The morning briefing.
While you slept: Honor unveiled a robot phone, Google Deepmind solved decades-old math problems, and Huawei redefined chip scaling to 1.4nm.
Frontier AI Breakthroughs. Google Deepmind's AlphaProof Nexus has autonomously solved nine open Erdős problems, including two that stumped mathematicians for 56 years, for just a few hundred dollars in inference costs. This system uses the Lean compiler for automatic proof verification, a distinct approach from natural-language methods. Meanwhile, GigaAI unveiled its Physical AGI "Dual Pyramid" architecture, aiming to overcome data and algorithmic bottlenecks in embodied AI scaling.
AI Reliability and Agent Security. Researchers found that leading AI models often give correct answers but cite incorrect sources, a phenomenon termed "attribution hallucination." This poses significant risks for regulated fields like law and medicine. Separately, the rise of self-running agents is creating a major security crisis, as these autonomous systems introduce new vulnerabilities and challenges for enterprises. Programmer George Hotz also warned that AI coding agents could become one of the most costly mistakes in software development due to hard-to-spot bugs.
Next-Gen Hardware. Honor unveiled its Robot Phone, the world's first smartphone with a motorized arm that functions as an integrated gimbal, leveraging on-device embodied AI for autonomous photography. In semiconductor news, Huawei introduced the Tau Law, a new scaling framework using logic folding to target transistor densities equivalent to 1.4nm by 2031. Additionally, Ganfeng Lithium Group has begun small-batch production of a 500Wh/kg solid-state battery, claiming it as the world's first 10Ah lithium-metal solid-state battery.
AI Governance and Commercialization. South Korea's deputy prime minister expressed concerns that AI wealth could exacerbate economic disparities or lead to job losses, emphasizing the need for AI benefits to reach the public. Anthropic announced plans to release its Mythos-class models to the public, while keeping a flaw-finder under wraps for now. On the commercial front, Salesforce faces criticism for what some call an "AI vaporware problem" with its Agentforce platform, and Finland’s Grundium acquired Denmark’s Visiopharm to build an end-to-end AI precision pathology platform.
Google Deepmind's AlphaProof Nexus solves decades-old math problems for a few hundred dollars
Google Deepmind's AlphaProof Nexus has autonomously solved nine open Erdős problems, including two that stumped mathematicians for 56 years. The system achieved this for just a few hundred dollars per problem in inference costs, using the Lean compiler to verify every proof step automatically.
Honor Robot Phone: A Smartphone That Literally Reaches Out and Takes Photos for You
Honor unveiled the Robot Phone, the world's first smartphone to incorporate a motorized arm that doubles as an integrated gimbal. This device is paired with on-device embodied AI for autonomous photography and smart home control.
GigaAI Unveils Physical AGI "Dual Pyramid" System, Targeting the Embodied Intelligence Scaling Wall
GigaAI revealed the world's first Physical AGI "Dual Pyramid" architecture at its launch event in Wuhan's Optical Valley. This dual-track framework aims to tackle the data and algorithmic bottlenecks preventing embodied AI from achieving true scaling.
Why self-running agents are creating the biggest security crisis of 2026
Securing autonomous agents is becoming the biggest security crisis of 2026, as enterprises struggle to bridge the gap in this new AI era. The rapid deployment of self-running agents introduces complex security challenges that traditional methods are ill-equipped to handle.
Huawei's Tau Law Redefines Chip Scaling: Logic Folding as the Post-Dennard Path to 1.4nm
Huawei Board Director He Tingbo introduced the Tau Law, a new semiconductor scaling framework centered on temporal minimization through logic folding. This innovation targets transistor densities equivalent to 1.4nm by 2031.
George Hotz says coding agents will be "one of the most costly mistakes" in software development
Programmer George Hotz warns that AI coding agents will become one of the industry's most costly mistakes after six months of testing. He states that LLMs deliver fast prototypes but produce bugs that are increasingly difficult to spot.
Salesforce Has an AI Vaporware Problem
Salesforce is facing criticism for what some describe as an "AI vaporware problem," with marketing for its agentic platform, Agentforce, being heavily focused on future capabilities rather than current offerings. This suggests a disconnect between promised AI features and their present reality.
Tequipy raises €3M+ to automate global IT operations across 180 countries
Tequipy, a Polish-British startup, has raised over €3 million in funding to automate global IT operations across 180 countries. The platform ships, services, and retrieves employee IT devices, with plans to expand into software and security operations.
AI models often give the right answers but point to the wrong sources
Leading AI models like GPT and Gemini routinely cite text passages in document analyses that do not actually support their answers, even when the answer itself is correct. Researchers at Peking University call this "attribution hallucination," posing a risk for regulated fields.
AI wealth must benefit the public, South Korea's deputy PM says amid Samsung labor tensions
South Korea's deputy prime minister stated that there are concerns about whether AI could worsen wealth gaps or lead to job losses. He emphasized that the wealth generated by AI must ultimately benefit the public.
Ganfeng Lithium Begins Small-Scale Production of the World's First 500Wh/kg Solid-State Battery
Ganfeng Lithium Group has commenced small-batch production of what it claims is the world's first 10Ah lithium-metal solid-state battery. This battery achieves an energy density of 500Wh/kg.
Anthropic to release Mythos-class models to the public
Anthropic is set to release its Mythos-class models to the public, extending access to more users, including governments. However, an AI flaw-finder remains under lock and key while the company develops appropriate guardrails.
Finland’s Grundium acquires Denmark’s Visiopharm to build an end-to-end AI precision pathology platform
Grundium, a Finnish digital pathology imaging startup, has acquired Visiopharm, a Danish provider of AI-driven precision pathology software. This merger aims to create an accessible end-to-end solution for diagnostic laboratories by combining their complementary capabilities.
MemAudit: Post-hoc Auditing of Poisoned Agent Memory via Causal Attribution and Structural Anomaly Detection
MemAudit proposes a method for post-hoc auditing of poisoned agent memory using causal attribution and structural anomaly detection. This addresses the security vulnerability where malicious records can be injected into an agent's persistent memory to steer its reasoning and actions.