National Security Rides on Robots: The Attack on GeekCon and China's Standardization Strategy for the Embodied AI Ecosystem

The advent of Embodied AI, with increasingly autonomous and integrated humanoid robots, marked a point of no return in global security. Cyber threats have moved beyond the realm of code to manifest as risks of physical harm and destruction. The news is dominated by the "GeekCon 2025 Fist," the incident in which a hacker attack on a Chinese robot demonstrated how easily a machine can be compromised and turned into a weapon, even a "digital Trojan horse" capable of infecting physically isolated (air-gapped) systems.

This endemic vulnerability is not just a technical problem, but a strategic criticality in the context of global competition. China has recognized that dominance in AI and robotics relies on infrastructural security. The country's response has been an acceleration in regulatory standardization, introducing a vast body of new sectoral standards (Edge Computing, 5G, Big Data, Blockchain) aimed at strengthening the Cyber Security Multi-Level Protection System (Dengji Baohu). This systemic effort aims to create a national protection ecosystem that shields the entire technological supply chain, shifting from a reactive patch-based approach to a preventive model of "Embodied Security". Security is no longer optional; it is the "passport" necessary for Embodied Intelligence to enter society reliably.

Unitree G1 Avatar AI humanoid agent - Credit UNITREE, official site

The Risk Shifted from Digital to Physical

For over three decades, cybersecurity focused on protecting information. Today, with the emergence of Embodied Intelligence, the security problem has acquired a body.

When AI gains mobility, strength, and autonomous decision-making, a hacked robot becomes a physical threat. As experts effectively summarized, if a virus in the past could "delete all your files," a security incident in the era of Embodied Intelligence today can "destroy your house and even harm people".

The urgency for a systemic and regulatory intervention is now undeniable, laying the groundwork for the transition to "Embodied Security".

The Dramatic Current Reality in China – The Shattered Illusion of the "Punch of GeekCon 2025"

The catalyst for this urgency was a globally resonant demonstration event held in China: GEEKCON 2025. The incident took place during the GEEKCON 2025 Security Geek Competition. The event, held in China, is a cybersecurity contest designed to discover and fix security flaws in new technologies. The focus of the incident was a Chinese-made humanoid robot, the Unitree G1. Two white-hat hackers first breached the security of a networked robot.

The real breakthrough, however, occurred when the hackers used the compromised robot as a vehicle to "infect" a second identical robot, even though it was physically disconnected from the Internet, or air-gapped. The first robot turned into a "digital Trojan horse," using near-field wireless communication to spread the threat to an isolated companion. The final result was the most dramatic: the robot executed a malicious command, forcefully punching a dummy.

This demonstration shattered the illusion of "physical isolation equals security". It proved that an attack can bypass the most basic barriers, and a machine can turn from a harmless assistant into a tool of destruction in moments.

The Diagnosis: A Sector "Full of Flaws"

Industry experts, such as the white-hat hacker Qushi Pei, found that many robot security systems are "full of flaws and loopholes" , and some manufacturers "haven't considered any security defenses". Reasons include:

• Functional Priority over Defense: Developers focus on functionality, neglecting the risks of attack and abuse.

• Talent and Budget Shortages: The pressure on corporate budgets exacerbated the threats.

• The Emergence of "Physical Logic" Attacks: Traditional defenses are ineffective against attacks exploiting "physical logic" (such as acoustic interference or visual stickers to trick sensors), a field that "is completely outside the traditional network security cognitive domain".

Robotics, National Security, and China's Strategic Response

Every robot is a potential access point to critical infrastructure. Compromising robots offers an opportunity for sabotage:

• Industrial Sabotage: A manipulated industrial robot can directly destroy production lines.

• Physical Intelligence Gathering: A hijacked domestic robot can become a "mobile spy" for stealing privacy.

• Fatal Risk in Vehicles: A breached autonomous driving system is not a "blue screen crash," but a deadly weapon.

Experts compare the risk of a cyber-attack on robots to that of an attack on a car or an airplane.

The Ethical Crisis and Asimov's Laws

The demonstration of a robot throwing a punch immediately evokes Asimov's Three Laws of Robotics.

Experts emphasize that ensuring the enforcement of these laws is a huge challenge. If a robot is hijacked, its "do no harm" code is ineffective, making cybersecurity equivalent to personal safety.

China's Strategic Response: Building a National Protection Ecosystem

Faced with these threats, China is significantly strengthening its security framework, turning vulnerability into a strategic imperative. The government has recognized that the traditional "patching" approach is failing and that a "paradigm shift" is needed, involving the entire production chain.

The response is not just limited to imposing single measures, but to constructing a comprehensive protection ecosystem, capable of ensuring technological sovereignty and national resilience against the era of Embodied AI. This ecosystem is based on extending protection standards to crucial enabling technologies.

The Defense Architecture – Chinese Standardization and the Dengji Baohu Model

The core of this new protection ecosystem is the systemic strengthening of the Cyber Security Multi-Level Protection System (Dengji Baohu) , which is transitioning from a protection system for traditional IT to an all-encompassing framework for Embodied AI.

The imminent implementation (scheduled for February 1, 2026) of six new Public Security industrial standards (GA/T) attests to the acceleration of this strategy. These standards extend the basic protection requirements (GB/T 22239—2019) or evaluation requirements (GB/T 28448—2019) to six technological areas that, together, constitute the infrastructure of Embodied Intelligence:

N.	Standard Number	Standard Name	Protection Type	Implementation
1	GA/T 1390.6-2025	Security Extended Requirements for Edge Computing	Extended Security Requirements	2026.02.01
2	GA/T 1390.7-2025	Security Extended Requirements for Big Data Systems	Extended Security Requirements	2026.02.01
3	GA/T 1390.8-2025	Security Extended Requirements for IPv6 Network	Extended Security Requirements	2026.02.01
4	GA/T 1390.9-2025	Security Extended Requirements for Blockchain	Extended Security Requirements	2026.02.01
5	GA/T 2347-2025	Security Evaluation Guide for Cloud Computing	Evaluation Guide	2026.02.01
6	GA/T 2348-2025	Security Evaluation Requirements for 5G Access	Extended Evaluation Requirements	2026.02.01

Ecosystemic Integration and Global Dominance

These standards are not isolated measures, but are designed to operate interconnectedly, creating a national protection matrix:

• AI Brain Protection (Big Data & Cloud): The Big Data and Cloud Computing standards ensure the security of centralized datasets and training platforms.

• Nervous System Protection (Edge Computing & 5G): The requirements for Edge Computing (with 5G technology) and 5G Access Security protect local decision nodes and communication channels crucial for robotic autonomy.

• Identity and Traceability Protection (Blockchain & IPv6): The Blockchain and IPv6 standards guarantee the immutability of audit logs and the security of network addressing, crucial for monitoring and incident response.

The geopolitical implication is that China is using standardization as a tool to ensure technological dominance over these new generations of systems. By setting internal rules, Beijing establishes the requirements for trust and access for foreign products, ensuring that the AI and robotics infrastructure aligns with national security interests, providing a "more solid guarantee for maintaining national cybersecurity".

Beyond Patches – The Vision of "Embodied Security"

Facing the complexity of attacks exploiting "physical logic" and AI vulnerabilities, the industry and research propose a radical paradigm shift.

The Need for Hardware-Level Ethical Veto

The true security of a robot must be embedded (embodied) in its fundamental design.

• Behavioral Safety: Focus must shift from "code compliance" to "behavioral safety". Any instruction implying high-speed movements toward a human must be directly intercepted or downgraded by the low-level system.

• Contextual Ethical Judgment: The robot must have a "limited autonomous veto power". Asimov's First Law must become unoverridable, hardware-level security logic.

• Irreversible Consequences: The consequences of a physical exploit (a punch, an impact) are immediate and irreversible.

  
  

Distributed Defense and AI Intelligence Vulnerability

An additional layer of risk comes from the robot's intelligence itself, often based on Large Language Models (LLMs).

• Sugar-Coated Poison (SCP) Attack: A new attack paradigm achieved an average success rate (ASR) of 87.23% against six mainstream LLMs. The attack exploits the decay of the model's attention on the initial prompt (Defense Threshold Decay - DTD) after generating a large amount of "benign" content , allowing the subsequent insertion of a malicious command.

• Part-of-Speech Defense (POSD): The Part-of-Speech Defense (POSD) strategy was proposed in response, using syntactic analysis to prioritize key verbs and nouns. POSD showed a reduction in the SCP attack success rate on the AdvBench dataset from 100% to 22.88% on DeepSeek-R1 and from 91.79% to 35.83% on GPT-4-0613.

• Supervisory Oversight (Distributed Defense): The defense model must shift from "centralized defense" to "distributed defense". Robots must be able to recognize abnormal behavior in their peers and undertake isolation or joint neutralization measures.

Security as the Foundation of Innovation

The advance of Embodied Intelligence is unstoppable. However, this promise is inextricably linked to security.

The robot hacking crisis has forced the political and technological spheres to face reality: the cost of security, perceived as high today, is infinitely lower than the cost of a national disaster.

The standardization in China, driven by the extension of the Dengji Baohu, is a national resilience imperative. It establishes the baseline (compliance) that companies can no longer afford to treat as "optional". This regulatory effort is not just a defensive barrier, but the construction of a protected and resilient national technological ecosystem.

In summary, security is not the obstacle to innovation, but its essential "passport". Only with a solid regulatory framework, capable of enforcing Embodied Security as a design principle, and with a geopolitical strategy that views robots as national security assets, can society confidently welcome these "next generations of steel bodies" and maintain technological leadership in an increasingly competitive and interconnected world.