Chinese academy of sciences model tops Meta-World benchmark with 91.9 score

completeaitraining.com Oct 5, 2026

Chinese academy of sciences model tops Meta-World benchmark with 91.9 score

Curated by Shubham · TechCapsules · Oct 5, 2026

AI-summarised brief · reviewed before publication

The Chinese Academy of Sciences’ Institute of Artificial Intelligence for Industries released Maxwell, a 1‑billion‑parameter embodied AI model that achieved a 91.9 score on the Meta‑World benchmark, surpassing the previous 90‑point record held by FabriVLA. Maxwell performed 50 everyday physical tasks—grasping, opening doors, using drawers—without fine‑tuning and completed over 200 tasks on a single‑card inference setup. It also reached a 99.1% success rate on the LIBERO benchmark. The achievement narrows the performance gap between Chinese and Western robotics models and demonstrates strong generalization with modest compute.

💡 Why It Matters

· Maxwell shows that high‑level robot control can be achieved with a single‑card inference model, enabling deployment in compute‑constrained industrial and construction settings.

Uncategorized

Read full article on completeaitraining.com

Orbex assets bought by Omnidea

Portuguese aerospace firm Omnidea acquired the majority of Orbex’s assets, including two Sutherland Spaceport buildings, manufacturing equipment, and intellectual property, following Orbex’s 2026 shutdown. The deal, finalized on September 30, grants Omnidea access to technology and processes needed to develop PrimeNEO, an evolution of Orbex’s small launch vehicle. Omnidea, founded in 2003 and a key ESA service provider, plans to use facilities in Scotland and the Azores for future launch operations. The acquisition positions Omnidea to expand its launch portfolio amid a growing small‑satellite market.

💡 By securing Orbex’s technology and infrastructure, Omnidea can accelerate its entry into the competitive small‑satellite launch sector, potentially reshaping launch options for European customers.

cellit.in · Oct 5

India achieves 5.56-km free-space Quantum Key Distribution link

India successfully demonstrated a 5.56‑km free‑space Quantum Key Distribution (QKD) link between BISAG‑N and IIT Gandhinagar on September 27‑28, using QNu Labs’ Armos QKD device and a Pointing, Acquisition and Tracking system. The trial maintained a Quantum Bit Error Rate below five percent and generated secure keys at 230‑260 bits per second, which were integrated with BISAG‑N’s Vedic Kavach post‑quantum cryptography platform for end‑to‑end encryption. The collaboration combined hardware‑based QKD with software‑based post‑quantum security, showcasing a resilient architecture for future quantum‑secure networks.

💡 The field test validates India’s capability to deploy practical quantum‑secure links without fiber, paving the way for satellite‑based quantum communication and strengthening national resilience against quantum‑era cyber threats.

martinfowler.com · Oct 5

Fragments: October 4

Martin Fowler discusses the shift from building deterministic software to nurturing inferential systems like large language models (LLMs). He cites the 1956 film Forbidden Planet to illustrate the dangers of creating autonomous technology that can behave unpredictably. Fowler contrasts traditional software bugs, which can be fixed or disabled, with LLMs, which lack simple corrective measures, potentially leading to unintended consequences. He also notes the recent deportation of his colleague Ola Bini from Ecuador, highlighting broader geopolitical tensions surrounding technology experts.

💡 The article underscores how evolving AI systems demand new governance models, as conventional debugging methods no longer apply. It signals a critical juncture where technical, ethical, and legal frameworks must adapt to manage autonomous technologies responsibly.

helpnetsecurity.com · Oct 5

How RMM abuse gives attackers a way in that looks like business…

Huntress reported that 45% of endpoint incidents in Q1 2026 involved attackers abusing legitimate remote monitoring and management (RMM) software, marking it as the most frequently observed tactic. The study ranked 11 attack methods by prevalence and damage, placing RMM abuse at the far right of the chart. Attackers install RMM tools such as Tiflux, then add additional remote‑access utilities like UltraVNC, Splashtop, and ScreenConnect, enabling persistent, covert control that mimics normal admin activity. The technique surged 277% year‑over‑year in 2025, underscoring its growing threat.

💡 The rise of RMM abuse shows attackers exploiting trusted tools to bypass security controls, turning routine IT operations into covert attack vectors. This trend forces organizations to scrutinize approved software and monitor for anomalous remote‑access activity.

citinewsroom.com · Oct 5

Trump unveils ‘Super Intelligence Force’ to oversee AI policy

President Donald Trump announced the creation of the “Super Intelligence Force,” a new task force to oversee artificial intelligence policy. The group, chaired by Director of National Intelligence Jay Clayton, will include Federal Trade Commission Chair Andrew Ferguson, Defense Undersecretary Emil Michael, and Office of Personnel Management Director Scott Kupor. It aims to maintain U.S. leadership in AI development, coordinate government engagement with consumers, public interest groups, religious organizations, critical infrastructure, and AI companies, and report directly to the president and White House Chief of Staff Susie Wiles. The initiative follows a “morally binding” pact signed by top tech firms and reflects Trump’s preference for industry-led regulation.

💡 By centralizing AI oversight, the administration could streamline policy responses to emerging technologies, potentially accelerating U.S. competitiveness while shaping global governance norms.

technologyreview.com · Oct 5

EmTech Future 2026: When AI Meets Everything

Yossi Matias, Google Research’s vice‑president, addressed the EmTech Future 2026 conference, emphasizing AI’s expanding role across biology, infrastructure, manufacturing and scientific research. He argued that AI’s most transformative impact will emerge when it converges with other disciplines, creating hybrid solutions that reshape how complex problems are tackled. The three‑day event featured sessions on AI’s industry effects, quantum computing progress, new energy systems, and evolving robotics, highlighting both opportunities and challenges of technological convergence. Recordings of all talks, including Matias’s analysis, are now available on demand, offering insights into AI’s current limits—such as the lack of genuine creativity in large language models—and the realistic timeline for recursive self‑improvement.

💡 The convergence narrative reframes AI from a standalone tool to a catalyst that can accelerate breakthroughs in traditionally siloed sectors.