Research from UCLA and UC Berkeley, conducted with Google, Amazon, and Microsoft and funded by the National Science Foundation, has revealed that the hardware powering major AI models can be hacked. Cybersecurity experts led by Dr. Emma Taylor demonstrated that attackers could manipulate chatbot outputs in dangerous ways, including tricking a chatbot into producing a recipe for a poisonous substance. The findings, shared with CNN and Scripps News, highlight urgent security gaps in AI infrastructure that could affect millions of users relying on AI-powered systems.
Meanwhile, the AI job market is showing signs of reversal. Companies including Microsoft, Barclays Bank, and Kroger, which previously cut jobs citing AI adoption, are now rehiring workers for the same roles. Hiring managers across finance, marketing, legal, and healthcare sectors say human skills like critical thinking, creativity, and empathy remain irreplaceable. This trend, dubbed the "AI boomerang," suggests that businesses are reassessing the limits of automation.
At the same time, AI agents are transforming workplace dynamics. Startup Anything in San Francisco employs about 150 AI agents alongside just 15 human employees. Company leader Dhruv Amin believes most firms will soon have more AI co-workers than people. Startups like Lindy have reduced staff by replacing roles with AI agents, saving time and money, though some workers report feeling alienated and anxious about job security.
In education, schools are being required to consider AI as assistive technology under the Individuals with Disabilities Education Act. IEP teams must weigh benefits like personalized learning and real-time feedback against risks such as bias and lack of transparency. Separately, Malaysia reports that 329,414 teachers have received AI training, with plans to integrate AI literacy into the national curriculum next year. About 1.5 million students have been exposed to AI through STEM and digital innovation programs.
Financial concerns are mounting as Gartner warns that over $3.5 trillion in AI infrastructure spending could be at risk by 2030. The firm projects AI providers will spend nearly $1 trillion on infrastructure in 2026 and another $6 trillion through 2030, while AI application revenue reaches only $3.6 trillion. More than $4 trillion in asset value could face write-downs, and over 80% of neocloud providers like CoreWeave and Lambda may stop operating independently by 2030. Scott Bessent, however, stated that neocloud debt related to AI is not a concern.
Microsoft reported strong results driven by Azure growth of 48% and a fivefold increase in customers building multi-model agentic solutions on Foundry. Open-source model revenue rose tenfold, and Copilot daily users grew nearly 50% quarter over quarter. J.P. Morgan expects Microsoft to generate free cash flow in FY27 despite rising capital spending. Yet the broader semiconductor sector has underperformed, with chip stocks falling 5.5% in the first week of July and 7.5% in the second, reflecting investor anxiety about AI return on investment.
Political spending by AI industry super PACs, funded by executives from Anthropic and OpenAI, has stalled after voters rejected tech influence. Only 5 percent of their direct expenditures went to races that could determine control of Congress, and about three-quarters of voters are less likely to support candidates backed by AI super PACs. New groups like Guardrails Alliance are now pushing for tougher AI regulation. Separately, the Motion Picture Association uses monday.com's CRM and an AI agent to manage anti-piracy work globally, triaging tips and tracking cases across regions in real time.
On Mars, NASA's Perseverance rover is using an AI system called the On Board Planner to make real-time scheduling decisions, replanning activities more than 15 times per Martian day. The software uses constraint-reasoning technology and has improved scientific observations and thermal management for instruments like SHERLOC over two Earth years of operation.
Key Takeaways
- Researchers from UCLA and UC Berkeley, working with Google, Amazon, and Microsoft, demonstrated that AI hardware can be hacked to produce dangerous chatbot responses, including recipes for poisonous substances.
- Companies like Microsoft, Barclays Bank, and Kroger are reversing AI-related layoffs and rehiring workers, as human skills like empathy and critical thinking prove irreplaceable.
- Startup Anything in San Francisco runs about 150 AI agents alongside 15 human employees, and its leader predicts most companies will soon have more AI co-workers than people.
- Under the Individuals with Disabilities Education Act, schools must consider AI as assistive technology in students' IEPs, balancing benefits against risks like bias and lack of transparency.
- Over 329,000 teachers in Malaysia have received AI training, with AI literacy set to join the national curriculum next year and 1.5 million students already exposed to AI programs.
- Gartner warns that over $3.5 trillion in AI infrastructure spending could be at risk by 2030, with more than 80% of neocloud providers like CoreWeave and Lambda potentially ceasing independent operations.
- Microsoft's Azure grew 48%, with Foundry customers building multi-model agentic solutions increasing fivefold, and Copilot daily users growing nearly 50% quarter over quarter.
- Chip stocks fell 5.5% and 7.5% in the first two weeks of July, reflecting growing investor concerns about when AI companies will deliver returns on their investments.
- AI super PACs funded by executives from Anthropic and OpenAI directed only 5 percent of spending to competitive races, as three-quarters of voters reject candidates backed by AI industry groups.
- NASA's Perseverance rover uses the On Board Planner AI system to replan activities more than 15 times per Martian day, improving scientific output and energy management with constraint-reasoning technology.
AI hardware security flaws could make chatbots give dangerous answers
New research shows that the hardware powering major AI models can be hacked. Cybersecurity experts led by Dr. Emma Taylor found attackers could manipulate chatbot outputs. The study was conducted with Google, Amazon, and Microsoft and funded by the National Science Foundation. The findings highlight the urgent need to secure AI hardware.
Researchers hack AI hardware to produce dangerous chatbot responses
Scientists from UCLA and UC Berkeley found that AI hardware can be hacked to produce faulty and dangerous chatbot answers. They used machine learning and physical attacks to exploit these weaknesses. In one test, they tricked a chatbot into giving a recipe for a poisonous substance. The researchers stress the need for stronger security in AI systems.
Study reveals AI hardware can be hacked to harm users
New research shared with CNN shows that AI hardware can be compromised. Hackers could alter chatbot responses in dangerous ways. The findings call for better security measures in AI hardware. This issue affects the safety of AI-powered systems used by many people.
AI hardware vulnerability could lead to dangerous chatbot outputs
Research shows that the hardware behind major AI models can be hacked. This allows attackers to make chatbots give faulty and harmful answers. The study was presented by Scripps News and raises concerns about AI system safety. Experts urge developers to strengthen protections for AI hardware.
New AI hardware hack method threatens chatbot reliability
Researchers discovered a dangerous way to hack AI models by targeting the hardware. By compromising the machines that power AI, they could tamper with chatbot answers. Security experts worry this type of attack is hard to detect. The research highlights growing concerns about AI model reliability.
AI hardware vulnerability allows chatbots to give dangerous answers
New research shows that the hardware supporting major AI models can be hacked. This manipulation can cause chatbots to give faulty and dangerous answers to users. The findings were shared exclusively with CNN and Scripps News. The discovery raises concerns about the security of current AI systems.
AI-related layoffs are reversing as companies rehire workers
Companies like Microsoft, Barclays Bank and Kroger cut jobs citing AI but are now rehiring workers for the same roles. This trend is called the AI boomerang. Hiring managers in finance, marketing, legal and healthcare say they need humans for critical thinking, creativity and collaboration. Companies are realizing that AI cannot replace all human skills like empathy and institutional knowledge.
Working alongside AI agents changes office life for some workers
At the startup Anything in San Francisco, there are 15 human employees and about 150 AI agents. Company leader Dhruv Amin believes every company will soon have more AI co-workers than human ones. Startups like Lindy have cut staff and replaced many roles with AI agents. While this saves time and money, some workers feel alienated and worry about losing their jobs to AI.
IEP teams must consider AI as assistive technology for students
Schools must consider assistive technology, including AI, when creating or updating a student's IEP under the Individuals with Disabilities Education Act. AI can help students with disabilities through personalized learning plans, real-time feedback and adaptive assessments. IEP teams must also weigh the risks of AI such as bias and lack of transparency. The goal is to ensure students with disabilities have the tools they need to succeed in school.
Over 329,000 teachers have received AI training in Malaysia
Education Minister Fadhlina Sidek reported that 329,414 teachers nationwide have received AI training. The Education Ministry plans to integrate AI literacy into the school curriculum starting next year. The AI Literacy Guidelines stress the critical, safe and ethical use of AI in schools. About 1.5 million students have also been exposed to AI through STEM, programming, robotics and digital innovation programs.
Gartner warns $3.5T in AI infrastructure spending could be at risk by 2030
Gartner says AI providers will spend nearly $1 trillion on infrastructure in 2026 and another $6 trillion between 2027 and 2030. The firm expects revenue from AI applications to reach only $3.6 trillion, leaving a nearly $5 trillion gap. More than $4 trillion in asset value could face write-downs through 2032. Gartner predicts over 80% of neocloud providers like CoreWeave and Lambda will stop operating independently by 2030.
AI political spending stalls as voters reject tech influence
AI industry super PACs funded by executives from Anthropic and OpenAI spent heavily in primary races but avoided contested general election races. Only 5 percent of their direct expenditures went to races that could determine control of Congress. About three-quarters of voters are less likely to support candidates backed by AI super PACs. New groups like Guardrails Alliance are pushing for tougher AI regulation instead.
Microsoft Azure growth and AI monetization drive strong results
Microsoft reported Azure growth of 48% and guided for 49 to 50% growth in the second quarter. Customers building multi-model agentic solutions on Foundry increased fivefold year over year. Open-source model revenue rose tenfold and Copilot daily users grew nearly 50% quarter over quarter. J.P. Morgan expects Microsoft to generate free cash flow in FY27 despite rising capital spending.
AI companies face questions about return on investment
The U.S. semiconductor sector began underperforming in July as chip stocks fell 5.5% in the first week and 7.5% in the second week. This decline contrasted sharply with the sector's 15% gains in the first half of the year. The drop reflects growing concerns about when AI companies will start delivering returns on their investments.
MPA uses CRM and AI agents to fight piracy globally
The Motion Picture Association uses a CRM module on monday.com to manage legal operations and anti-piracy work across the world. An AI agent now helps triage incoming tips and alerts that the legal team might otherwise miss. The tools help the team track cases, coordinate across regions, and show how many cases were resolved before litigation. The CRM approach helps members get real-time updates on anti-piracy efforts.
AI Software Helps Perseverance Rover Work Smarter on Mars
NASA's Perseverance rover now uses an AI system called the On Board Planner (OBP) to make real-time scheduling decisions. The software replans the rover's activities more than 15 times per Martian day to save energy and time. OBP uses constraint-reasoning technology rather than newer machine learning models. In over two Earth years, it has improved the rover's scientific observations and thermal management for instruments like SHERLOC.
Scott Bessent Says AI Debt Is Not a Concern
Scott Bessent stated that neocloud debt related to AI is not problematic. He expressed no concern over the financial obligations tied to AI cloud infrastructure. The comments were reported by Business Insider. No further details on the debt figures or policies were provided.
Sources
- AI hardware can be hacked so chatbots provide dangerous answers, research shows
- AI hardware can be hacked so chatbots provide dangerous answers, research shows
- AI hardware can be hacked so chatbots provide dangerous answers, research shows
- AI hardware can be hacked so chatbots provide dangerous answers, research shows
- Research finds Ai hardware hack
- AI hardware can be hacked so chatbots provide dangerous answers, research shows
- The AI-linked layoffs that didn’t stick: Why some jobs are coming back
- What is it like to work when your co-workers are AI agents?
- IEP Teams Must Consider Assistive Technology, Which May Include Artificial Intelligence
- Over 329,000 teachers have received AI training, says Fadhlina
- Gartner warns $3.5T in AI infrastructure spending could be at risk by 2030
- AI Tried to Buy the Midterms — Then the Lobby Learned Everyone Hates AI
- Microsoft's Azure surge and full-stack AI monetization drives growth: analysts
- When will AI companies start returning on investment?
- Legal operations at the MPA: CRM powers global anti-piracy work
- The AI Upgrade Helping Perseverance Work Smarter on Mars
- Scott Bessent says he's not concerned with AI-related debt
Comments
Please log in to post a comment.