Humanoid robots · Deployment · Reality check
The humanoid revolution sold about 7,000 robots last year
Humanoid robotics has become one of technology's most visually persuasive industries. The machines run, dance, box, carry cargo and perform backflips. The International Federation of Robotics has supplied the less cinematic number: about 7,000 humanoids were sold globally in 2025 for industrial and professional-service applications.
The figure does not count every humanoid machine on Earth, but it is one of the first standardized measurements of the commercial category. Many buyers were research institutions and companies using the hardware for AI development, testing and pilots. Automakers are early adopters, yet small experimental fleets remain very different from production lines built around thousands of proven machines.
Forecasts can still be right: costs are falling, capabilities are improving and investment is enormous. Counting deployed machines keeps the timeline honest. Demonstrations ask whether a robot can perform a task; commercial adoption asks whether enough customers will build operations around it. For now, the professional humanoid revolution would fit inside a moderately busy concert venue.
Source: Reuters ↗AI · Cybersecurity · Continuous testing
The AI red team no longer needs to go home at five
Palo Alto Networks has launched Unit 42 Continuous Frontier AI Defense, an annual subscription service that perpetually searches enterprise systems for exposures. Its multi-model harness includes Anthropic's Claude Mythos 5, OpenAI's GPT-5.6-Cyber and open-weight models, routing tasks to different systems as corporate environments change.
The service begins with a full-estate baseline scan and continues testing web applications, APIs, cloud infrastructure, source repositories and network assets. It validates possible attack paths, then provides prioritized repairs, code-level guidance and virtual-patch recommendations. Palo Alto says six months of internal development and more than 100 customer engagements preceded the launch.
A conventional penetration test is a snapshot; new code and configuration changes begin aging it immediately. Continuous testing changes the cadence to persistent adversarial inspection. Attackers and defenders gain speed from the same underlying capability, shrinking the interval between introducing a mistake and discovering it. The practical question becomes which AI gets there first.
Source: Palo Alto Networks ↗Quantum physics · Decoherence · Fundamental limits
Quantum systems apparently have a speed limit for falling apart
Quantum states exchange energy and information with their environment, producing decoherence and decay. In many-particle systems, that decay can become correlated and collectively enhanced. Researchers reporting in Nature Physics asked how the maximal decay rate of a large quantum system scales with its size, a calculation that is generally intractable.
The team derived rigorous upper and lower bounds for a broad class of Markovian many-body systems. For several physically relevant cases the bounds become asymptotically tight, yielding exact scaling laws. In large free-space atomic arrays, the scaling depends on array dimensionality rather than microscopic details such as lattice geometry or atomic polarization.
This is theory rather than a device suddenly hitting a measured wall, but it supplies a ceiling for systems that exploit or must survive collective emission. The result may constrain quantum processors, simulators and metrology as atomic arrays scale. Even quantum systems falling apart are required to remain within a particular amount of strange.
Source: Nature Physics ↗