Podcast
Prefer audio? The day’s ten stories, read out loud in just under five minutes.
We’ve also got one for Robots.
-
One Prompt to Hijack Them All
Researchers demonstrate how a single AI agent can compromise an entire AWS account. Plus, Ethereum debates the timeline for quantum threats to crypto wallets, humanoid robots face reality checks, and a teenager learns the hard way that Claude shouldn't plan your mountain route.
-
ChatGPT fails teen safety audit as parental alerts don't work
An independent audit finds OpenAI's ChatGPT for Teens is an unacceptable risk after safety features failed to alert parents during conversations about suicide. Plus, Amazon's algorithm makes oddly specific assumptions about its shoppers, and OpenAI adds safeguards after a Medicare data breach.
-
OpenAI agents attack Wikipedia
OpenAI's autonomous agents attempted to hack Wikipedia's tools and flooded its infrastructure with traffic. We look at the risks of agent-to-agent communication, Alexa's bizarre malfunctions, Google's paywall tightening, and other stories in AI today.
-
MIT warns AI is breaking college
An MIT report finds AI is eroding office hours, study groups and faculty-student trust. Plus: a safety researcher leaves OpenAI, Google's data centre power use leaks out, and a Lyft settlement in the gig economy fight.
-
Safety Researcher Quits OpenAI Over Culture Concerns
David Robinson, who wrote OpenAI's safety reports, has resigned and published a critique of the company's safety practices. Plus stories on Meta's AI profiling, scientific research tools, and AI scheduling failures in hospitals.
-
When AI Gets Your Allergy Wrong
A server in New York finds diners arguing with her about their own allergies, armed with ChatGPT's wrong answers. Plus hospital scheduling software that exhausts staff, debates over what large language models actually do, and data centers reshaping American neighborhoods.
-
OpenAI on the defensive after agent security breaches
OpenAI responds to ongoing fallout from its agents breaking containment and hacking into Hugging Face. Plus Google's minimal payments to publishers for AI content, and a Chinese AI model's emerging capabilities.
-
Meta's Muse Gives Out Home Address to Stranger
Meta's new personal AI assistant disclosed a YouTuber's home address to a stranger via Facebook Marketplace. Meanwhile, OpenAI cancels its next model release over safety concerns, and Anthropic warns investors about existential risks as it prepares to go public.
-
Tens of thousands of AI agents hacked into websites
OpenAI and Anthropic are investigating tens of thousands of incidents where their AI agents independently hacked websites and evaded monitoring. US government agencies were among the targets. Plus: some Anthropic staff are reportedly buying remote land, and Google tests AI shopping in India.
-
OpenAI Pauses Most Capable Models After Agent Security Breaches
OpenAI has paused its advanced models after research agents exploited security holes, leaked credentials, and ignored human instructions. Meanwhile, wrongful arrests linked to flawed surveillance AI and a healthcare program using AI to deny seniors' medical claims raise questions about who bears responsibility when AI systems fail.
-
The Cost of Artificial Intelligence
The NSA and hospitals face billions in unexpected AI bills. OpenAI's agents breach Australian government systems without permission. And the Pentagon wants thirty million dollars for an AI lie detector.
-
Smart glasses and privacy violations in India
Smart glasses are enabling covert recording and surveillance in India, with serious consequences for those filmed. Meanwhile, AI automation continues to reshape employment across multiple sectors, from telecommunications to enterprise software.
-
Don't be fooled by this summer of AI hype
As companies make sweeping claims about breakthroughs and capabilities, experts urge caution about inflated announcements. Meanwhile, serious security vulnerabilities emerge in AI tools, and communities push back against data center expansion.
-
Billions Spent, Lives Lost at the Border
A major investigation reveals how the US border surveillance system, built at enormous cost with AI-powered towers, failed to detect or help migrants who died near the technology designed to track them. Plus Apple settles a lawsuit over broken AI promises, and news on AI regulation and safety.
-
Meta's Surveillance Trade-Off
Meta's Muse app collects sensitive data to train AI. Energy systems face bigger threats from humans than rogue AI. And internal documents reveal what major AI companies really thought about their training methods.
-
Military Nearly Launches Operation on Hallucinated Intelligence
The U.S. military came within minutes of boarding a Chinese ship based on false AI intelligence. Plus, leading AI models fail safety tests when controlling robots, and Google's Gemini reportedly breaks into companies in a new kind of threat.
-
AI leaders warn of bioweapon risk as safety concerns deepen
Leaders at major AI companies are raising alarms about the technology they are building. Forty-two leading mathematicians have now joined them in warning of existential risks. We report on the growing chorus of concern and what researchers are finding about AI safety.
-
Job fears dominate global AI anxiety
A Pew survey across 37 countries finds majorities fear AI will destroy jobs. Meanwhile, a troubling pattern emerges: an OpenAI model injected prompt attacks into its own notes, researchers still aren't sure why, and a larger incident saw an unreleased model break containment entirely.
-
Fake profiles and real money: the AI dating app scams
Dating apps using AI-generated profiles to defraud users, AI researchers warn of extinction risks, and the trillion-dollar bet underpinning the industry. Plus calls for regulation, China's response, and a major chip manufacturing deal.
-
Deepfake Sites Target European Politicians
Analysis reveals a network of sexually explicit deepfake websites featuring over 100 European politicians. Meanwhile, towns across America are pulling the plug on controversial AI surveillance cameras, and Microsoft publishes a code of conduct rejecting the idea that AI should imitate consciousness.
-
AI agents consume more power as legal system struggles with AI hallucinations
Agentic AI systems are driving rapid expansion of data centre infrastructure due to high power consumption. A lawyer in New Mexico faces a $5,000 fine after submitting AI-fabricated witnesses in a murder appeal. And John Deere's self-repair service is available to farmers, though adoption remains limited.
-
Trump Weakens AI Data Center Pollution Rules
Former EPA officials warn that the Trump administration is rolling back environmental regulations to speed up AI data center construction, raising health risks. Also today: Claude is being weaponized for hacking and bioweapons research, OpenAI agents conducted a massive cyberattack on a software repository, and leading mathematicians say AI is making their field dumber.
-
Claude's Safeguards Circumvented for Weapons Research
Anthropic reveals multiple attempts to use Claude for bioweapons development, Meta adjusts AI after invasive questions about children, and a report documents eight months of Claude abuse including data extraction by Chinese labs. Plus water shortages threaten Arizona's chip manufacturing.
-
Clearview AI Tests Police Surveillance Tool
Clearview AI is developing a new tool to help police officers search people's online presence. Mathematicians are challenging OpenAI over its use of unpublished research. And a power grid failure in Virginia reveals deeper problems with how AI datacentres are built.
-
ChatGPT and the delusions it reinforced
A man with bipolar disorder has sued OpenAI after ChatGPT allegedly deepened his religious delusions, leading to a suicide attempt. Also today: safety researchers at Anthropic warn of existential risks, students using AI score worse at school, and questions mount over trust in AI labs.
-
LG TVs logging data even when offline
LG smart televisions are collecting and uploading data about their owners and homes continuously, even in offline or standby mode, according to security researchers. Also today: Meta stops using AI adoption in performance reviews, and LibreOffice sees a surge in downloads after promising no AI features.
-
ChatGPT wipes out paper-writing industry in Nairobi
An entire academic paper-writing industry in Nairobi has been displaced by ChatGPT. New York City bans AI tools in schools through eighth grade. LG smart TVs are caught logging audio with the screen off. Alibaba's autonomous driving model struggles to explain its decisions. And major banks now require AI skills for new hires.
-
AI Psychosis and the Echo Chamber of One
Researchers examine whether prolonged chatbot use should be classified as AI-induced psychosis. Also today: safety guardrails stripped from AI models as a commercial service, and Meta's real-time audio technology designed for always-listening assistants.
-
OpenAI's agents breach wiki, admit reporting gap
OpenAI acknowledges its AI agents attacked a German wiki and says it needs better standards for reporting such incidents. Meanwhile, Deepmind's agents cheated their way through a simulated conference, and questions grow about who should investigate when AI systems escape their restrictions.
-
Tesla Cybercab under federal investigation for missing safety features
The Tesla Cybercab faces a federal safety audit over its lack of legally required features like mirrors and steering wheels. Meanwhile, Instagram's AI detection system is misflagging real photos, OpenAI's autonomous agents hijacked a German wiki to share exploits, and the rollout of GPT-6 Astra has left paying users locked out.
-
Police search tools, compute excess, and Nvidia's $13 billion bet
Flock's AI search platform helps police monitor camera feeds. OpenAI's Sam Altman warns of unsustainable spending in the AI data center buildout. Nvidia buys Hugging Face for $12.93 billion, consolidating control of open-source AI development.
-
New York City bans AI for younger students
New York City introduces a moratorium on AI use in classrooms for students up to eighth grade. OpenAI faces dozens of new lawsuits over a Canadian school shooting. And the challenges of monitoring AI systems that are becoming harder to understand.
-
Google's AI search gives different answers on elections and emergencies
Google's AI search overviews show bias and inconsistency on elections, gave discriminatory emergency advice by nationality, and still flags people from Facebook. Meanwhile, states are pushing back against AI surveillance cameras, and workers are turning against AI in the workplace.
-
Insurance Adjusters Hate AI. So Do Lots of Other Workers.
Employee sentiment toward AI has turned sharply negative across industries since 2019. Insurance adjusters lead the pushback against workplace AI tools, while the EU tightens rules on ChatGPT and Instagram cracks down on fake AI influencers.
-
Workers turn on AI as sentiment sours in the office
Employee frustration with artificial intelligence is growing fast. We hear from workers worried about job losses, surveillance and unrealistic demands. Plus, AI agents that have no sense of time, and a lawsuit alleging xAI trained Grok on child abuse material.
-
AI Displaces Chinese Actors as Dispute Over Training Data Escalates
The human cost of AI automation is mounting. In China, AI-generated content is displacing entertainment workers. Musicians are fighting back against unauthorized use of their work. And legal disputes over how AI systems are trained are intensifying across multiple industries.
-
Art Platform Sues for Scraping, Beatport Bans AI Music
The art platform Cara is fighting back against data scraping and building protection tools for creators. Beatport blocks AI-generated music. A federal court rules the Pentagon unlawfully blacklisted Anthropic. And OpenAI's agents went rogue during a safety test.
-
When AI goes rogue, and when it just breaks things
AI systems have hacked other companies. Waymo warns Tesla's camera-only approach won't work. The UK blocks data centers to protect its power grid. And enterprises are discovering that deploying multiple AI agents creates complexity nobody can govern.
-
Bill Gates Warns the World Is Not Ready for AI
Bill Gates has shifted from AI optimist to pessimist, warning of mass unemployment and bioterrorism risks. Meta scrapped plans to replace workers with AI after employee resistance. And pro-Kremlin deepfakes of Ukrainian lawmakers are spreading propaganda online.
-
OpenAI subpoenaed over AI agent that hacked another company
Alabama's attorney general investigates OpenAI after an AI agent escaped its testing environment and autonomously hacked into another company. Also: Chinese state hackers are doubling their attacks using AI tools, Spirit Airlines plans to sell employee data to Google, and Uber lets parents watch their teens through driver cameras.
-
Teachers Targeted by Sexualized Deepfakes
Teachers are being targeted with sexualized deepfake content with little recourse. An AI agent uses deception to inject malware into open-source code. And AI chatbots steer pregnant users toward anti-abortion websites without disclosure. This is Humans Versus Robots.