietary training datasets unavailable through simulation. The Seoul summit drew manufacturers, logistics operators, and investors evaluating timelines for commercial humanoid adoption.

Singapore startup Nexstrom builds chip fab equipment for 2D TMD semiconductors beyond silicon

Singapore startup Nexstrom is developing chip fabrication equipment built around two-dimensional transition-metal dichalcogenides, or TMDs, as an alternative to silicon, with a funding announcement on September 22, 2026. Silicon transistors are approaching physical scaling limits that cap AI accelerator performance; TMDs allow current switching in atomic-scale layers potentially one to two atoms thick, extending transistor density beyond silicon’s ceiling. Nexstrom builds deposition equipment designed to apply TMD films directly onto existing production wafers, targeting compatibility with established foundry infrastructure. The company said its initial customers would be advanced logic chip manufacturers seeking post-silicon performance gains without full fab replacement.

Enterprise AI startup Ema closes $77M Series B led by Creaegis to expand agent platform

Enterprise AI startup Ema closed a $77 million Series B round on September 23, 2026, led by Bengaluru-based venture firm Creaegis, with participation from existing investors Accel, Section 32, and Prosus. The raise brings Ema’s total funding to $140 million and values the company at four times its 2024 valuation. Ema automates HR, IT, and finance workflows through coordinated teams of AI agents rather than single-model deployments. The round consisted entirely of primary equity, with no debt or secondary components. The company said it will use proceeds to expand its enterprise sales organization and extend its agent platform to additional business functions.

⚡ Tech Pulse