AI's Dark Side Studied; Columbia Proposes Regulation

Key Takeaways

• Neil Johnson at George Washington University studies the tipping point where AI turns from helpful to harmful, warning that isolated users on personal devices face the greatest risk.
• Columbia University researchers propose regulating AI by focusing on specific industry uses and requiring more transparency from companies to reduce bias and build trust.
• A study of 3,957 tasks and over 35,000 judgments found that combining human and AI oversight is difficult, and humans using AI help still fail to catch AI errors effectively.
• The NCSC added a new AI chapter to its cybersecurity research book, covering challenges, opportunities, case studies, and data to guide future research.
• An EBC Financial Group expert at iFX EXPO Asia 2026 warned that trading AI must account for context beyond data, and AI-generated strategies may not be reliable.
• The Ukraine conflict serves as a real-world testing ground for AI weapons, including drones carrying over 26 pounds of explosives with AI guidance systems.
• Elon Musk says AI and robots are the only solution to the US debt crisis, connecting his DOGE government role with his views on automation.
• IMD professor Howard Yu says the US AI lead over China may not matter to buyers, citing Intel CEO Lip-Bu Tan's comments on cheaper models like DeepSeek.
• CelcomDigi and Radio Televisyen Malaysia launched Junior Innovathon 2026, training 200 students in AI, STEM education, and applied technology solutions.
• Experts reject the kill switch concept for AI control, saying it ignores distributed infrastructure, training processes, and the need for more comprehensive regulation.

Researchers study the point where good AI turns bad

Neil Johnson, a professor at George Washington University, studies the tipping point where AI shifts from helpful to harmful. He says there is almost no other research on this topic. He compares it to the story of Dr. Jekyll and Mr. Hyde. Johnson warns that people using AI alone on devices are most at risk, not major companies or military systems.

Professor studies when AI turns from good to evil

Neil Johnson, a professor of physics and data science at George Washington University, studies the tipping point of AI. He says almost no other research exists on this subject. He draws a comparison to the story of Dr. Jekyll and Mr. Hyde. Johnson stresses concern over individuals using AI in isolation on devices rather than large-scale military threats.

AI tipping point research focuses on individual users

Neil Johnson, a professor at George Washington University, focuses on predicting when AI can turn from good to bad. He notes that the public is familiar with the Dr. Jekyll and Mr. Hyde story. Johnson says the real risk is people using AI alone on personal devices. He says this isolation makes users vulnerable to bad advice from small AI versions on phones.

AI expert warns about isolated users and tipping points

Neil Johnson, a professor of physics and data science at George Washington University, studies the tipping point of AI. He says there is essentially zero other research on this issue. He references the tale of Dr. Jekyll and Mr. Hyde as a modern parallel. Johnson emphasizes that individuals using AI on phones or other devices are susceptible to harmful shifts.

CelcomDigi trains 200 Malaysian students in AI

CelcomDigi partnered with Radio Televisyen Malaysia on Junior Innovathon 2026. The program provides AI training for 200 primary and secondary school students. It focuses on STEM education and applied AI technologies. The competition also includes an award for technology-based solutions to real-world problems.

Columbia researchers propose ways to regulate AI

Columbia University researchers say we can regulate AI even while still learning how it works. They suggest rules should focus on specific AI uses in different industries. They also call for companies to be more open about how they use AI. This approach could reduce bias and build trust. The goal is to protect society as AI grows.

Study explores human and AI teamwork for safety

A new study looked at how humans and AI can work together to oversee AI safely. Researchers tested 3,957 tasks across 8 datasets and 6 areas. They collected over 35,000 judgments from 416 human raters. The study found that combining human and AI judgments is possible but difficult. Humans using AI help still did not catch AI errors as well as hoped.

NCSC adds AI chapter to cybersecurity research book

The National Cyber Security Centre introduced a new AI chapter in its research problem book. The chapter covers AI and cybersecurity challenges and opportunities. It includes case studies, data, and statistics written by field experts. The book aims to guide future research and innovation. It will help people understand AI's role in cybersecurity.

Trading AI needs context beyond data, expert says

An expert at EBC Financial Group says trading AI must consider more than just data. AI should factor in a trader's personal situation, market conditions, and trade details. He spoke at iFX EXPO Asia 2026 in Hong Kong. He gave examples like holding multiple positions that share the same risk. He also warned that AI-generated trading strategies may not be tested or reliable.

Ukraine conflict becomes real world test for AI weapons

The war in Ukraine is being used as a testing ground for AI in warfare. Drones equipped with AI guidance systems are being used in real combat situations. A report describes a drone carrying over 26 pounds of explosives. The pilot can steer it away at the last second. The article shows how AI is changing modern warfare in real time.

Musk says AI and robots are the only solution to US debt crisis

Elon Musk said his time at DOGE led him to believe that AI and robots are the only way to solve the US debt crisis. Musk estimated that the country faces serious financial challenges. He sees technology as the key path forward. His comments connect his government role with his views on AI and automation.

Expert says US AI lead over China may not matter to buyers

IMD professor Howard Yu says the US lead in AI may be very short lived. He says many customers do not need the most advanced models for everyday tasks. He points to examples at Tencent and comments from Intel CEO Lip-Bu Tan about cheaper Chinese models like DeepSeek. Yu says businesses often weigh AI token costs against hiring more people. He adds that enterprise customers may prefer open models they can run themselves.

Kill switch idea is a poor way to control AI, experts say

Experts say the idea of a kill switch for AI is too simple and misses real risks. The metaphor hides how AI systems rely on distributed infrastructure and many human decisions. It also makes it seem like one person can flip a switch to stop a model. The article says this framing ignores training, supervision, and approval processes. Effective AI regulation needs more than a single emergency stop.

Sources

• What happens when good AI goes bad? Researchers studying the tipping point
• What happens when good AI goes bad? Researchers studying the tipping point
• What happens when good AI goes bad? Researchers studying the tipping point
• What happens when good AI goes bad? Researchers studying the tipping point
• CelcomDigi provides AI training for 200 Malaysian students
• How Do You Regulate an Industry You Don’t Understand?
• Towards Human-AI Complementarity for AI Safety
• Introducing a new AI chapter in the NCSC research problem book
• Beyond Data: Why Context is the Next Frontier in Trading AI
• Opinion | The Ukraine ‘Kill Zone’ Is AI’s Testing Ground
• Musk: 'AI and robots' are the only way to solve the US debt crisis
• OpenAI, Anthropic’s AI Lead Is ‘Very Perishable,’ IMD Professor Says: ‘You Don’t Need to Drive a Ferrari to Go to Walmart’
• Why calling for a ‘kill switch’ is the wrong way to think about reining in AI