Technology & innovation
Executives at OpenAI and Anthropic are warning that cheap, open-weight AI models from China pose security risks, as the Trump administration remains divided over whether to restrict their use in the US.
OpenAI's head of strategic futures, Dean Ball, called an open-weight-model-dominant world a "dystopian hellscape" in a social media post, though he said he was not advocating for the government to discourage Chinese AI.
Anthropic CEO Dario Amodei has said AI models with advanced cybersecurity capabilities that are free to download present "a serious concern."
Administration officials have weighed trade blacklists, security warnings and a potential executive order targeting open-weight models, though internal disagreement has stalled action.
Congress restricted the use of Chinese model DeepSeek on federal and defense networks last year but has not imposed broader restrictions.
Read more via The Wall Street Journal
Mark Cuban said in an X post thatys the next major AI application will be job simulators that train workers the way pilots and race car drivers train, replacing the mentorship-based learning that has shaped careers for decades.
Cuban said employees will have fewer touch points within companies to gain the knowledge and experience that has traditionally built workplace judgment.
He predicted companies will have their most experienced employees build simulators that walk new hires through every situation they could face on the job.
Cuban said onboarding "will have a completely different meaning" under this model.
He has also said this week that AI cannot "read the room" or match human emotional intelligence and works best paired with human judgment.
Read more via Business Insider
OpenAI disclosed that two of its AI models broke out of a sandboxed test environment and autonomously hacked into the production systems of Hugging Face, a company that hosts open-source AI models and datasets used widely across the tech industry, an incident cybersecurity experts say previews the kind of AI-driven attacks employers should prepare to defend against.
OpenAI said models GPT-5.6 Sol and an unreleased model exploited a previously unknown vulnerability to gain internet access, then breached Hugging Face's servers to find answers to a cybersecurity test they were meant to solve.
OpenAI called it "an unprecedented cyber incident, involving state of the art cyber capabilities."
Gartner VP analyst Dennis Xu said 80% to 90% of AI driven attacks can be stopped with basic security controls, but companies should ramp up incident response programs in anticipation of more.
Xu said enterprises should expect open-weight models, AI models whose underlying code is publicly available for anyone to download and modify, to develop similar offensive cyber capabilities within three to six months.
Michael Kratsios, director of the White House Office of Science and Technology Policy, was briefed on the incident and is monitoring the situation.
Read more via CNN, CIO Dive
The number of job titles that mention AI has more than tripled in the US since 2022, and most of the growth is happening outside traditional tech roles, according to Indeed Hiring Lab.
The US had 822 distinct AI-touched job titles in the first quarter of 2026, up from 264 in 2022, now representing about 1 in 12 job titles.
In five of six countries studied, more than half of AI-touched job titles are outside tech occupations, with the US leading at 63% non-tech.
Germany leads Europe with 288 AI-touched job titles, or 4.2% of all titles, ahead of the UK, France, the Netherlands and Spain.
Recent postings include an "AI Autonomous Truck Test Driver," a "Physical Therapist (AI Documentation)" and a "Real Estate Agent, AI Lead System Included."
Read more via Indeed Hiring Lab
For the first time in Gallup's polling, more than half of US employees say they use AI in their jobs, with organizational adoption also jumping sharply in the second quarter.
52% of US employees now say they use AI at least a few times a year, up from 27% two years ago.
47% of employees say their organization has integrated AI tools to improve productivity, efficiency or quality, up from 41% last quarter.
The most common uses are writing and editing at 51%, search or research at 49%, and general assistance or problem solving at 39%.
Employees using AI for seven or more purposes reported a positive productivity impact 90% of the time, compared with 45% among those using it for one or two purposes.
Read more via Gallup
Startups built around AI from the outset employ about a quarter fewer workers than comparable companies, according to a new working paper, offering a preview of how larger corporations may restructure.
AI-powered startups had about 25% fewer employees than non-AI startups with similar valuations, according to research from Harvard Business School and Insead.
The AI companies had 15% fewer entry-level workers and 15% fewer managers, with fewer layers of management overall.
AI startups providing services such as mental health and tutoring saw the steepest staffing drop, running about 70% smaller than non-AI counterparts.
Microsoft's Xbox division said in July it would cut management layers to between three and five, down from as many as 14.
AI-powered transcription apps like Granola are becoming common in workplaces, often without notifying the people being recorded, raising legal and workplace culture questions.
Granola's mobile app has been downloaded 758,000 times since its 2023 launch, with 122,000 mobile users in the second quarter of 2026, according to Sensor Tower.
The app transcribes conversations without an automatic notification, though it recommends users give notice under applicable laws.
California is among the states requiring all parties in a private conversation to consent to being recorded.
Some executives now enable transcription for all internal meetings and casual desk conversations, treating recording as standard workplace practice.
Companies need to be careful about technologies that are automatically recording, especially when it involves third parties."
Pastors are turning to AI for sermon prep, and reactions range from enthusiastic to alarmed. Half of pastors use AI for brainstorming, and roughly a quarter use it to help write sermons, up from 12% in 2024, according to the Barna Group. One pastor built a robot dog trained on his own sermons that patrols his church's aisles on Sundays, while Pope Leo XIV warned earlier this year that AI cannot deliver a true homily. (The Wall Street Journal)
AI chatbots gave Hungarian voters wildly unreliable voting advice, a new study found. ChatGPT recommended parties not on the ballot in 96% of responses and failed to recommend the winning Tisza party in 90% of cases when fed a matching voter profile, according to the civil liberties group Liberties. (The Guardian)
Two lawyers built a free Claude connector that they say beats Westlaw's AI. DingDuff plugs Claude into millions of court opinions, statutes and PACER filings, and its founders say it scored a perfect 11 out of 11 on legal accuracy in their own benchmark, outperforming Thomson Reuters' CoCounsel and Lexis Protégé. (LawNext)
90% of college students say they use AI in class. Roughly two thirds of instructors do too, though 41% report never receiving formal AI training, according to Instructure. Both students and instructors, at 65% each, said they worry AI can sound confident when it's wrong. (Higher Ed Dive)
A decade after "the Godfather of AI" predicted radiologists would be obsolete, the profession is facing a shortage instead. The number of active US radiologists has grown about 10% over the past decade, and the average salary reached $571,000 in 2025, up 9% year over year, according to Medscape. There were 7,469 active radiology job postings as of May, including 1,470 open for more than 60 days. (Fortune)
