Welcome to the August 22, 2025 edition of ACM TechNews, providing timely information for computer professionals three times a week.

HEADLINES AT A GLANCE

• EU to Curb AI Chip Flows to China as Part of U.S. Trade Deal
• FBI Warns of Russian Hacks Targeting U.S. Critical Infrastructure
• Top Law Schools Boost AI Training as Legal Citation Errors Grow
• Chipotle Teams Up with Zipline to Test Drone Delivery
• Humanoid Robot Does Complex Tasks with Little Code Added
• Microsoft Curbs Early Access to Notifications About Cybersecurity Flaws
• Smart Glasses Help Train General-Purpose Robots
• Compact Quantum Logic Gate Demonstrated Using GKP Codes
• School Cuts Costs for Students by 3D-Printing Violins
• Space Station Crew Gains AI Assistant
• Tool Sniffs 5G Traffic to Launch Attacks
• Sensor Survives Being Cut in Half
• AI-Generated Responses Undermine Crowdsourced Research Studies

EU to Curb AI Chip Flows to China as Part of U.S. Trade Deal

Under the terms of the recent EU-U.S. trade agreement, the European Union has agreed to purchase $40 billion of U.S. AI chips and to adopt U.S. security standards to prevent “technology leakage to destinations of concern.” EU trade chief Maros Sefcovic (pictured) stressed that the chips must stay in Europe and benefit its economy, and not be re-exported because they might “fall into the wrong hands.”
[ » Read full article *May Require Paid Registration ]

FBI Warns of Russian Hacks Targeting U.S. Critical Infrastructure

The U.S. Federal Bureau of Investigation (FBI) and researchers at Cisco Talos warn that Russian hacker groups are using a seven-year-old vulnerability in Cisco IOS software to target critical U.S. infrastructure. The Cisco researchers said the hackers are undertaking en masse extraction of device configuration information. The FBI issued a separate advisory noting it had identified incidents over the last year in which hackers collected configuration files "for thousands of networking devices associated with U.S. entities across critical infrastructure sectors."
[ » Read full article ]

Top Law Schools Boost AI Training as Legal Citation Errors Grow

Law schools at the University of Chicago, University of Pennsylvania, and Yale University are among those adjusting their curricula to train students to understand AI’s limitations and to check their work. The changes come after attorneys have been fined or faced sanctions for their usage of AI in legal proceedings, which often includes errors. Said William Hubbard, deputy dean of University of Chicago Law School, “You can never give enough reminders and enough instruction to people about the fact that you cannot use AI to replace human judgment."
[ » Read full article ]

Chipotle Teams Up with Zipline to Test Drone Delivery

Chipotle has partnered with drone delivery company Zipline to launch a pilot program offering autonomous food delivery. Starting in the greater Dallas area of Texas, select users of the Zipline app can order from the full Chipotle menu, with wider access planned in coming weeks. Orders are placed in a “Zipping Point,” picked up by aircraft, and lowered by droid from about 300 feet in the air to customers’ homes.
[ » Read full article ]

Humanoid Robot Does Complex Tasks with Little Code Added

A humanoid robot developed by Boston Dynamics and Toyota Research Institute researchers employs a large behavior model to facilitate the addition of new capabilities without the need for hand-programming or new code. Researchers demonstrated the Atlas robot's ability to self-adjust by interrupting it mid-task with unexpected challenges. Boston Dynamics' Scott Kuindersma said, "Training a single neural network to perform many long-horizon manipulation tasks will lead to better generalization."
[ » Read full article ]

Microsoft Curbs Early Access to Notifications About Cybersecurity Flaws

Microsoft is restricting Chinese companies’ access to its cybersecurity vulnerability program after investigating whether a leak contributed to the recent SharePoint breaches that affected 400 organizations, including the National Nuclear Security Administration. The Microsoft Active Protections Program previously gave partners, including a dozen Chinese firms, early details and proof-of-concept code for flaws. Now, those in countries requiring vulnerability reporting to governments, like China, will receive general written descriptions only when public patches are released.
[ » Read full article *May Require Paid Registration ]

Smart Glasses Help Train General-Purpose Robots

A smart-glasses system developed by New York University researchers aids robot learning to train general-purpose robots more easily. The EgoZero system uses smart glasses from Meta's Project Aria to collect "egocentric" training data from the perspective of wearers performing everyday tasks. The system trained a robot on seven manipulation tasks using only 20 minutes of human data per task, with a success rate of 70% when autonomously completing tasks with a robot.
[ » Read full article ]

Compact Quantum Logic Gate Demonstrated Using GKP Codes

Researchers at Australia's University of Sydney and Q-TRL demonstrated the potential for hardware-efficient quantum computing with the development of a universal quantum logic gate using Gottesman–Kitaev–Preskill (GKP) code encoded in a single trapped ytterbium ion. The researchers entangled two logical qubits stored in the natural vibrations of a single ytterbium atom. They manipulated the ytterbium ion, which was held in a Paul trap at room temperature, to store the error-correcting GKP codes, then entangled the logic gates between the encoded states.
[ » Read full article ]

School Cuts Costs for Students by 3D-Printing Violins

Laura Jacob, superintendent of California Area School District in Pennsylvania, is cutting costs for students interested in music by 3D-printing violins. With over 70% of her students from low-income families, Jacob wanted to make the instruments accessible without the rental fees that run to hundreds of dollars per year. She experimented with online models and eventually succeeded in producing violins for about $50 each. Over five years, she has printed more than 200 violins, and now teaches a weekly violin club.
[ » Read full article ]

Space Station Crew Gains AI Assistant

China’s Tiangong space station crew recently completed their third spacewalk with the aid of a new large-scale AI assistant. Delivered by the Tianzhou 9 cargo craft on July 15, Wukong AI is built on a domestic open-source model tailored for aerospace missions. It supports astronauts with scheduling, mission planning, and data analysis with its intelligent question-answering system.
[ » Read full article ]

Tool Sniffs 5G Traffic to Launch Attacks

An open source framework developed by security researchers at the Singapore University of Technology and Design exploits vulnerabilities in 5G mobile networks during the pre-authentication phase when devices connect to networks. The Sni5Gect framework can detect uplink/downlink traffic with more than 80% accuracy from up to 20 meters away to launch attack payloads. The tool targets the brief unencrypted messaging period between the base station and target handset that occurs during handshaking and authentication.
[ » Read full article ]

Sensor Survives Being Cut in Half

A stretchable sensor developed by researchers at Belgium's Vrije Universiteit Brussel can self-heal after being cut completely in half. The sensor is made with a polymer with Diels-Alder crosslinks that break when damaged and reform upon contact, restoring its original structure within 24 hours at room temperature. The researchers found the sensor performed at 80% capacity after six break-and-heal cycles. They also observed minimal drift, amounting to less than 5% for pristine sensors and under 10% for previously severed sensors, through 800 stretch cycles.
[ » Read full article ]

AI-Generated Responses Undermine Crowdsourced Research Studies

Researchers at Germany's Max Planck Institute for Human Development found crowdsourced research studies may be contaminated by AI-generated responses. In a study of the Prolific platform, they identified 45% of participants copying and pasting content into an open-ended question and noted "overly verbose" or "distinctly non-human" language in the responses. In a second study, the researchers added traps using reCAPTCHAs to distinguish entirely human responses from bot-generated responses.
[ » Read full article *May Require Paid Registration ]