EXPLORINEER
en

KR — ZDNet Korea

"I Thought It Was a Robot Malfunction, but It Was a Hack": Physical AI Faces Growing Security Threats

Security2 min read

Research firm Gartner warns that 'physical AI' in robots, drones, and self-driving cars can be attacked in new ways. Hackers could bend the AI's view of the world, and that could cause injuries or equipment damage.

What happened

Gartner published a report on physical AI and its effect on chief information security officers (CISOs), the people who lead a company's cyber defense. The report says old network-focused security is not enough to keep physical AI safe. Physical AI is AI that senses the world through cameras and sensors and then moves on its own, as robots, drones, and self-driving cars do.

Key numbers and facts

  • An IEEE study cited by Gartner found that a subtle attack could make a system drift up to 8 meters off its path.
  • The same attack could make the system 'see' obstacles that are not there.
  • Delays of only about 15 milliseconds (ms) during network security checks can break real-time control.
  • Gartner says most industrial cyber-physical systems do not yet use physical AI at full scale.

Why this is hard to spot

Physical AI is built to adjust to changes around it. So when an attacker slowly nudges its sensor data, the change can look like normal learning. Also, AI models can behave in unpredictable ways. A chatbot that makes up an answer is a small problem. A robot arm that moves wrongly or a motor that overpowers can cause real harm.

Why it matters

When a machine with physical force makes a mistake, people and factories can get hurt. Gartner says software alone may not stop these accidents. Companies may also be held responsible if they did not add proper safety devices or run their systems safely. Security teams need to plan for this before physical AI spreads widely.

What to watch

  • Whether companies add hardware safety switches that work apart from the AI.
  • Whether security checks can be designed so they do not slow real-time control.
  • Separate monitoring systems that check actual movement and switch to a safe mode when something looks wrong.
  • How responsibility rules change if an AI-driven machine causes an accident at work.

Glossary

Gartner
A global market research company that publishes reports on technology trends and advice for businesses.
Physical AI
AI that senses the real world through sensors and moves machines such as robots, drones, or self-driving cars.
IEEE
A large international professional group of engineers that publishes research and technical standards.

Same story, different eyes

Ad

Add 150 words. Learn 10 a day.

Dicter word books send only the words you learn to your review notes.

More in this topic

SecurityArs Technica

Ships loiter near subsea cables amid Iran's warnings to Trump

Several Iran-linked tankers and an Iranian fishing boat were tracked loitering over subsea Internet cables in the Gulf of Oman near the Strait of Hormuz in October, according to maritime intelligence firm Windward, as reported by the Financial Times. Windward said the movements could indicate preparations to map or sabotage the cables, with some ships switching off their AIS transponders. Six active cables cross the strait, which links the Middle East to Europe, Africa and Asia.

SecurityArs Technica

Chainalysis says hackers stole credentials and tokens for all of its customers

Blockchain analysis firm Chainalysis says hackers obtained sensitive personal information, including names, usernames, password hashes, and live tokens, for all of its users after breaching its network security. The company said an external actor used AI agent-assisted techniques in an October incident and later described an on-path attack that exposed some customer passwords. Chainalysis has not said where the infrastructure compromise occurred or whether the stolen data was TLS encrypted, and it is unclear whether the theft exposes customers to additional threats.

SecurityTechCrunch

Hundreds of Artex AI hacking tools found publicly exposed to the web

Security researchers at Censys found more than 800 deployments of Artex, a Chinese-made open source autonomous penetration testing tool, accessible from the web without a password. The tool has been linked to several cyberattacks, including against South Korean banks, and to an attack on IT support provider ConnectWise that let Chinese-linked hackers spy on thousands of organizations. Artex has since announced it will go closed-source, though dozens of copies of its code remain on code-sharing sites.

SecurityThe Verge

Just about anyone can rent access to White House IP addresses

Security firms Layer3 Intel and Synthient found residential proxy activity on the White House network, indicating outsiders could rent access for "a few dollars." Layer3 found 28 residential proxy services routed traffic out of the network between October 5 and November 3, typically between 5PM ET and midnight. Synthient linked the activity possibly to the watch party app Rave, which it says bundles 12 proxy SDKs.