A court lifted the block on Perplexity's Amazon shopping agent, OpenAI settled a U.S. hiring case, and Texas paused data-center hookups.
 |
Wednesday, August 5, 2026 |
AI collided with the law and public trust today. A federal appeals court lifted the block on Perplexity's Amazon shopping agent, OpenAI settled a U.S. hiring-discrimination case, Texas paused new data-center connections, medical researchers found a trust trap, and customers demanded human backup. |
|
Top 5
What matters today
|
| 01 |
The Ninth Circuit found Amazon was unlikely to prove a federal computer-hacking violation because customers used the agent to access the site. The first appellate ruling on platform access by AI agents could shape shopping, booking and browsing tools across the web.
|
|
| 02 |
The Justice Department said OpenAI and Statsig discouraged domestic applicants for fewer than 10 visa-sponsored roles through hidden listings, mailed applications and late-night radio ads. The deal includes $1.2 million in penalties, a $2 million back-pay fund and required hiring changes; OpenAI denied wrongdoing.
|
|
| 03 |
ERCOT has more than 474 gigawatts of connection requests, over five times its record peak demand, and about 90% come from data centers. The review makes power, water, public subsidies and local impacts gatekeeping questions for the state's AI buildout.
|
|
| 04 |
In skin-disease tests, nonexperts trusted LLM explanations even when the prediction was wrong, while clinicians caught errors. Consumer-facing medical AI needs interfaces that encourage an independent judgment before a confident explanation becomes an anchor.
|
|
| 05 |
Gartner's survey of 3,566 customers found that 50% said generative AI made service easier, while 87% considered access to a person essential. Companies that bury escalation risk turning automation into another barrier.
|
|
|
|
|
AI Workflows
Make AI remember the maintenance your brain refuses to
Put your car, HVAC system and major appliances into one project, then turn manuals and receipts into reminders before the expensive surprises arrive.
|
|
Predictive maintenance gets pitched with factory sensors and six-figure software. Most households have a low-cost version sitting in a junk drawer: owner's manuals, warranty cards, oil-change stickers, repair invoices and phone alerts nobody reads. AI can turn that mess into an upkeep plan for the things that punish your budget when ignored.
Give every expensive thing a file
Create a ChatGPT Project called Home + Car Maintenance. Projects are available on free and paid ChatGPT accounts and can hold PDFs, spreadsheets, documents, images and pasted text. Start with your vehicle, heating and cooling system, water heater, refrigerator or freezer, and washer and dryer. Upload the owner's manual, warranty, latest service invoice and a photo of the model plate for each one. Add the vehicle's current mileage and the date of its last service. Crop addresses, account numbers and payment information first. Then give the project this assignment:
First, ask me for any missing model numbers, purchase or installation dates, current mileage and last-service dates. Then build a 12-month maintenance plan using only the manufacturer documents and service records I provide. For every task, list the equipment, date or mileage trigger, official interval, source file and page, warning signs that may justify earlier service, and whether the job is suitable for an owner or a qualified professional. Flag warranty requirements and safety-related work. Mark anything you cannot verify as "needs confirmation." Do not invent an interval.
Check the cited page before putting a date on your calendar. Manufacturer guidance wins when the chatbot's answer and the manual disagree.
Make the plan tap you on the shoulder
Move the verified dates into the reminder system you already check. ChatGPT Scheduled Tasks can send one-time or recurring notifications on paid plans. A simple task would read: "On the first Saturday of every month at 9 a.m., remind me to open my Home + Car Maintenance project and review what is due in the next 30 days." Scheduled Tasks cannot read files stored in a project, so use the notification as the nudge to open the plan. Free-account users can place the same recurring review in Google Calendar or the reminder app already on their phone. Use the warning systems you may already own
CARFAX Car Care is free and tracks reported service history, upcoming service and open recalls. Eligible Nest thermostats can warn when a forced-air system runs longer than usual, loses heating or cooling, or needs a filter. On compatible appliances, Samsung SmartThings Home Care shows error codes and troubleshooting steps, while LG ThinQ can run Smart Diagnosis and help arrange service. Start with the manufacturer app that matches equipment you already own.
Turn an alert into a better service call
When a dashboard light, thermostat warning or appliance code appears, save a screenshot and add it to the project with the relevant manual. Use this follow-up:
Explain this alert in plain English using the attached manual. Tell me what I can safely check now, whether the manual says to stop using the equipment, what information a technician will need, and what questions I should ask before approving a repair. Quote and cite every safety instruction. Separate facts from possible causes.
Add the final invoice, mileage and repair date after the work is finished. The project becomes a running history that can update the next due date, preserve warranty records and help you compare a future estimate with earlier repairs. The first win may be a filter changed on time, a recall handled before a road trip, or a service call booked before the emergency weekend rate kicks in.
|
|
|
AI Tool Spotlight
Hey Noah follows scheduling conversations through to the invite
|
|
Hey Noah is an AI executive assistant that works through text, email, WhatsApp, iMessage, Slack and phone. It can check calendar availability, coordinate with attendees, handle time zones, reschedule meetings, send reminders and place the confirmed event on the calendar.
Its practical value is follow-through: the assistant joins the back-and-forth with other people and closes the scheduling loop. That requires access to calendars, messages and contacts, so review its permissions, security information and privacy terms before connecting work accounts.
|
|
|
More worth knowing
Headlines
|
|
Work, service and skills
|
|
Gartner's survey of 3,566 customers found that 50% said generative AI made service easier, while 87% considered access to a person essential. Clear escalation should be part of the service design from the start.
|
|
Only 13% of CHROs strongly agreed their companies' job designs were ready, compared with 28% of executives overall. The gap suggests technical rollouts are getting ahead of job design, learning, compensation and career-path work.
|
|
The company reports more than 18,500 live-session attendances and confidence gains of up to 72%. Thirty days later, as many as 90% of participants reported saving 45 to 60 minutes a week, offering a measurable model for role-based training.
|
|
A cited Upwork study found that 77% of employees said AI increased their workload, and many high-productivity users also reported burnout. Faster output still carries review work, decision fatigue and pressure to produce more.
|
|
The Justice Department said OpenAI and Statsig discouraged domestic applicants for fewer than 10 visa-sponsored roles. The settlement includes $1.2 million in penalties, a $2 million back-pay fund, revised policies, staff training and federal monitoring; OpenAI denied wrongdoing.
|
|
|
|
Health, schools and public service
|
|
Nonexperts became more confident in wrong skin-disease judgments when an LLM supplied a convincing explanation. Clinicians were more resilient to bad advice and performed best with a prediction alone.
|
|
Roughly half were built for emergency work or had documented use there. Pricing was usually opaque, privacy risk was nearly universal and verifiable governance assurance was rare, leaving local agencies with a difficult buying problem.
|
|
A qualitative study of 34 Queensland patients found that acceptance depended on clear responsibility, strong data rules, clinician control and tools that make care more personal and responsive. Legal approval alone did not create public trust.
|
|
A Turnitin report found that teachers and administrators led implementation in 48% of respondents' schools, compared with 17% for technology leaders. Teacher leadership can move quickly, though districtwide coordination still matters.
|
|
The discussion follows the firing of a Nashville teacher accused of altering student images in ways that made students feel unsafe. Possible changes include tougher penalties for failing to report abuse and greater transparency around misconduct dismissals.
|
|
|
|
Rules, trust and security
|
|
Projects must disclose expected grid demand, on-site generation, water use, cooling systems, public incentives and local impacts before moving forward. The order gives state agencies a much fuller picture of an unusually large development queue.
|
|
The Ninth Circuit found Amazon was unlikely to prove a computer-hacking violation because customers used the agent to access the site. The ruling is the first federal appellate decision on platform access by user-directed AI agents.
|
|
The law requires covered tools to preserve origin and modification information and offer free verification. Social platforms face their next compliance deadline on January 1.
|
|
The coalition, which includes Arkansas, demanded records, logs and communications tied to the breach and similar tests. No lawsuit has been filed, though the letter signals possible consumer-protection and data-privacy action.
|
|
FBI figures cited by CNN show investment scams accounted for more than $632 million. Known-contact verification, family safe words, multifactor authentication and passkeys remain useful defenses as deepfakes and voice cloning improve.
|
|
|
|
Business and infrastructure
|
|
The company reported $7.8 billion in revenue and $15.83 billion in AI capital spending for the quarter. Shares fell after hours as investors weighed the growth against more than $18 billion in overall capital spending.
|
|
Second-quarter revenue rose 50% to $11.54 billion, and AMD forecast third-quarter sales above expectations. Shares still fell more than 7% after hours, showing how much growth investors already expect from the AI buildout.
|
|
Quarterly revenue rose 24% to $20.54 billion, and shares gained 12% after the results. Construction equipment and backup power are turning the AI buildout into revenue well beyond chip and software companies.
|
|
Oracle has $129.5 billion in debt and roughly $260 billion in data-center lease commitments, while S&P cut it to BBB-, one notch above junk. The pressure shows the financing risk behind the AI infrastructure boom.
|
|
The FCC measure is still being developed and could change. A ban would address security concerns inside the AI supply chain while potentially raising costs for U.S. cloud providers that need alternatives at scale.
|
|
|
|
Products, media and culture
|
|
Anyone can now download the app and hail a fully autonomous ride. Nearly 150,000 people used the earlier interest-list service, and Waymo is testing airport and freeway routes next.
|
|
The coming paid add-on will credit and compensate participating artists and direct listeners back to the original work. Consent is becoming a product requirement for consumer-facing AI music features.
|
|
The study covered nearly 147,000 agent actions at 40 companies. Human approval of proposed actions rose from 23% to 41% over three months, while higher-risk identity and access work still required close supervision.
|
|
New products cover vulnerability remediation, model observability, clean backup recovery, identity exposure and autonomous investigations. These launches put governed actions inside existing security systems.
|
|
Muck Rack data cited in the analysis found that 29% of organizations had no owner for AI visibility and 39% did not measure it. Communications teams need a process for auditing answers and correcting weak source material.
|
|
|
|
mAIn Street is built for nontechnical readers who want the signal, not the sludge.
|
|
|
|