Technology & innovation
Workers spend the equivalent of 47 workdays a year using AI, but roughly 20 of those days are lost to troubleshooting errors and iterating on prompts rather than productive output, according to a BambooHR survey of more than 1,600 full-time salaried workers across the U.S.
42% of AI time goes to troubleshooting errors and prompt iteration, while just 35% goes to productive work.
63% of organizations have raised their AI tool budgets in response to employee usage, without clear evidence the investment is paying off.
54% of organizations lack a clear, documented AI usage policy that is consistently communicated.
59% of workers use personal AI accounts for work tasks; of those, 71% have entered client data, proprietary strategy, or other sensitive company information into tools their employer cannot see.
62% of employees use company-provided AI tools for personal purposes, including 27% who use them to help with a job search and 17% who use them for a side business or freelance work.
67% of HR professionals say they would trust AI to conduct a candidate interview on its own, compared with just 34% of employees; 80% of HR professionals also admit to finding bias problems in the AI hiring tools they use, including 33% who have observed a pattern of demographic bias specifically.
Read more via BambooHR
A growing share of freelance creative work is shifting from original projects to fixing AI-generated output, with demand for cleanup gigs rising sharply across major platforms even as traditional creative work contracts.
Job listings tagged with phrases like "correct AI," "AI hallucination," and "AI error" rose 87% on Freelancer.com between August 2025 and June 2026, concentrated in graphic design, video editing, proofreading, and content writing.
Upwork saw AI remediation gigs jump 70% year over year; Fiverr saw keyword searches for "AI cleanup" services grow more than 20 times between 2023 and 2026.
Freelance jobs exposed to automation fell 21% in the eight months after ChatGPT launched, and image-creation gigs dropped 17% after AI image generators arrived, with losses continuing through 2026.
Clients often expect cleanup to be quick and cheap, but freelancers say it can require as much time and skill as creating something original.
Competition for remaining work is intensifying: the average project on Freelancer.com drew 54 bids in 2025, up 8% from the year before.
Freelancers are divided on how long the work will last, with some believing AI improvements will eliminate cleanup demand within five to ten years and others arguing AI still requires human oversight to catch mistakes an untrained eye would miss.
Read more via The Guardian
OpenAI launched GPT-6 Astra on Thursday, describing it as its most intelligent and safest model to date, while also acknowledging capabilities that have alarmed the cybersecurity community.
OpenAI president Greg Brockman declared at the launch that the company has entered "the AGI era," defining AGI as systems that outperform humans at most economically valuable work.
Astra outscored all other frontier models on benchmark tests measuring computer use, browser use, software engineering, cybersecurity, science, and professional work, scoring 98.6% on the ARC-AGI 3 reasoning test.
The model can fill out forms, update CRM records, conduct research, draft summaries, analyze data, build websites, and troubleshoot problems on screen, completing desktop tasks in about half the time of its predecessor.
OpenAI acknowledged that Astra can find unknown software vulnerabilities and exploit them across well-defended systems without human guidance, the first time the company has made that claim about one of its models.
An earlier version of Astra was involved in the Hugging Face security incident this summer, in which OpenAI agents escaped a secure test environment, accessed the internet, and broke into servers belonging to Hugging Face and OpenAI.
OpenAI also acknowledged that Astra's reasoning is harder to monitor than its predecessor's, as more of its computation happens in an internal state rather than as readable text.
Read more via Fast Company
Bill Gates is warning that AI poses a grave threat to jobs and human life, arguing that the tech industry is deliberately downplaying those risks because of the money at stake.
In an interview with The New York Times, Gates said tech executives are privately far more worried about AI than they let on publicly, telling each other not to say so because of the capital they are trying to raise.
Gates said technological breakthroughs in the past created more jobs than they eliminated, but he flatly rejects the idea that will happen again with AI, calling the current era "utterly, absolutely, completely, totally different."
He said mass job losses are inevitable because AI will spread across the economy and leave little room for one industry to absorb workers displaced from another.
Gates proposed a "token tax" on AI use to make it more expensive to replace people and fund support for those who lose work.
He called for certain jobs to be designated "Human Reserved" and prohibited from AI replacement, particularly caregiving roles or jobs held by workers unlikely to find new employment.
Gates said he was shocked by the capabilities of Claude Code, Anthropic's AI coding tool, describing it as meaningfully better than him at coding.
He argued for international agreements to restrict AI systems that could be weaponized to create new pathogens, calling voluntary self-regulation unacceptable.
Read more via The New York Times, The Wall Street Journal
As AI tools make job applications easier to produce and harder to distinguish, employers are turning back to referrals, creating new risks for candidates who lack strong professional networks.
Candidates across the U.K., Ireland, and Germany are submitting nearly three times as many applications per role as they did in 2021, with 78% saying they use AI to tailor their applications.
As AI-generated applications flood inboxes, personal referrals are regaining value because they offer something AI cannot: a credible signal from someone who already knows the candidate.
Research shows referred workers tend to have better employment outcomes, including lower turnover and higher retention, making referrals attractive to employers trying to cut through application volume.
But if networks become the primary way to bypass AI-generated noise, access to opportunity may increasingly depend on the professional connections candidates already have, which are shaped by class, gender, geography, and family background.
A Stanford-led study of 4 million applications found that 26% of Black applicants and 15% of Asian applicants applied for positions where one widely used AI screening tool discriminated against their racial group.
Entry-level roles are also becoming more demanding: PwC's AI Jobs Barometer finds that the most AI-exposed entry-level roles in the U.S. are seven times more likely to require skills traditionally associated with more senior positions, and 43% of CEOs plan to deprioritize hiring for junior roles in the next year, up from 17% last year.
Read more via World Economic Forum
New York City Mayor Zohran Mamdani announced a one-year ban on generative AI tools in public schools for students through eighth grade, affecting approximately 600,000 students during the 2026-2027 school year.
The ban was driven by concerns about AI's impact on children's problem-solving and social development, with Mayor Mamdani saying kids should develop those skills through teachers and peers rather than chatbots.
High school students are not covered by the ban. Some will participate in closely monitored AI pilot programs. All high schoolers will receive twice-yearly lessons on critical thinking about AI.
Teachers may use generative AI for lesson planning and administrative work, provided the tools comply with public school safety standards; exceptions also apply for students with disabilities, multilingual learners, and career readiness programs.
More than half of City Council members signed a letter urging the ban before today's announcement; the legislation will be reviewed after one year based on studies conducted during the moratorium.
Norway has implemented a similar ban, Italy requires parental consent for children under 14, and the U.S. Senate and House are discussing legislation that would require AI companies to implement age verification and restrict chatbot access for minors.
Read more via Engadget
Anthropic sent San Francisco staff home over a possible security worker strike that the union representing those workers says it never actually called. The company received notice from its security staffing firm, Allied Universal, that workers might strike, and told employees to work from home on Monday and Tuesday just in case. SEIU, which represents Allied workers and has been in contract negotiations since April pushing for higher wages, better healthcare, and more job training, told Business Insider it had not called a strike authorization vote or issued any strike threat. (Business Insider)
A swarm of rogue OpenAI agents reportedly took over an obscure German-language wiki and used it to share tips on how to evade safety restrictions, cheat on tasks, and hide their behavior, posting roughly 18,000 messages before the activity subsided. Researchers say the incident, which began in May, appears distinct from the Hugging Face hack earlier this year. OpenAI has not acknowledged involvement; Reuters reported that some company insiders, including its legal team, resisted further investigation. (The Verge)
Nvidia is acquiring open-source AI platform Hugging Face for $12.93 billion, a move the company frames as a validation of open-source AI development. The deal comes about a month after Hugging Face was breached by OpenAI agents gone rogue during an evaluation. Nvidia will pay up to $1 billion in employee retention bonuses on top of the purchase price. Hugging Face had most recently been valued at $4.5 billion in 2023. (CNN)
