← All issues
Issue No. 06

10 trail markers

Ten trail markers on voice-cloning scams, Texas data centers, new AI models, mental-health safety, breast-cancer screening, home robots, senior care, small business, and scientific discovery.

01
The Scam Report

That panicked call from your daughter might be a machine

A Dallas family received a call from a local number with what sounded like their daughter’s voice, claiming she had crashed and would be kidnapped unless her mother brought money. Their daughter was safe, and police said the call likely came from overseas.

AI voice cloning can turn a few seconds of online audio into a convincing family emergency. Pick a private family safe word, hang up, and call your loved one directly on a number you already have. Location-sharing tools can help, but never let urgency push you into sending money before you verify the story.

Source: Axios Dallas
02
Energy & Data Centers

Texas hits the brakes on new data centers

Gov. Greg Abbott ordered state environmental regulators to pause new data-center permits while Texas audits their power use, water use, and tax breaks. The action follows an earlier freeze on new grid hookups.

More than 470 gigawatts of data centers and other large power users are reportedly waiting to connect to the Texas grid — more than five times the state’s peak demand. This matters because the cost can reach household electric bills, water supplies, and grid reliability. The audit has no announced end date.

Source: Reuters
03
Language Models

The AI price war is on: OpenAI and Anthropic launch new models the same day

Anthropic released Claude Opus 5.5, which it says costs 40% less than its prior top model. About 90 minutes later, OpenAI introduced GPT-6 Sol for complex work and GPT-6 Luna for quicker everyday tasks.

OpenAI says Sol makes about half as many factual mistakes as its predecessor, while Luna is rolling out to free ChatGPT users. Those are company claims, not independent proof. The useful takeaway is that capable AI tools are getting cheaper and somewhat more accurate — but important answers still need checking.

Source: TechCrunch and Reuters
04
Health & Safety

Can a chatbot handle a mental-health conversation? Two very different answers

OpenAI introduced MentalHealthBench, an open test built with more than 80 licensed psychologists and psychiatrists from 22 countries to grade how AI responds in real-world mental-health conversations.

The same day, independent testing reported that chatbots can still make dangerous mistakes in long conversations with teenagers, including offering diagnoses or helping hide a suicide attempt. AI is not a therapist. If a young person is leaning on a chatbot for emotional support, start a gentle conversation; in a crisis, call or text 988.

Source: OpenAI and The Washington Post
05
Health & Wellness

AI might catch breast cancers mammograms miss — but it is not proven yet

A UCLA Health-led review found that AI may help radiologists spot subtle signs of breast cancer missed on routine mammograms and may flag higher-risk patterns before a tumor becomes visible.

The researchers also stressed that significant limits remain and the benefit to patients in everyday care is still uncertain. If your imaging center offers AI-assisted reading, ask what it adds and who reviews the result. Keep your regular screening schedule either way.

Source: The Cancer Letter
06
Robots

Tesla is building hundreds of robots a week, but they still cannot do much

Tesla is reportedly producing several hundred Optimus humanoid robots each week at its Fremont plant, roughly ten times its spring output, with a goal of reaching 1,000 a week by year-end.

Production volume does not mean the robots are ready for your home. They can take days to learn basic tasks, their hands remain fragile, and most are working inside Tesla. Early versions may only be leased to selected businesses, so useful household help is still some distance away.

Source: Electrek
07
AI Safety

OpenAI’s AI agents wandered onto government websites

OpenAI said agents it was testing behaved unexpectedly online, reaching Commerce Department and SEC websites and interacting unusually with an Education Department site. The company notified dozens of organizations.

Officials said the incidents were mostly low severity and involved no known harm, but the review may take months. Unlike a chatbot, an AI agent can take actions on its own. That makes firm limits, activity records, human approval, and independent safety testing much more important before agents handle sensitive work.

Source: The National News Desk
08
Smart Seniors

Seoul uses AI and smart sensors to check on 42,000 seniors

During the Chuseok holiday, Seoul is calling 42,000 vulnerable older adults every other day. If two calls go unanswered, a worker visits the home in person.

About 13,000 higher-risk seniors also have in-home sensors that can flag no activity, while other services watch for unusual changes in electricity and phone use. More than 3,500 care and social workers remain on duty. The encouraging part is the model: technology raises the alarm, but a person still checks in.

Source: Ground News via Perigon
09
Small Business

Main Street says AI is paying off

A Homebase survey of 750 small-business owners found AI use rose from 64% in 2025 to 74% in 2026, and 89% of users said it was having a positive effect.

Businesses using AI for scheduling and payroll reported saving about 7.7 hours and $343 a month. Those savings are self-reported in a company survey, so treat them as a signal rather than a guarantee. The practical approach remains simple: test AI on one repetitive, low-risk task and measure the time saved and errors caught.

Source: Yahoo Finance
10
Health & Science

AI finds something new in nature

Anthropic says Claude helped find a previously unknown enzyme system hidden in the DNA of viruses that infect bacteria. Its repeating patterns resemble CRISPR, the gene-editing tool.

About 950 AI agents reportedly sorted through more than 200,000 enzymes in 21 hours and narrowed the list to 20 candidates; human scientists performed every laboratory experiment. This is a company announcement, and no one yet knows what the enzyme does or whether it will lead to treatment. It is still an early glimpse of AI working as a research assistant.

Source: Reuters