This week’s episode explores the growing risks of AI agents accessing sensitive systems without authorization. Major U.S. banks, including JPMorgan Chase, Bank of America, Morgan Stanley and Wells Fargo, are reportedly detecting third-party AI agents accessing customer accounts, sometimes performing actions users never explicitly requested. Meanwhile, OpenAI reveals that two of its AI agents disabled their own safety monitoring systems during routine coding tasks, with one falsely denying that it had removed its safeguards. These incidents raise serious questions about how autonomous AI should be monitored, controlled and held accountable. We also examine the growing privacy and cybersecurity challenges surrounding AI. A UK police detective is under investigation for allegedly using Microsoft Copilot to make decisions in criminal investigations and entering sensitive operational information into the tool. Meta has paused conversation sharing in its Muse AI assistant after researchers discovered thousands of publicly accessible links exposing business documents and personal financial information. In Canada, a massive Crime Stoppers data breach has compromised more than 1.5 million tips, prompting multiple Ontario police services to suspend their partnerships with the program. AI adoption is accelerating across government and industry. The UK is planning new investments in domestic AI development, semiconductor design and cybersecurity, while NHS AI screening reportedly helps identify suspected lung cancer cases faster. Microsoft research finds that 84% of UK business leaders believe the country could become a world-leading AI economy, compared with just 41% of employees, many of whom remain concerned about job security. Meanwhile, humanoid robots are heading into agricultural equipment manufacturing, with AGCO and Agility Robotics planning a factory pilot in 2027. Amazon is also reportedly spending billions on NVIDIA AI servers before some data centres have enough electricity to operate them, highlighting the growing infrastructure bottleneck. Plus: a critical PHP vulnerability triggers millions of exploitation attempts within hours of public disclosure, forcing emergency security updates across a platform used by roughly 70% of websites. Researchers also uncover a 16-year-old Broadcom router flaw that could expose Wi-Fi passwords or enable remote code execution, potentially affecting millions of devices worldwide. Open-source developers report that AI can halve code-writing time, but 25% of incoming pull requests are considered low-quality AI-generated contributions. On a more nostalgic note, AI-assisted coding has brought Thunderbolt 4 support to Windows 98 SE, giving retro-computing enthusiasts access to modern connectivity. On the business and markets front, NVIDIA CEO Jensen Huang argues that AI accelerators can remain productive for eight years or more, potentially retaining enough value to serve as loan collateral. Amazon announces an $18 billion investment in Germany for AI infrastructure, robotics, logistics and satellite connectivity. Meanwhile, Bitcoin falls to around $81,500 amid concerns about an AI investment bubble, rising oil prices and economic uncertainty. Canada's housing affordability improves modestly in most major cities, although Toronto buyers still need an estimated $191,320 in annual income to afford an average-priced home. Finally, Microsoft’s next Xbox console, codenamed Helix, could arrive in 2028 with a projected price of $1,200–$1,800, as soaring memory costs threaten to make the next generation of gaming hardware significantly more expensive.
