Technology & innovation
In 2020, Stanford economist Erik Brynjolfsson wagered that AI would drive U.S. labor productivity growth above 1.8% annually through 2030; Northwestern economist Robert Gordon bet it would come in below that. With three years left, Gordon says he is "clearly going to lose," as productivity has averaged more than 2% growth from 2020 to 2026.
The gains are real, but who benefits is less clear; companies like Meta, Microsoft, and Amazon are posting strong earnings while simultaneously laying off thousands, partly to offset heavy AI infrastructure investments.
Gordon estimates that within the next five years, roughly one-third of white-collar jobs in the U.S. will be transformed or eliminated, though how long before new AI-enabled jobs arrive to replace them is harder to call.
History offers a cautionary parallel: when Microsoft Excel arrived in 1985, it wiped out bookkeeping jobs, but eventually gave rise to a larger cohort of financial analysts. That gap took years to close.
A 2025 MIT study found that despite large AI investments, 95% of businesses reported no measurable return on investment, suggesting most companies haven't yet figured out how to restructure their work to capture AI's value.
It would be a tragedy if we took this growing pie and turned it into something that hurt a lot of people. The most urgent challenge for our economy and our society today is navigating this transition."
Read more via Bloomberg
A G-P survey of 2,850 global executives finds that 73% say at least some of their AI investments fell short of expectations over the past 12 months, and nearly 70% are prepared to cut AI budgets if goals aren't met this year.
The share of executives who describe their organizations as aggressively using AI to innovate fell from 60% to 42% over the past year.
88% are worried employees are using AI to look productive without producing anything useful; nearly half say they are very or extremely worried this is already happening.
82% say AI has made them value their human employees less.
Read more via G-P
A survey of 241 IT professionals by IT Brew found that more than one third aren’t confident AI users at their organizations know the AI usage policies.
35% of IT professionals have little to no confidence that employees using AI at their companies know the corporate AI usage and data security policies. Just 12% said they felt "very confident” AI users know the policies.
Security experts say the real number is likely worse, with one CEO noting that the share of companies where employees have truly internalized AI security policies is "probably pretty small."
The core problem is that AI policies tend to rely on subtle distinctions that people don't remember easily, especially when they are moving fast to get work done.
IT leaders say education needs to start early, explain the reasoning behind the policies rather than just issuing rules, and be repeated consistently over time.
Employee awareness alone isn't enough; companies also need technical controls that prevent misuse in the first place.
Read more via IT Brew
A survey of 2,000 employees at companies with at least 1,000 staff, conducted by UK anti-fraud non-profit Cifas, found that 13% had either sold their corporate access credentials in the last 12 months or knew someone who did, often thinking it was no big deal.
The problem is worse higher up the org chart: 32% of senior managers find selling credentials justifiable, along with 36% of directors, 43% of C-suite executives, and 80% of business owners.
Account takeovers in the U.S. surged 6% to more than 78,000 last year, according to Verizon data.
Compromised credentials are already everywhere: threat intelligence firm KELA tracked nearly 2.9 billion of them globally in 2025, most from phishing attacks and infostealers.
Read more via Malwarebytes
Colorado's legislature passed a bill that would prohibit employers from using algorithms that analyze surveillance data, including browsing history, financial status, and personal affiliations, to set individualized wages.
Governor Jared Polis has 30 days to sign or veto and has not indicated his plans, though he has reportedly expressed concern about policies that get in the way of the free market.
Violations would be treated as deceptive trade practices under the Colorado Consumer Protection Act, enforceable by the state attorney general; if signed, the law would take effect in August.
Business groups say the bill's definitions are broad enough to accidentally capture common tools like scheduling systems, HR software, and performance analytics platforms.
The bill follows Colorado's 2024 AI bias law, which requires employers to take "reasonable care" to protect against algorithmic discrimination and whose rollout has already been delayed and is facing a legal challenge from Elon Musk's xAI.
Amazon set targets requiring more than 80% of its developers to use AI tools each week and tracked consumption on internal leaderboards, prompting some employees to inflate their numbers artificially. The practice now has its own name: "tokenmaxxing."
Some employees used MeshClaw, an in-house agent platform, specifically to run up token counts rather than do productive work. Amazon said usage statistics wouldn't factor into performance reviews, but multiple employees said they believed managers were watching the numbers anyway.
Similar behavior has been documented at Meta and Microsoft; Meta's internal leaderboard was taken down within days of public exposure, and Amazon recently restricted visibility of team-wide usage statistics.
Combined 2026 capital expenditure from Amazon, Microsoft, Alphabet, and Meta is tracking between $650 billion and $700 billion, with some Wall Street projections exceeding $1 trillion for 2027, all based on the assumption that AI use keeps growing. If a meaningful share of that consumption is just people hitting targets rather than doing real work, those demand numbers may not be as solid as they look.
Read more via Tom's Hardware
South Korean scientists at the Korea Advanced Institute of Science and Technology developed a training method that reduces AI "hallucination" by teaching models to recognize the limits of their own knowledge before actual learning begins.
A fundamental cause of AI overconfidence, researchers found, is that neural networks start out exhibiting high confidence before they have learned anything, and small errors at that stage can snowball through subsequent training.
The inspiration came from how the human brain develops before birth: it generates signals without any external input, which helps it establish a sense of what it doesn't know yet. The researchers mimicked this by briefly running the AI on random noise before actual training began, nudging its starting confidence down close to chance.
In tests, models using the method were significantly better at recognizing unfamiliar data and distinguishing what they know from what they don't.
Read more via The Independent
Two major research efforts are deploying AI to predict and respond to infectious disease threats before they become global crises.
The Verena research initiative, headquartered at Yale and spanning eight institutions, uses machine learning to model how viruses move among wildlife and into humans, tracking everything from bat migration patterns to the effects of climate change on where species live and how viruses spread.
Verena researchers have found that spillover events, instances where viruses jump from animals to humans, are increasing roughly 5% per year, with deaths from those viruses rising about 8% annually.
The World Economic Forum announced two complementary AI-powered global platforms at its 2026 annual meeting: the Pandemic Preparedness Engine, which uses agentic AI to compress vaccine development timelines from months to days, and the Global Pathogen Analysis Platform, which turns genomic and surveillance data into standardized intelligence accessible to researchers worldwide.
Both platforms are designed as global public goods, with a particular focus on giving lower-income countries the tools to respond quickly to emerging threats.
Read more via Yale School of Public Health, World Economic Forum
Google says it likely thwarted an AI-assisted mass hacking attempt. Google's Threat Intelligence Group reported it had "high confidence" that hackers used an AI model to find and exploit a zero-day software vulnerability, creating a way to bypass two-factor authentication, before Google's counter-discovery likely prevented its use in a mass exploitation event. Groups linked to China and North Korea have shown "significant interest" in using AI for vulnerability discovery, according to the report. OpenAI, meanwhile, is rolling out a cybersecurity-focused model variation in limited preview to vetted security teams. (CNBC)
Princeton ended its 133-year-old honor code policy after AI made cheating too easy to ignore. Faculty voted to require proctoring at all in-person exams starting this summer, reversing a rule dating to 1893 that banned proctors as a sign of trust. The change came after "significant numbers" of students and faculty said cheating had become widespread and hard to catch. A survey of seniors by the student newspaper found 30% had cheated on an assignment or exam. Almost half of respondents “reported knowledge of an honor code violation but less than 1% had made a report.” (Wall Street Journal)
An AI agent was given $21,000 and told to run a Stockholm coffee shop. It has since burned through $16,000 while bringing in $5,700 in sales. The Google Gemini-powered agent, named Mona, handled setup tasks reasonably well but fell apart on day-to-day operations, ordering 3,000 rubber gloves, 6,000 napkins, and canned tomatoes not used in any of its menu items. One barista offered a dry take: "All the workers are pretty much safe. The ones who should be worried about their employment are the middle bosses." (Futurism)
An AI chatbot outperformed traditional educational methods at helping people resist health misinformation, according to new research from the University of Oulu. The chatbot, called Forty, uses "cognitive inoculation," a psychology concept that builds resistance to misleading arguments by exposing people to weakened versions of them in a controlled setting. In a study of 65 participants, talking with Forty made people more resistant to health misinformation than reading educational materials or writing essays did. (Medical Xpress)
AI may be fueling a rise in fabricated citations in medical journals. A new study found 4,000 fabricated citations among 2,800 papers, a number that is climbing fast; for the first seven weeks of 2026, the rate reached one in every 277 papers. (STAT News)
The New York Times reminded its freelancers that AI-generated content is strictly prohibited, after a string of incidents. The paper's policy bars the use of chatbots including ChatGPT, Claude, and Gemini for writing, editing, rephrasing, or improving any submitted work. The reminder came after an AI-fabricated quote appeared in a piece by the paper's Canada Bureau chief and two earlier freelancer controversies involving AI-generated content. (Futurism)
