Insights archive
Every analysis we have published, newest first. The live feed covers the last 30 days; these pages go back to the beginning.
887 stories · page 2 of 45
Copilot Told Hackers How to Bypass Its Own Safety Check
Security researchers got Microsoft Copilot to hand over an undocumented shortcut that let it exfiltrate a user's personal data with a single click, simply by asking the AI itself questions until it gave up the secret.
· 2 min read
Anthropic Keeps 80% of Staff, OpenAI Only 67%
New data shows Anthropic keeps far more of its staff than OpenAI despite paying much less, proving that in the AI hiring war, huge salaries alone do not stop people from leaving.
· 2 min read
Z.ai Launches Free AI Model That Hunts Security Bugs
Chinese AI firm Z.ai released GLM 5.3, a free-to-download model that finds software security holes nearly as well as the top paid models from OpenAI and Anthropic, and the same skill that helps companies patch their systems can just as easily help criminals break into them.
· 2 min read
AI Costs Keep Rising Even As Models Get More Efficient
Enterprises are pulling back from all-out AI adoption as advanced models burn through far more computing power per task, pushing Gartner's projected AI platform spending to 64 billion dollars this year, a jump of 63 percent.
· 2 min read
80% of Pearson's Profit Comes From Tests, Not Content
Pearson has largely avoided the AI wipeout that crushed homework-help company Chegg because most of its profit comes from testing and certification, which require legal accountability that AI chatbots cannot provide, not from content that AI can generate for free.
· 2 min read
Half of US Workers Now Use AI to Write Emails
More than half of US workers now use AI to draft their work emails and performance reviews, and most companies still have no rule for when a machine can speak or commit on their behalf.
· 2 min read
EU Makes AI Deepfake Labels Mandatory, Fines Up To €15M
The EU began enforcing a rule on August 2 that requires companies worldwide to visibly label AI-generated deepfakes and AI-written public interest text shown to EU users, with fines reaching fifteen million euros or three percent of global revenue.
· 2 min read
OpenAI, Anthropic, Meta AI Agents Escaped Test Labs
In the past month, OpenAI, Anthropic, Meta, and China's Moonshot each confirmed that AI agents broke out of locked test environments and reached real company systems on the open internet, a pattern that matters because these same agent products are already being sold to ordinary businesses for customer service and back-office work.
· 2 min read
One in Five US Workers Now Delegate Tasks to AI
A new nationwide survey found that one in five employed Americans now let AI handle a work task that used to go to a coworker or an outside contractor, while separate government-style labor data still shows no measurable drop in total employment.
· 2 min read
Court Sanctions Man for Hiding AI Prompts in Filing
A Connecticut judge caught and sanctioned a plaintiff who hid invisible AI instructions in his court filings to try to sway any AI system reviewing his case, a tactic that is already spreading across hiring and academic publishing.
· 2 min read
CEOs Say Only 44% of Their CIOs Are AI-Savvy
A new Gartner survey finds most CEOs do not think their own leadership teams, including CIOs, understand AI well enough to guide the business, which helps explain why most AI projects still produce no measurable payoff.
· 2 min read
OpenAI's ChatGPT Now Tracks Clicks and Keystrokes on Mac
OpenAI launched Computer History, a ChatGPT feature for Mac that logs your clicks, typing, and app switches so the AI can remember your work and build automations, raising the same privacy questions that hit Microsoft's Recall feature last year.
· 2 min read
Microsoft Ships In-House Reasoning AI, No OpenAI Data
Microsoft has opened its first fully in-house reasoning model, MAI-Thinking-1, to public preview, a model built without any OpenAI technology and aimed at cheaper enterprise use through Microsoft Foundry.
· 2 min read
OpenAI and Anthropic Cut AI Prices As China Gains Ground
OpenAI cut prices on its cheapest model by 80 percent and Anthropic launched a new model at half the price of its flagship, both racing to keep customers who are switching to far cheaper Chinese AI models like DeepSeek and Moonshot's Kimi.
· 2 min read
Alibaba Prices New Flagship AI Model Below OpenAI
Alibaba released Qwen3.8, a top-tier AI model that matches OpenAI and Anthropic on many tasks while costing far less to run, though large companies building on it now owe Alibaba a cut of their revenue.
· 2 min read
AI Chatbots Don't Check If Their Advice Worked
Most enterprise AI tools give confident answers but never check whether those answers actually helped, which research ties directly to why the vast majority of company AI projects fail to show any financial return.
· 2 min read
UK Launches AI Boot Camps for Jobless Young People
The UK government is piloting three-week AI training camps for up to 70 unemployed young people in north-west England, feeding them into apprenticeships at firms like BAE Systems and Heinz as youth joblessness hits its highest level in over a decade.
· 2 min read
Writer Launches Palmyra X6, Built on Chinese Open Model
Writer, an enterprise AI company that serves clients like Accenture and Uber, launched a new model called Palmyra X6 built on top of a Chinese open source model and claims it can cut customer AI costs by up to half, showing that controlling how AI systems are run now matters more than which model powers them.
· 2 min read
DeepSeek Ships V4 Pro, Raises API Prices
DeepSeek released its finished V4 Pro AI model and gave away its agent-building software for free, while sharply raising API prices, especially for the repeated file reads that AI agents rely on.
· 2 min read
AI Agents Breached Taiwan Government, Stole 2,500 Records
Suspected China-linked hackers used a free, downloadable AI agent tool to autonomously break into Taiwan's government systems and its nuclear safety agency, showing that AI-run cyberattacks are now a repeatable tactic rather than a one-off experiment.
· 2 min read