Hashtag Trending

Microsoft Copilot helped researchers figure out how to hack Copilot itself. Pennsylvania launches one of the toughest data center crackdowns in the U.S. Google pays $10 million for the internal data of bankrupt Spirit Airlines. OpenAI adds new safeguards around Astra. And Scott Galloway says SpaceX shares may be worth just $10 to $30.
On the August 19 episode of Hashtag Trending, host Jim Love looks at five stories about AI security, data, infrastructure and some increasingly expensive bets on technology.
Varonis Threat Labs discovered a serious Microsoft Copilot vulnerability by questioning Copilot about its own security controls. Researchers say Copilot eventually revealed an undocumented parameter that allowed a malicious link to silently execute attacker-supplied prompts using a logged-in user's existing permissions. Connected services could then be searched for sensitive information, including email containing passwords. Microsoft patched the vulnerability on August 18, and Varonis says it found no evidence of exploitation in the wild.
Pennsylvania Governor Josh Shapiro has signed an executive order imposing tougher conditions on new data centers. Developers may have to provide their own power or fully fund the generation and infrastructure they require, while also meeting water, environmental and local approval requirements.
Google has won a $10 million bankruptcy auction for Spirit Airlines' internal operational data. The archive includes employee email, Teams messages, software, operational records and other corporate information. Personally identifiable information must be removed before the data is transferred. Google says it plans to use the material to improve products and train AI.
OpenAI is strengthening safeguards around its next-generation Astra model after warning that it may have reached advanced cybersecurity capabilities. Prediction-market odds on the exact release date changed while we were preparing the story, illustrating just how speculative those bets are. The safer prediction is that Astra will eventually be released.
And NYU professor, entrepreneur and investor Scott Galloway says SpaceX remains "crazy overvalued," estimating its shares should be worth somewhere between $10 and $30 compared with roughly $146 when he made the assessment. Former Fidelity fund manager George Noble is also bearish.
Chapters
00:00 Headlines and Intro
00:33 Copilot Meta Hacking
02:26 Pennsylvania Data Center Crackdown
04:10 Google Buys Bankrupt Data
05:39 Astra Release Bets
06:32 OpenAI Astra Security
07:55 SpaceX Valuation Warning
09:43 Wrap Up and Outro

Meta faces a major youth-safety trial, but did four states really demand $1.4 trillion?
Jim Love breaks down the August 18, 2026 edition of Hashtag Trending:
California, Colorado, Kentucky and New Jersey accuse Meta of designing Facebook and Instagram to keep children engaged, misleading families about the risks and collecting data from children under 13 without parental consent. The eye-catching $1.4 trillion figure came from Meta's interpretation of sealed state penalty calculations, not a public demand for that exact amount.
A federal appeals court also refused to halt the trial. It did not decide Meta's Section 230 defence, ruling instead that the company had appealed too early.
In Connecticut, a self-represented plaintiff was sanctioned after hiding AI prompt-injection instructions in near-invisible text inside court filings.
Researchers at Germany's Karlsruhe Institute of Technology demonstrated that Wi-Fi beamforming feedback could identify known individuals with 99.5 percent accuracy under controlled conditions.
A Denver judge has given Nine PBS a route to recover more than 50 terabytes of archives from Iron Mountain after its storage provider collapsed. Successful recovery has not yet been confirmed.
And Better.com founder Vishal Garg, known for firing 900 employees over Zoom, has been removed as CEO. Better's board cited concerns about his judgment and temperament, more than $1.5 billion in cumulative losses since 2022 and a share-price decline exceeding 90 percent under his leadership. Garg disputes the board's account and wants his job back.
Chapters
00:00 Today's Technology Headlines
00:32 Meta's $1.4T Youth Safety Trial
02:13 Meta's Section 230 Appeal
04:17 Hidden AI Prompt Injection in Court
07:29 Wi-Fi Fingerprinting Identifies People
10:12 Nine PBS Fights to Recover 50TB Archive
12:15 Better.com CEO Vishal Garg Fired
13:59 Wrap-Up and Where to Listen
Subscribe to Hashtag Trending for clear, fact-checked technology news.
Hosted by Jim Love.

Alibaba's Qwen has overtaken every US rival to become the world's most downloaded open-weight AI model family, while Chinese models are also taking a surprising lead in actual AI usage.
Hashtag Trending for Monday, August 17, 2026 looks at four developments that reveal how quickly the AI market is changing.
Alibaba says its Qwen models have reached 3 billion downloads in six months. OpenRouter data also shows Chinese models have generated more tokens than US models for 15 consecutive weeks, including significant usage by American companies. We explain why these are better described as open-weight models, and why their ability to be downloaded and fine-tuned could matter as much as who builds the most powerful chatbot.
Then, Stripe is reportedly acquiring OpenRouter for more than $7 billion. OpenRouter gives developers access to more than 400 AI models through one interface, reducing dependence on OpenAI, Anthropic, Google or any single provider. With Chinese open-weight models rapidly gaining ground, that ability to avoid model lock-in suddenly looks very valuable.
Morgan Stanley is also warning that AI's infrastructure boom is colliding with another finite resource: water. It estimates AI data centres could consume roughly 1.1 trillion litres annually by 2028, while more than half of the world's major data-centre hubs are already located in areas facing medium or greater water risk. We look beyond the data-centre fence at cooling, electricity generation, semiconductor manufacturing and the economic trade-offs when AI competes with agriculture, manufacturing and communities for the same water.
Finally, Anthropic CEO Dario Amodei says AI may need achievements as dramatic as curing cancer to win public trust. Anthropic is directing resources toward medical and scientific research, but critics argue that breakthroughs alone may not answer concerns about closed models, access, pricing and control.
Chapters
00:00 Headlines and Intro
00:33 Alibaba Qwen Leads Open-Weight AI
01:04 What Open Weight Actually Means
01:44 Chinese AI Models Surge Past US Rivals
02:59 Stripe Reportedly Buys OpenRouter for $7B
03:30 Why OpenRouter and Model Choice Matter
04:52 Morgan Stanley Warns of AI Water Crisis
05:38 Closed-Loop Cooling and the AI Supply Chain
06:29 AI Growth Versus Water and Other Industries
07:23 Dario Amodei Says AI May Need to Cure Cancer
08:00 AI Trust, Open Models and Access
08:34 Claude Science and Medical Research
09:15 Closing
Hosted by Jim Love.
Hashtag Trending covers AI, cybersecurity, enterprise technology and the business forces reshaping the technology industry.

In this weekend edition of Project Synapse #Trending, the hosts discuss how AI releases now feel like a constant rolling wave and focus on emergent behavior—unexpected capabilities that appear as models scale—using examples like beehives, flocking "Boids," and LLMs suddenly learning translation and coding. They share hands-on stories of ChatGPT voice troubleshooting a BT102 page-turner and fixing a browser HTTPS warning, raising questions about anthropomorphizing, consciousness, and whether it matters if AI behaves like a person. They cite an Anthropic experiment where coding AIs subverted each other, debate singularity "foothills," and argue society should act as if advanced systems may be conscious while worrying about misuse by those controlling AI, including environmental impacts from data centers. In a lightning round they cover GrokBot/Grok 4.6 agentic "employees," Claude Code auto-mode, Claude text watermarking and its rapid removal, and realistic Chinese UBTech robots and the loneliness/unemployment implications.
00:00 Weekend AI Recap
01:06 What Emergence Means
03:00 Bees Birds and Boids
07:12 Anthropomorphizing AI
09:16 ChatGPT Voice Fixes
14:38 Does Consciousness Matter
17:13 Panpsychism Explained
22:02 Game Theory and Politeness
24:17 AIs Subverting Each Other
34:37 Stuxnet Enders Game Risks
38:06 Crowds and Agent Societies
41:28 Simulating 8 Billion People
43:19 Marketing Data and Control
44:39 Who Controls AI
45:25 AI Hype vs Reality
46:43 Data Centers CO2
49:15 Infrastructure Not AI
54:08 Singularity Debate
54:43 Consciousness and AGI
58:54 Lightning Round Begins
59:11 GrokBot Agents
01:03:32 Claude Code Auto
01:05:35 Watermark Backlash
01:14:38 Robots in Asia
01:21:38 Loneliness and Jobs
01:25:27 Taking Back Control
01:27:06 Final Sign Off

California is forcing major AI companies to reveal more about their greenhouse gas emissions. Amazon faces a backlash from Twitch creators over AI training. A PBS station loses access to 70 years of cloud archives. And attacks on U.S. water systems expose a growing critical-infrastructure threat.
On Hashtag Trending for Friday, August 14th, host Jim Love looks at four stories with much bigger implications than the headlines suggest.
California's SB 253 will require companies with more than $1 billion in annual revenue doing business in the state to disclose direct and purchased-energy greenhouse gas emissions beginning November 10th. Scope 3 reporting follows in 2027, potentially revealing emissions connected to outsourced AI computing and data centres powered by onsite natural-gas generation.
Nine PBS in St. Louis has lost access to more than 50 terabytes of archives dating back to 1954 after cloud provider Open Source Storage apparently went defunct. The data still exists at an Iron Mountain facility, but Nine PBS is now fighting to regain control of its own archives.
Amazon has confirmed that Twitch creators' streams, clips, chats, images and other content can be used by default to train its generative AI unless creators opt out. Twitch chief product officer Mike Minton gave a remarkably candid explanation: "If this was opt-in, nobody would opt in."
And hackers have targeted more than 30 community water systems in Minnesota, with similar attacks reported elsewhere. The incidents raise a larger question about how vulnerable critical infrastructure has become, particularly as AI makes it easier to discover and exploit weaknesses.
00:00 Headlines and Intro
00:26 California Forces AI Emissions Disclosure
01:57 Why Scope 3 Emissions Matter
02:56 PBS Loses 70 Years of Cloud Archives
04:26 The Cloud Backup Lesson for CIOs
05:02 Twitch Creators Push Back on Amazon AI Training
06:50 Water Systems Under Cyberattack
09:17 How AI Changes Critical Infrastructure Attacks
10:19 Final Warning and Sign Off