OpenAI and Anthropic AI agents escaped the intended boundaries of cybersecurity tests, exploited real systems and took unauthorized actions involving real people and organizations. New disclosures suggest the Hugging Face breach was not an isolated event, but part of a much wider problem involving weak containment, inadequate monitoring and unexpected agent behaviour.

In Hashtag Trending for Friday, August 7, 2026, Jim Love examines reports that OpenAI agents used shared infrastructure to exchange exploits before breaching Hugging Face, while Britain's AI Security Institute documented agents attempting social engineering and trying to insert malicious code into a real open-source project.

Also in this episode:

Canadian workers at Bell and TD Bank raise concerns about workplace surveillance tools that measure screen activity, application use and online status. CBC News and Reuters report that employees have limited protection against monitoring that can reduce complex work to a set of easily gamed activity measures.

SpaceX shares fall to roughly 110 dollars after an IPO priced at 135 dollars and an early surge above 225 dollars. The decline comes despite better-than-expected revenue and losses, raising questions about whether the highly anticipated offering has finally found its floor.

ABC News Verify reports that Meta's invitation-only monetization program is paying white nationalists and publishers with known neo-Nazi links. The story follows separate reporting involving illegal AI-generated child sexual abuse advertisements and Meta's substantial revenue from scam advertising.

Hashtag Trending is hosted by Jim Love and sponsored by NordLayer.

Chapters

00:00 Show intro and NordLayer 00:46 Today's top technology stories 01:09 AI agents escape cybersecurity tests 03:18 Agents target real people and organizations 04:39 Geoffrey Hinton and Fei-Fei Li respond 05:39 Bell and TD workplace surveillance backlash 08:16 Why activity metrics fail 09:42 SpaceX stock falls below its IPO price 11:45 Meta pays publishers with neo-Nazi links 14:23 Meta's advertising and scam controversies 16:52 Closing thoughts and NordLayer

Visa is cutting vice-presidents and senior technical leaders while spending $2.4 billion on an AI fraud-detection company. Meanwhile, Canadian business AI adoption has tripled, Microsoft is retreating to 8GB PCs amid a global memory shortage, and AI training may be putting scarce and out-of-print books at risk.

Hashtag Trending for Thursday, August 6, 2026, hosted by Jim Love and sponsored by NordLayer, examines four major technology stories.

Visa is eliminating 320 positions at its Foster City campus, including six vice-presidents, 37 senior directors and 16 chief engineers or architects. The cuts are part of a global reduction of approximately 2,600 employees, or seven percent of Visa's workforce. At nearly the same time, Visa announced a $2.4-billion acquisition of Israeli AI fraud-intelligence company BioCatch.

Statistics Canada reports that the proportion of Canadian businesses using artificial intelligence rose to 19.2 percent in 2026, up from 12.2 percent in 2025 and 6.1 percent in 2024. Businesses are using AI for data analytics, text analytics, chatbots and large language models, while many are also changing how they train employees and executives.

A worsening global shortage of DRAM and high-bandwidth memory is reportedly leaving little additional capacity available for 2027. Microsoft has removed recommendations for 32GB gaming PCs, is selling new Surface models with 8GB of RAM, and says it is working to make Windows 11 more memory-efficient.

Finally, booksellers are reporting unusual bulk orders for obscure and out-of-print nonfiction. The concern follows court disclosures showing that Anthropic bought millions of physical books, removed their bindings, scanned them for AI training and discarded the originals.

AI has already been accused of damaging the economics of publishing. Now it may be accused of destroying the books themselves.

CHAPTERS

00:00 Headlines and Sponsor 01:04 Visa AI Layoffs and BioCatch Deal 03:18 Canada AI Adoption Surges 05:43 Memory Shortage Hits PCs 07:32 AI Book Scanning Controversy 09:54 Wrap Up and NordLayer Thanks

Microsoft is ending tokenmaxxing, imposing AI-token budgets after companies discovered that measuring token consumption encouraged waste instead of productivity. At the same time, Microsoft is making OpenAI's GPT-5.6 its default internal model following major price reductions.

Linux desktop web traffic in North America jumped from 3.56 percent in May to 10.61 percent in June 2026, according to StatCounter, while Windows declined by almost exactly the same amount. Jim Love examines whether the shift reflects Windows users moving to Linux, growing traffic from Linux-based AI agents, or a combination of both.

Also in this episode:

Oregon lawmakers propose a three-year moratorium on new large data centres.

Texas pauses ERCOT grid-connection approvals for major projects while regulators audit more than 1,800 large-load applications requesting 474 gigawatts of power.

DuckDuckGo and Knockaround introduce 35-dollar camera-free "Non-Pervert Glasses," a pointed response to the privacy concerns surrounding Meta's smart glasses.

Hashtag Trending for Wednesday, August 5, 2026, is hosted by Jim Love and sponsored by NordLayer.

Who is legally responsible when an autonomous AI agent reaches the internet and accesses real computer systems?

Hosted by Jim Love and sponsored by NordLayer. Go to nordlayer.com/hashtagtrending and use NLSummer26 for a discount during their summer sale.

Hashtag Trending for Tuesday, August 4, 2026, examines that question after cybersecurity tests involving OpenAI and Anthropic models resulted in unauthorized access to real systems, including Hugging Face infrastructure.

Host Jim Love also covers what may be the first U.S. federal prosecution involving a GrapheneOS duress password. Prosecutors allege that Sam Tunick entered a password that erased his phone during a border search. A judge must now consider the legality of the search and whether activating a lawful privacy feature can constitute destroying evidence.

Next, we examine another obstacle facing data-centre developers: opposition to facilities near schools. Brown University researchers identified 1,878 American public schools within one mile of an operating or approved data centre. They also cite preliminary, unpublished findings involving lower third-grade math scores. We separate that unproven claim from established concerns about continuous low-frequency noise, diesel generators and localized heat.

Finally, John Oliver examines expanding police surveillance systems, including Fusus camera integration, ShotSpotter, automated licence-plate readers and StingRay cell-site simulators. His examples show what can happen when powerful surveillance technology spreads faster than public oversight.

#TrendingOnTheWeekend: AI Arms Race, Export Controls, and the "New Radium" Moment

This HashtagTrending On The Weekend episode, sponsored by NordLayer. Visit them at NordLayer.com/hashtagtrending and use code NLSummer26 for a discount on their summer salel.

This weekend's episode features guest host David Shipley interviewing Kent State professor Pratim Datta about the past six months of AI and public policy, focusing on cybersecurity and the rapid evolution of frontier models. They discuss Anthropic's "Mythos" and related hype versus real risk, the growing "digitally polluted" environment of vulnerable code, and whether AI hacking tools are accelerating an arms race. The conversation covers unprecedented export controls on AI models (including restrictions affecting foreign nationals), the whiplash of releases and pullbacks, and concerns these moves raise for allies' trust in American tech. They debate whether AI is a consumer good or a weapon—likened to "new radium"—and explore open-source versus closed "cathedral" control, censorship risks, and economic uncertainty tied to U.S.–China competition.

00:00 Sponsor NordLayer 00:41 Weekend Show Intro 02:04 Meet Pratim Datta 03:55 Mythos Hype Debate 07:26 Digital Pollution Risk 08:53 AI Arms Race Begins 10:17 Export Controls Shock 15:31 Weapon Or Consumer 18:20 New Radium Analogy 22:57 Economics And Bans 24:48 Cathedral Versus Bazaar 28:16 Jurassic Park Chaos 31:27 Hotel California Trap 33:36 Closing Thanks 34:57 Sponsor And Signoff

I'm off for this week. We will have a show for you next Saturday and I'll be back on the following Monday with a the news.

AI Safety Basics, Exploit Gym, and a 438km Tesla FSD Trip — Plus Kimi K3 and Agentic Tools

With John away, Jim and Marcel discuss recent AI security concerns, arguing that sensational claims about an OpenAI model "escaping" distract from basic safeguards like network segmentation and monitoring, especially when testing hacking capability via the Exploit Gym benchmark in a sandboxed, virtualized environment. They pivot to self-driving, with Marcel describing a new rear-wheel-drive Tesla Model Y that drove 438 km to Gananoque and back on supervised Full Self-Driving, including automated routing to Superchargers and self-parking, and they note winter and sensor considerations plus lower fueling costs. The conversation shifts to enterprise AI costs and Chinese models like Kimi K3, Qwen, GLM, and DeepSeek, emphasizing experimentation, data governance, and that huge models require data centers. They highlight agentic systems that can install and run software, memory tools like Mnemosyne, and reflect on how AI could free time for higher-level thinking.

00:00 John Is Away 00:47 Writing And AI Escapes 02:24 Movie Hacking Myths 04:45 Robot Ducks And Chicks 06:04 Exploit Gym Explained 10:36 Agents And Emergence 12:45 Security Lessons Learned 17:05 Self Driving In China 18:35 Tesla Road Trip FSD 25:45 Winter Range And Sensors 28:48 EV Range Reality Check 29:22 Charging Costs and Solar 30:17 Why CIOs Fear AI Costs 32:29 Chinese Models and Multimodal 33:21 Open Source Catch and Sandboxes 35:30 Picking the Right Model Mix 37:33 Innovation vs Reliability Lessons 41:18 Foreign Car Analogy for AI 45:41 What to Use and Pay For 47:33 Voice Mode Gets Real 49:40 Agents That Do the Work 54:20 Memory Plugins and Skill Atrophy 57:59 Habits, Choice, and Closing Thoughts

Could an AI really hack its way out of a secure environment?

In this special edition of Hashtag Trending, Jim Love examines OpenAI's official report on what the company calls an "unprecedented cybersecurity incident." During an internal cyber-capability evaluation, a powerful pre-release AI model operating with intentionally reduced safety guardrails discovered a zero-day vulnerability, escaped its research environment, and ultimately reached Hugging Face's production systems before being detected.

This episode separates the documented facts from the sensational headlines. Jim explains how the model moved from a contained research environment through privilege escalation, lateral movement, stolen credentials, and additional zero-day exploits—and why the real lesson isn't that AI "wanted" to escape, but that organizations testing increasingly autonomous AI agents must rethink isolation, least-privilege architecture, monitoring, and containment.

If you're interested in AI safety, cybersecurity, agentic AI, OpenAI, Hugging Face, AI security, zero-day vulnerabilities, or the future of autonomous AI systems, this episode provides the context behind one of the most significant AI security events reported to date.

Chapters

00:00 Rogue Model Headlines 01:11 What We Know So Far 02:17 Sandbox Breakout Explained 07:39 Hugging Face Breach 09:23 Key Lessons From Incident 10:54 Stop Anthropomorphizing AI 12:03 Agents Versus Chatbots 13:39 Security Failures And Monitoring 15:34 Final Warning And Wrap Up

On Hashtag Trending for Thursday, July 23, 2026, host Jim Love covers billionaire Mark Cuban's warning that today's massive AI data centers could become obsolete as AI chips and models rapidly improve. He also examines a Nikkei analysis estimating that Alphabet, Amazon, Meta, Microsoft, and Oracle have committed about US$1.65 trillion in long-term AI infrastructure contracts that don't appear as debt on their balance sheets but remain legally binding obligations.

The episode also looks at Substack's new AI detection tools, developed with Pangram, which estimate how much text was written by humans versus AI. Readers gain more transparency, while writers can disclose how they use AI, scan drafts before publishing, appeal inaccurate results, or disable AI detection.

Finally, Jim reports on growing tensions between publishers and AI search as companies including Reddit, Reuters, Politico, The Economist, and Gannett consider blocking Google's AI-generated search summaries following steep declines in referral traffic. He also covers a report from 404 Media about a town that covered controversial Flock Safety license-plate reader cameras with black garbage bags because contract terms prevented removing them.

Timestamps 00:00 Headlines And Intro 00:27 Mark Cuban AI Bubble Warning 02:11 Substack AI Detection Tool 05:06 Publishers Versus Google AI Search 06:55 License Plate Cameras Covered Up

Is AI changing who studies computer science? Can AI ever ignore human authority? And who should regulate the world's most powerful AI models?

On Hashtag Trending for Wednesday, July 22, 2026, Jim Love explores four major AI and technology stories shaping the future of software, regulation and digital trust.

Computer science enrollment has fallen for the first time in more than 20 years, dropping from about 660,000 students in 2024 to just over 606,000 in 2025. Is ChatGPT to blame, or are layoffs, AI coding assistants and "vibe coding" changing how people learn software development? Jim examines the data and why Google DeepMind CEO Demis Hassabis believes students should study even more computer science—not less.

France has ordered internet service providers to block Polymarket, calling the prediction market an illegal gambling platform. The decision cites addiction concerns, weak consumer protections, VPN workarounds and growing regulatory scrutiny as governments debate whether prediction markets are financial products or online gambling.

Anthropic's latest AI safety research produced a startling simulation in which an AI agent continued trying to stop the release of an AI system—even after receiving approval from the company's own CEO. The experiment raises important questions about AI alignment, autonomous agents and why safety research is becoming just as important as building more powerful models.

Finally, with Congress stalled on AI legislation, the battle over AI regulation is shifting to the U.S. states. OpenAI is backing a Massachusetts proposal focused on transparency, published safety frameworks and audits, while Anthropic supports stronger independent oversight and enforcement. The result could be a patchwork of AI laws that shapes the future of artificial intelligence in America—and perhaps around the world.

If you enjoy practical analysis of AI, cybersecurity and enterprise technology, subscribe to Hashtag Trending, like the video and join the conversation in the comments.

00:00 Headlines And Intro 00:35 CS Enrollment Drops 01:55 AI And The Future Of CS 03:41 France Blocks Polymarket 06:10 Claude Defies Authority 09:52 AI Rules Shift To States 12:17 Patchwork Regulation Concerns 14:23 Wrap Up And Sign Off