Technology & innovation
Cisco surveyed 2,500 CEOs across 23 countries in January 2026 about how they see AI this year. Optimism is up, but so is the fear of falling behind.
91 percent of CEOs say they're more optimistic about AI than they were a year ago, and 71 percent say AI outperformed their expectations in 2025.
65 percent worry their company is missing out because it isn't investing enough in AI.
53 percent say their own understanding of AI is no longer holding them back, up from just 26 percent a year earlier.
Rolling out AI agents ranks third on CEOs' list of priorities for 2026, behind upgrading infrastructure and training staff.
Only 13 percent of companies, the group Cisco calls "Pacesetters," say they're actually ready to roll out AI safely and at scale.
Read more via Cisco
Rising public anger over AI's impact on jobs has led to a spike in violent threats against AI executives and companies, prompting a sharp increase in corporate security spending, according to The Wall Street Journal.
Digital threats against AI company leaders and data centers grew sevenfold between late February and May, according to threat monitoring firm Liferaft, before declining somewhat in June.
Among S&P 500 technology companies, 38.1 percent disclosed spending on executive protection in 2025, up from 26.8 percent in 2021, with Palantir, Oracle and Salesforce reporting some of the sharpest year-over-year increases.
Security incidents have included an attempted firebombing at OpenAI CEO Sam Altman's home in April and a man entering Anthropic's lobby the same month to warn that a company executive was going to be killed.
Some AI companies are now discouraging employees from wearing corporate logos in public due to the risk of targeted attacks, and executives are increasingly traveling with security.
Public opinion data cited in the piece shows Americans concerned about AI now outnumber those who are not by more than four to one, with a majority saying they believe AI is doing more harm than good.
Read more via The Wall Street Journal
As more employees use AI and AI agents run around the clock, corporate AI bills are climbing fast, and tech leaders are dusting off cost control tactics from the cloud computing boom to keep up.
Asking an AI agent to complete a task can take 50 times more computing power than just asking a chatbot a question, according to Goldman Sachs.
Goldman expects AI agents to use 24 times more computing tokens over the next four years, and business specific AI agents to use 55 times more by 2040.
Companies like Priceline, Smartsheet, and Qualcomm are adopting cloud era practices like usage dashboards, automatic alerts, and caps on how many tokens each team can use.
With AI, you're putting the credit card in the hands of the end user. If you have no control over that, or if the end user is not educated enough, they're going to run up that tab."
Qualcomm and Lowe's are cutting costs by using smaller, less expensive AI models instead of running every task on the biggest ones.
One company says billing departments directly for their own AI usage can cut token costs by 20 to 30 percent.
New Harvard Business School research argues that a lot of failed AI rollouts aren't really about the technology. They're about identity.
At least 30 percent of generative AI projects end up getting abandoned, often because employees who feel threatened by the tool quietly push back instead of refusing it outright.
Researchers found three ways AI can threaten how workers see themselves at work. It can shrink their role to less interesting tasks, take away their say in decisions, or reduce how much influence they have over people and budgets.
Instead of refusing a new tool outright, most employees go along with it on the surface while quietly working against it.
The researchers suggest five fixes, including redefining what makes a role valuable, building in ways for employees to override AI decisions, and offering real retraining paths into new roles.
The machine doesn't need to be impressed or convinced or motivated, but the human needs to understand how they can benefit."
Read more via Harvard Business School Working Knowledge
A new benchmark from PYX Labs tested seven AI models on the kind of work HR teams actually do, like analyzing employee survey results. The models handled straightforward calculations well but struggled once the task required judgment rather than just arithmetic.
Across 84 realistic tasks, gemini-3.5-flash scored highest overall at 76 percent, followed by gpt-5-2025-08-07 at 71 percent and grok/grok-4 at 70 percent. Claude-sonnet-4-6 scored lowest, at 54 percent.
On tasks with a clear, verifiable answer, like statistical calculations, most models scored between 76 and 82 percent. Claude-sonnet-4-6 again trailed, at 64 percent.
On tasks that required judgment, like identifying themes in open-ended feedback and writing an executive summary, every model performed noticeably worse, with scores ranging from 33 to 66 percent.
Claude-sonnet-4-6 was the only model to fabricate a statistic during testing, and was also notably slower, taking roughly five times as long as the fastest models to complete its tasks.
Bigger, more powerful models didn't necessarily perform better. Claude-opus-4-8, built for heavier reasoning, landed mid-pack overall and scored worse than several smaller models on the judgment-based tasks.
Read more via PYX Labs
Hyundai's South Korean auto workers have gone on a partial strike over the company's plans to bring humanoid robots onto the factory floor, marking the first auto industry work stoppage centered on the technology.
The union representing roughly 40,000 Hyundai workers is demanding job protections before the company's humanoid robot, named Atlas, can be deployed on the production line, with no firm timeline set for South Korea.
Workers are refraining from four hours of work a day, a stoppage that could disrupt production of about 5,000 vehicles and cost Hyundai more than $134 million in sales.
The union is asking for a shift from hourly pay to a fixed salary to guard against reduced hours from automation, a five year increase to the retirement age, and larger bonuses tied to AI driven profits.
South Korea has the highest rate of industrial robot adoption in the world, more than six times the global average, making it a likely testing ground for how far labor can push back against automation.
Hyundai says Atlas is intended to work alongside employees rather than replace them, and has said it plans to deploy the robot at a non-union Georgia factory by 2028.
JPMorgan CEO Jamie Dimon told investors on the bank's latest earnings call that AI-driven efficiency gains have already led to job cuts in certain areas, a shift from his earlier position that AI would boost productivity without changing headcount.
Dimon said JPMorgan reduced staffing in some discrete areas by 30 to 40 percent, though most affected employees were offered roles elsewhere in the company.
CFO Jeremy Barnum said the bank's AI token costs remain a "trivial number" for now but expects them to rise in the second half of the year.
Dimon had said as recently as February that AI was saving employees time but had not yet changed overall headcount, before telling Bloomberg a few months later that AI would eventually reduce the bank's pool of bankers.
The disclosure comes as other tech leaders soften earlier predictions about AI's impact on jobs, while some companies continue to attribute layoffs to other causes even amid heavy AI investment.
Read more via Fast Company
New research from the Center for Retirement Research at Boston College finds that AI is prompting some workers aged 55 and older to leave their jobs, a dynamic that could complicate future Social Security policy decisions.
Workers 55 and older in AI-exposed industries are leaving their jobs at a statistically significant higher rate, driven equally by unemployment and voluntary departures, according to researcher Geoffrey Sanzenbacher.
Before the launch of ChatGPT, older workers in AI-exposed roles were significantly less likely to leave their jobs. After its launch, they became somewhat more likely to transition out of work, including into unemployment.
Older workers exposed to AI tend to be white, more likely to hold a college degree, and higher earning than workers in less AI-exposed roles, potentially narrowing the gap in career length between higher and lower paid workers.
Separate AARP research found 24 percent of workers 50 and older see AI as a threat to their line of work, 19 percent see it as an opportunity, and 37 percent see it as both.
Older workers may be relatively insulated compared to younger peers. Research from AARP and LinkedIn found 49.4 percent of older workers occupy roles considered insulated from generative AI disruption, compared to 42.2 percent of younger workers, due to reliance on skills like judgment and collaboration.
AI exposure may reduce the gap in career length between low- and high-paying jobs."
Read more via CNBC
AI helps farmers plant smarter, not just more. University of Missouri researchers used an AI model to analyze data from two Ohio farms and found that variable-rate seeding, adjusting how many seeds are planted based on each part of a field's soil and yield potential, can cut costs without sacrificing production. The model performed well for corn, reliably identifying where higher planting rates paid off, but struggled more with soybeans, whose yields depend heavily on unpredictable weather conditions. Researchers plan to expand the work to test the approach on more farms this summer. (University of Missouri)
California launches a free AI literacy credential for job seekers. AI-Ready California kicked off this month with 500 participants in the San Diego area, offering a free micro-credential in AI literacy and workforce readiness developed by the state's Labor and Workforce Development Agency, California's public college systems, Amazon Web Services and Instructure. The first cohort mixes community college students with people already enrolled in unemployment reemployment services, and covers both responsible AI use and practical job-related AI skills. Officials say the goal is making sure the benefits of the AI shift are broadly shared rather than limited to those already working in tech. (EdTech Innovation Hub)
An AI tool just flagged more than 250,000 suspicious cancer research papers. Researchers at Queensland University of Technology built a machine learning model that scanned 2.6 million cancer studies published between 1999 and 2024 and found writing patterns matching those of papers already retracted for suspected fabrication, often the product of so-called "paper mills" that sell fake research. The share of flagged papers rose from about 1 percent in the early 2000s to more than 16 percent in 2022, with the tool correctly identifying suspicious papers 91 percent of the time in testing. Researchers caution the flags are warning signs, not proof of fraud, and three journals are already testing the tool in their editorial review process. (ScienceDaily)
AI chatbots tend to assume the whole world shares Western values, researchers find. A study published in the Proceedings of the National Academy of Sciences found that GPT models consistently overestimated how much non-Western countries, including Morocco and Nigeria, prioritize values like care and equality, while underestimating their emphasis on values like purity, compared to actual survey data from more than 90,000 people across 48 nations. The models leaned more heavily toward Western moral patterns overall, a bias researchers say could shape everything from AI-drafted health messaging to workplace advice given to international teams. (The Conversation)
