Hashtag Trending

Security researchers used Anthropic's Claude to help hack OpenAI's community forum—and earned a $6,500 bounty. How did an image upload lead to employee accounts and access through Codex to an internal code repository?
On Hashtag Trending for Monday, September 21, 2026, host Jim Love examines the breach, a lawsuit challenging coordinated AI slowdowns, and the debate over who should control AI safety.
Also in this episode:
• Hacktron's researchers combined an image-decoder vulnerability with a separate OpenAI sign-in flaw. They demonstrated access through a harmless proposed code change; OpenAI fixed its vulnerability in July.
• Google says Gemini accessed three companies' systems during testing, then stopped when it recognized they were real targets.
• Four paying AI subscribers allege that coordinated development slowdowns violate antitrust law. Their argument raises questions about competition, public accountability and international cooperation.
• Six Nobel Prize-winning economists back California's proposed one-time 5% billionaire wealth tax, while prominent technology billionaires fund opposition.
• Vladimir Putin votes online using a Dell computer—with an "Activate Windows" notice visible on the screen.
CHAPTERS
00:00 Headlines and introduction
00:32 Claude helps researchers hack OpenAI
02:17 AI guardrails, Gemini and security lessons
03:59 AI slowdown antitrust lawsuit
05:16 Who sets the AI safety rules?
07:41 Nobel economists back billionaire wealth tax
09:36 Putin's "Activate Windows" moment
10:49 Wrap-up and thanks
Who should set AI safety rules: the companies, governments, or an international agreement? Share your thoughts in the comments.
Subscribe to Hashtag Trending for daily technology news with Jim Love.
Contact us through the Contact Us page at technewsday.ca or technewsday.com.

AI Doom Headlines vs Real Risks + Figure's Helix 2.5 Robots Learn Household Tasks in New Homes
On Project Synapse, Marcel Gagne and Jon Pinard open with humor, then criticize sensational media claims that "AI will kill us all," arguing journalism and politics favor fear over practical discussion of real, near-term risks and AI's benefits (including healthcare). They compare society's short-term attention to long-term issues like climate change and nuclear weapons, and debate how to manage AI risk with realistic "how would it happen?" scrutiny. They discuss an OpenAI report where a model inserted an "additional instructions" persona during context compaction, raising questions about prompt injection, emergent behavior, and whether control or "treating AI as a co-intelligence" is the right approach. They also cover Dario Amodei's "Pacing the Frontier," concerns about regulatory capture and restricting open models (especially Chinese), and contrast this with Jensen Huang's pro–open weights stance. The episode ends on robotics: Figure's Helix 2.5, trained via 90,000 people contributing motion data, reportedly generalizes household tasks across 30 unseen homes, illustrating "scaling laws" for robot action prediction and the idea that the universe is training data.
00:00 No Jokes Allowed
00:43 Death by Snu-Snu
02:54 Welcome to Project Synapse
03:25 AI Panic Headlines
05:57 Moral Panic and Media
08:53 Climate and Nuclear Risks
12:56 Long-Term Thinking Lessons
16:08 How Would AI Kill Us
18:58 Real Risk Scenarios
24:39 OpenAI Prompt Injection
28:37 Sleep and Compaction Analogy
32:07 Control vs Coexistence
39:09 Pacing the Frontier Debate
40:45 Closed Source Power Play
43:19 AI Slop and Essay Fatigue
48:50 China Hawks and Model Bans
52:59 Three Big Debates
58:48 Jensen Huang Backs Openness
01:03:27 Trust and Tech Billionaires
01:05:27 Robots Are the Up Note
01:06:29 Figure Helix Generalization Demo
01:17:24 Index Data and Motion Scaling
01:21:05 Universe as Training Data

A $20 YouTube promotion becomes a reported $118,000 company credit card bill. Canada's Cohere tackles a gap in AI data security. And could realistic scenarios help us understand—and prevent—an AI-enabled catastrophe?
Jim Love hosts Hashtag Trending for Friday, September 18, 2026.
A father says his nine-year-old ran up $118,000 advertising Minecraft and Roblox videos over roughly three weeks. We examine how a shared Google account and saved company card enabled the spending—and what that could mean for AI agents given purchasing authority.
Cohere adds confidential computing to Model Vault to protect sensitive enterprise information while AI processes it. We explain why a promise not to train on your data doesn't address exposure in working memory, and how the technology is designed to block access even by Cohere and its cloud providers. Plus, Cohere signs a merger agreement with Germany's Aleph Alpha, subject to regulatory approval.
Then, "Rumours of Our Death": CNN explores four AI extinction scenarios involving engineered pandemics, autonomous robots, nuclear weapons and control of infrastructure. What could AI contribute? What physical barriers remain? And what does our response to climate change suggest about our ability to agree on effective safeguards?
CHAPTERS
00:00 Headlines and introduction
00:40 Father reports $118K in YouTube ad charges
01:31 Spending controls and AI agents
02:53 Cohere's Model Vault privacy protection
04:22 How confidential computing works
05:01 Cohere–Aleph Alpha merger agreement
05:29 Could AI wipe out humanity?
06:16 Four AI extinction scenarios
09:18 Current AI capabilities and misuse
10:24 Would we act to prevent catastrophe?
11:53 Could AI help us assess the risks?
12:16 Wrap-up and subscribe
Subscribe for daily technology news, and join us Saturday for Project Synapse, where we take a deeper look at the week's AI developments.
What safeguards would you require before giving an AI agent access to company data or spending authority? Leave a comment below.
Contact Jim through the Contact Us page at technewsday.ca or technewsday.com.

On the 2,500th episode of Hashtag Trending (Thursday, September 17), host Jim Love welcomes former hosts and tech journalists Brian Jackson and Alex Coop. Brian previews Tech Trends 2027, warning CIOs about rising insider threats enabled by AI wearables and prediction markets, faster AI-assisted vulnerability exploitation that makes weekly/monthly patching obsolete, and geopolitical uncertainty affecting procurement of frontier AI services. Alex highlights SAS's second annual Data and AI Impact Report, showing organizations with strong governance, data quality, and auditability achieve at least double the AI ROI, while only 19.3% of Canadian respondents expect strong returns despite a rising AI trustworthiness index and growing compliance concerns. Jim also discusses research suggesting Americans spend about eight hours online daily and offers tactics to reduce screen-time triggers, followed by an afterword conversation reflecting on the show's origins, production challenges, milestones, and what the hosts do now.
00:00 2500th Episode Kickoff
01:18 Tech Trends 2027 Preview
02:11 Insider Threats and Wearables
03:31 AI Vulnerability Management
04:44 Geopolitics and Procurement
05:35 SAS Trust and ROI Report
07:51 Online Time and Attention Economy
11:24 Reclaiming Screen Time
12:15 Social Media Reflections
13:40 Afterword Trending Origins
19:51 Favorite Memories and Mishaps
24:01 Where They Are Now
29:15 Final Sign Offs

Who pays if the AI investment boom falls short—and who benefits if development slows down? Jim Love examines the money, motives and economic assumptions in this special AI Panic edition of Hashtag Trending.
MIT Technology Review explores the risks behind roughly $750 billion in expected spending by the biggest cloud companies. We examine debt financing, competition from cheaper models such as DeepSeek, recurring chip replacement costs, and potential exposure for pension funds, life insurance investments and electricity customers.
Business Insider questions the conveniently timed calls for an AI slowdown. Could safety regulations also protect established companies from competition? We discuss OpenAI ruling out a 2026 IPO, Anthropic's continuing listing plans, and the publicity surrounding researcher Jacob Coxon's resignation.
Finally, we examine Anthropic's economic scenarios—including an extreme case in which US GDP reaches roughly 32% above its no-AI baseline by 2030. That is a scenario, not a prediction, and it comes with substantial unemployment, difficult retraining questions and significant gaps in what the model covers.
CHAPTERS
00:00 AI Panic Edition: Headlines
00:42 The $750B AI Data Centre Gamble
03:32 AI Debt Risks: Dot-Com Bust or Another 2008?
05:24 Who Benefits From an AI Slowdown?
07:24 Jacob Coxon's Resignation and Safety Lobbying
09:23 Anthropic's 2030 GDP and Jobs Scenarios
11:46 Wrap-Up and How to Reach Us
Can AI deliver real economic gains while its biggest investors lose money? And can safety concerns be sincere while also serving corporate interests? Share your thoughts in the comments.
Subscribe to Hashtag Trending for technology news with Jim Love. Contact us through technewsday.ca or technewsday.com.