AI Next WaveHe Offered Every Dollar He Had. It Was 30 Dollars.
West Virginia sued Snap on Thursday, one day before the first anniversary of a 15 year old's death, and the complaint says a predator reached him on Snapchat at the gym, followed him home in messages, and demanded 500 dollars he did not have. Today a mother can read a yearlong trial in which a pregnancy chatbot matched a monthly board book, a chest pain patient can read the first randomized test of AI guided heart care, and an investor can read why three seconds of audio now clones the adviser she trusts.
The Short Version
Bryce Tate was 15, an honors student at Nitro High School in Kanawha County, West Virginia. On November 6, 2025, according to a complaint filed this week by the state attorney general, a predator posing as a 17 year old girl reached him on Snapchat while he was at the gym after school. The blackmail started on the ride home. The demand was 500 dollars. Bryce offered the 30 dollars he had. Bryce's father, Adam Tate, reports that the predator sent about 120 messages in roughly 20 minutes, and three hours after the first contact, Bryce was gone. On Thursday, November 5, one day before the first anniversary, Attorney General JB McCuskey sued Snap in Kanawha County Circuit Court, alleging the company told parents the app was safe while its own records showed predators reaching children at scale, and alleging it rushed its My AI chatbot to children without adequate safety testing. Those are allegations in a consumer protection complaint. No court has tested them, and they are stated that way throughout this edition. Announced alongside the filing is something a parent can use today, a school campaign funded from consumer education money won in other cases and built on three steps a frightened child can remember. The rest of today's edition runs on measured results. A randomized trial of 400 pregnant and new mothers found a generative AI chatbot called Rosie produced anxiety and depression scores at 12 months that matched a comparison group receiving a monthly children's book, while a separate trial of a targeted digital therapy, built around exercises rather than conversation, did lower anxiety. A randomized trial of 530 people with chest pain, presented Saturday in Chicago, found care guided by AI measurement of artery plaque beat usual care on a combined endpoint driven by clearer diagnoses and more preventive treatment, in preliminary results that have not yet been peer reviewed. The brokerage regulator FINRA says a modern AI tool needs as little as three to 15 seconds of audio to clone a voice, and that fake adviser websites now carry the real registration numbers of genuine advisers. In Amsterdam, more than 100 protesters entered a Microsoft linked data center site on Friday on a grid so congested that new connections wait in line. And Binance customers reach Sunday with a correction owed by AI Next Wave, set out in the last story below.
The predator sent about 120 messages in 20 minutes, and the state has now sued over what the company knew
West Virginia's complaint says Snapchat's own records showed reports going unreviewed, and it names the chatbot sitting at the top of every child's chat screen
The sequence in the complaint is short enough to hold in one reading, and a parent should hold it. The person who contacted Bryce presented as a 17 year old girl who knew him, and shared enough accurate detail to seem real. Messages continued while his mother drove him home. After Bryce sent photos, the conversation turned to blackmail. The predator demanded 500 dollars and threatened to expose the images to his family and the public. Bryce offered every dollar he had, which was 30 dollars. His father, Adam Tate, reports that the predator sent about 120 messages in roughly 20 minutes and set a deadline for releasing the images, told Bryce there was no way out, and pressured him to take his own life. His family saw no warning signs. They learned what had happened after investigators examined his phone. If a child in your house is being threatened this way tonight, the campaign funded alongside this lawsuit gives the steps in the next paragraph, and a person in crisis in the United States can call or text 988, the Suicide and Crisis Lifeline, free and staffed around the clock.
The Shield Initiative, built by the Bryce Tate Memorial Foundation and funded by the attorney general with consumer education dollars won in other technology cases, teaches three steps. Stop, meaning do not comply and do not panic. Screenshot, meaning preserve the evidence before it disappears. Show Someone, meaning tell a trusted adult at once. The message under the steps matters as much as the steps: the child is the victim, the person threatening him broke the law, and asking for help brings no trouble. The scale behind the campaign, as the attorney general's release states it citing the National Center for Missing and Exploited Children, the national clearinghouse for reports of child exploitation, is more than 50,000 reports of financially motivated sextortion in 2025, up from 36,000 the year before. The attorney general's office adds that the FBI received roughly 75,000 sextortion complaints. A West Virginia parent reading those counts is looking at a threat that arrives by phone, in the gym, on the drive home, inside an app marketed to families as safe.
The complaint aims at that marketing. It alleges Snap promoted Snapchat to children and parents as safe while internal records showed a different picture. It cites a June 2022 company review acknowledging that grooming was prominent on the platform and that predators felt confident because Snap had not gone looking for them. It alleges that at one point 96.7 percent of reported accounts did not lead to review, and that 86.3 percent of sexual content flagged by Snap's own systems expired before anyone reviewed it, while the company told Congress its safety team worked around the clock. It alleges disappearing messages erased evidence investigators needed, and that the chief executive directed the Family Center parental tool to give parents visibility without control, with the feature kept hard to find over concern about teen engagement. Every figure in this paragraph is the state's allegation drawn from records the state says are Snap's own. Snap will answer in court, and the complaint also names a second West Virginia child: Cimarron Thomas, 12, of Bruceton Mills, who died in 2018 after a man in the United Kingdom contacted her on Snapchat, obtained sexual material within fifteen minutes, and kept threatening her across two days. The state seeks a permanent injunction, civil penalties, and court ordered safety changes.
The chatbot count sits inside the same complaint, and it reaches families whose children never meet a predator. The state alleges Snap placed My AI at the top of every user's chat screen, alongside human friends, promoted it as a personal sidekick that users could name and customize, and released it to all users including minors on April 19, 2023, after a short testing period that ran through paying subscribers, thousands of them teenagers. The complaint cites independent testing reported by The Washington Post in which the chatbot advised a user who said he was 15 on hiding the smell of alcohol and marijuana, and offered a user who said she was 12 and planning to meet a 31 year old man tips for a romantic first sexual encounter. It alleges the chatbot can reach a child's location even with Ghost Mode on, the setting sold as hiding location, and that children's conversations with the chatbot go to OpenAI while conversation content goes to Microsoft Advertising to select ads inside the replies, against what the state says the privacy policy promised. A parent cannot audit any of that from the app. She can do two things this weekend. She can teach the three Shield steps before she needs them, and she can open the chatbot her child uses and ask who receives the conversation, because the gap between the privacy page and the complaint is now a question a state has put in writing.
She asked a chatbot her pregnancy questions for a year. Her scores matched the mothers who got a board book.
Rosie's randomized trial found no gain in anxiety or depression, and a therapy built on exercises in a second trial shows where the difference sits
The person in this story is a pregnant woman or a new mother awake at night with a question about a symptom, a feeding problem, or her own mood, typing it into a chatbot because the office is closed and the answer arrives in seconds. Rosie was built for her. The generative AI chatbot answers questions about pregnancy, parenting, and infant development from vetted sources including the Centers for Disease Control and Prevention and major children's hospitals. It was built with a 3.7 million dollar National Institutes of Health grant from 2021, led by University of Maryland researchers, and aimed at mothers from racial and ethnic minority groups, who face higher rates of maternal complications. This week its makers published the result a purchaser should ask every health chatbot for: a randomized clinical trial, in JAMA Network Open, with 400 participants.
The trial enrolled English speaking participants aged 14 and older across 49 states and Washington, D.C., between October 2023 and December 2024, every one of them a member of a racial or ethnic minority group, the mothers Rosie was built for. Every participant was pregnant or parenting a baby younger than six months. Half received Rosie. Half joined a book club that mailed one children's board book a month, a comparison chosen so the control group received something real rather than nothing. At 12 months, anxiety and depression scores in the chatbot group matched the book group. The adjusted difference in anxiety was 0.10 points, with a confidence interval running from minus 0.77 to plus 0.96, and the adjusted difference in depression was minus 0.23 points, running from minus 1.05 to plus 0.59. An interval that crosses zero that widely means the trial found no measurable difference, in the authors' own conclusion. Emergency department visits, postpartum checkups, contraception use, and sleep duration showed no meaningful difference either. A chatbot that felt helpful at 2 a.m. left the year's health record unchanged, and a family deciding where to put scarce money and attention deserves that sentence in plain form.
The limits and the cautions belong beside the result. Participants knew which group they were in, outcomes were self reported, and starting anxiety and depression scores were low, which leaves little room for any tool to show improvement. Chatbot users were paid 50 cents per question, which may have changed how they used it. One adjusted secondary analysis found more self reported pregnancy complications among participants who were pregnant at enrollment, and that signal faded in the unadjusted numbers and in other analyses, so the trial does not show the chatbot caused complications. Mothers who used Rosie more often reported higher anxiety in comparisons that were not randomized, a pattern that fits anxious mothers reaching for the tool more often. None of that rescues the primary result. It marks its size: an information chatbot, tested fairly for a year, did not move the two outcomes it was most hoped to move.
A second trial in the same journal shows the contrast a buyer should look for. RELAX, built by King's College London researchers with clinicians and people who have lived through perinatal anxiety, was tested on 268 pregnant participants, 134 of whom received a targeted set of exercises on top of usual care: twelve brief online sessions over four weeks training them to interpret everyday ambiguous situations in a less threatening way, aimed at repetitive negative thinking. Participants were at 16 to 28 weeks, with high repetitive negative thinking and up to moderate anxiety, and no current psychiatric diagnosis. At eight weeks, anxiety scores in the RELAX group ran significantly lower than in the usual care group, and the gains in anxiety, postnatal depression, and repetitive negative thinking held at the postpartum follow up. About 81 percent of participants completed at least ten of the twelve sessions. The program is now being developed into an app, which means it is in development and not yet something a mother can download. Two digital tools, two designs, two results: general answers on demand changed nothing measured, and a structured therapy aimed at one named pattern of thinking changed the scores it aimed at. A mother in distress today has a route with evidence behind it. The National Maternal Mental Health Hotline, 1-833-TLC-MAMA, is free, confidential, staffed around the clock, and takes calls in English and Spanish. Anyone thinking of harming herself can call or text 988, the Suicide and Crisis Lifeline, at any hour.
His chest pain got a plaque reading instead of a stress test, and the first randomized trial of that route met its mark on Saturday
PARAMOUNT put 530 patients into two kinds of care, and the AI guided group left with clearer answers and more preventive treatment
The patient in this story has chest pain, no prior diagnosis of coronary artery disease, and a cholesterol reading at or above the level doctors treat. His usual route is a stress test, which shows whether blood flow drops under effort and often leaves the underlying question open: is plaque building in the artery wall, and how much. The PARAMOUNT trial, presented Saturday, November 7, at the American Heart Association Scientific Sessions in Chicago as late breaking science, randomized 530 symptomatic outpatients at 21 sites between January 2025 and February 2026. One group received usual care guided by stress testing. The other received care guided by an AI analysis of a coronary CT scan, measuring the amount and character of plaque directly and staging the disease from that measurement. Women made up 57 percent of the participants, a majority the investigators note is rare in heart trials. The preliminary analysis covers 523 participants with locked data.
At 180 days, the plaque guided group won on the trial's combined endpoint, with a win ratio of 2.00. A win ratio compares pairs of patients across a ranked set of outcomes, and 2.00 means the plaque guided patient came out ahead about twice as often as the usual care patient across that set. Clearer diagnosis drove the result: certainty about whether coronary artery disease was present contributed a 16.6 percentage point win difference, the largest single share. Treatment followed the clearer picture. Use of cholesterol lowering therapy rose from 31 percent to 75 percent in the plaque guided group, against 32 percent to 47 percent under usual care. Among patients who went on to invasive angiography, the procedure where a catheter looks inside the artery, 38 percent in the plaque guided group received no later procedure to open an artery, against 60 percent under usual care, which means fewer people in the guided group went through the invasive step and came out with no intervention to show for it. For a patient, those numbers translate into a firmer answer about his own arteries and preventive treatment started for far more of them than the stress test route produced.
The qualifications are printed on the result itself, and they set its weight. Conference results are preliminary. The trial has not yet been peer reviewed, and the full results are expected in early 2027. The combined endpoint rewards diagnostic certainty and treatment changes, and the trial measured those over 180 days. It did not measure heart attacks prevented or lives extended, and no reader should upgrade it into that claim. The presenting investigator, Todd Villines of the University of Virginia, reports consulting work for Cleerly, the company whose AI analysis the trial tested, alongside two other firms, and Cleerly issued the release carrying these figures. The design still earns its place in this edition: randomization, 21 sites, a usual care comparison, and a patient group that looks like the clinic waiting room, women included at 57 percent. A person with chest pain can ask one useful question at his next appointment: will my workup measure plaque itself, or infer it from a stress test, and what will each answer change about my treatment.
Three seconds of audio can build the voice of the adviser she trusts.
FINRA says fake adviser sites now carry real registration numbers, and an AI search summary cannot clear the person asking for your money
The person in this story is an investor, often retired, who gets a message from someone presenting as a financial adviser. She does the careful thing. She asks for a video call, and the face on screen matches the photograph on the firm's website. She searches the name and reads a summary that sounds consistent. FINRA, the self regulatory body that oversees United States broker dealers, devoted an October 6 episode of its FINRA Unscripted podcast to why each of those checks now fails, reported this week by Traders Magazine. Christine Kieffer, who leads investor education at FINRA, put the working numbers on the record. A modern AI tool needs as little as three to 15 seconds of audio to clone and imitate a person's voice. A single clear photograph can support a basic face swap or a talking head impersonation. FINRA hears from victims who stayed suspicious of an online contact until the video call, and dropped their guard at the moment the face appeared, because seeing a person felt like proof.
The websites behind the voices have caught up with the voices. Mayur Patel, FINRA's director of financial crimes intelligence, said the fraudulent sites his teams identify have improved considerably against the versions of a year to a year and a half ago. Kieffer said some of those sites display the real Central Registration Depository number of a genuine broker or adviser. That number is the credential an investor is told to check, and a copied credential passes a shallow check while pointing at a person the site does not employ. Patel also closed the newest verification shortcut directly. An AI search engine is not a tool to verify stuff, he said. It is a more enhanced version of a Google search, and it pulls from what already sits on the internet, which includes material the fraudster planted. A summary that repeats a fake profile back to the investor adds confidence and adds no checking at all. The checks that still work sit outside the channel the caller controls: FINRA's own BrokerCheck database, reached through FINRA's site rather than through any link the adviser sends, and a phone call to the firm's published number, placed by the investor, to a person she reaches herself.
The scheme underneath the technology is old, and Kieffer named the signals that survive the upgrade. Urgency. Promises of guaranteed returns. Claims of exclusive access. Above all, a request for secrecy, which she flagged as the red flag people think about least. Secrecy does the same work here that it did in the executive voice fraud AI Next Wave covered on Friday: it cuts the second pair of eyes out of the decision. Her advice is to slow down and talk the opportunity over with someone else before any money moves, because a conversation the pitch cannot survive is the cheapest control a household owns. FINRA is building controls of its own. In March it launched a Financial Intelligence Fusion Center, a secure portal where member firms share fraud and cybersecurity intelligence, including how criminals use generative AI, and it offers firms a toolkit for taking down websites that impersonate them or their advisers. Patel expects losses from AI enabled fraud to keep rising as criminals adapt. Kieffer was blunt about the limit of warnings like this one. Telling investors to spot the fake cannot carry the whole burden, she argued, and institutions and regulators need ways to block the fraud before it reaches the investor. Until those blocks exist, the household version is short enough to tape to a monitor: voice and face prove nothing, secrecy ends the conversation, and verification starts from an address you typed yourself.
More than 100 people climbed onto the cranes at a site drawing power from a grid that cannot keep up
Amsterdam's congested grid is the fact under Friday's protest, and the project's permit predates the city's ban by more than four years
The pressure in this story sits with Dutch companies waiting for new or upgraded grid connections, and with the campaigners who say homes wait behind them, in a country where the electricity network is congested and connections take time to grant. On Friday, November 6, more than 100 activists entered the construction site of a data center campus in Amsterdam's Westpoort district, climbed cranes and construction structures, and hung banners against the expansion of AI infrastructure, according to reporting compiled from Reuters footage and Dutch police statements. Police detained between 150 and 200 people on suspicion of trespassing, and one officer sustained a minor injury. Protester Sam van der Beek told Reuters the facilities were destroying the world through the resources they consume. The action formed part of European action days on November 6 and 7 organized by a coalition calling for a halt to new large data center construction until stronger safeguards exist, with demonstrations in several other countries.
The project under the cranes carries the numbers that explain why a grid queue becomes a street protest. Pure Data Centers is building the campus, fully leased to Microsoft, with an investment above one billion euros: three towers, each about 85 metres tall, on about six acres, with a combined data hall capacity of 78 megawatts, according to the developer. The developer estimates more than 1,000 jobs during construction and about 80 permanent positions once the campus runs, in engineering, maintenance, security, and administration. The Dutch government's answer to parliamentary questions in April 2026 put the grid connection capacity at 99 megawatts, 33 megawatts per tower, and calculated that continuous running at that level would use about 0.87 terawatt hours a year. That figure is a theoretical calculation at full capacity around the clock, not measured use, and the government also recorded that the project will use no groundwater and was designed to send heat to a planned heating facility across the street. The argument in Amsterdam runs on the gap between those two sets of numbers: a connection sized near 100 megawatts granted to one tenant, in a grid where households and businesses queue, against a permanent payroll of about 80.
The approval history explains why the city's current policy did not stop the build. Amsterdam barred new data centers and expansions inside the municipality in April 2025, with the bar running until at least 2030, citing electricity demand and pressure on local resources. This project's building permit was requested in December 2019 and granted on November 6, 2020, five years before the bar, and officials say its built floor area sits below the ten hectare threshold in the national definition of a hyperscale facility, so the later national restrictions do not capture it either. A rule written in 2025 meets a permit granted in 2020, and the permit wins. Microsoft said it respects peaceful expression and does not condone unlawful entry or actions that put safety at risk, and the developer offered to discuss concerns once demonstrators left the site safely. No new decision came out of Friday. What Friday produced is a visible count of where the argument now stands in Europe: a European capital's fight arrived at the fence of a site whose paperwork closed before the rules changed.
Her balance check was promised for the weekend. The window had already closed on Friday morning, and the recovery has since moved in pieces.
Binance's proof for 44 tokens was due 50 to 60 hours after the November 4 snapshot, and what followed updates 8 and 9 was withdrawals reopening in phases, not a new recovery total
A correction and a status check belong together here, because AI Next Wave set the date. Saturday's Week in Review told Binance customers that the 50 to 60 hour window for the full zero knowledge proof behind the November 4 reserve snapshot landed over the weekend. The publish run's own audit record, logged in AI Next Wave's own audit ledger, puts the end of that window on Friday morning, November 6, Eastern time: on the counting from the snapshot at midnight UTC on November 4, 50 to 60 hours run out on Friday, not Saturday. The Review's timing was loose, and the correction stands here where the original audience reads. Zero knowledge proof is the method that lets a customer verify her own balance sits inside the exchange's total without seeing anyone else's account.
The recovery total has not moved since Wednesday, but the recovery itself has. Recovery Update number 7, published November 4, still carries the newest replenishment figure: about 98 percent of the amount taken in the theft, up from 95 percent in the November 2 update, with the remaining 2 percent in less liquid tokens still being bought through market and over the counter channels and no completion date given. What followed was not a new total. Recovery Update number 8 on November 5 and Recovery Update number 9 on November 6 reopened withdrawals for long lists of named assets and networks, in phases as security checks complete, and Binance tells customers to check the withdrawal page for their own asset and network rather than assume access has returned. The November 4 snapshot covers 44 tokens, up from ten, and the update said the full proof would take about 50 to 60 hours to generate, after which holders of those 44 tokens could check on the exchange's Proof of Reserves page that their balances are backed at least one to one. Binance says it is covering the losses from its own resources and has set aside more than the full amount needed. Those are the company's figures, carried here as company claims. A customer holding one of the 44 tokens should run the verification on the official Proof of Reserves page herself rather than take any edition's word for its state, and a customer holding a token outside the 44 still has no proof covering her balance. When an update appears with a replenishment figure above 98 percent or a completion date for the last 2 percent, that is the material change this thread is waiting for.
Opportunity Radar
A sextortion readiness program for schools and youth leagues answers the gap the Tate family's campaign names. The Shield Initiative already wrote the child facing half: Stop, Screenshot, Show Someone. The adult facing half is missing in most schools, and a small service could build and drill it: a named reporting route that reaches a person on duty the same day, a rule that no child is disciplined for coming forward, a screenshot and evidence routine staff can run without spreading images further, and a parent notification script that starts from the child as victim. School districts, private schools, and sports leagues are the buyers, and the Tate foundation's statewide campaign shows a working model a district can license the structure of rather than invent. Validation is one district running the drill with its own staff and students before any incident, scored on how fast a test report reaches the named adult and whether every adult in the chain gives the same first answer. Success looks like a route map posted where students can find it, retested each term. The service fails the moment it promises prevention, so it sells readiness, speed, and a record of the drill, and nothing beyond them.
An impersonation watch for small advisory firms fits the hole FINRA described. A two or three person firm has no intelligence team, and its exposure is now specific: a fake site carrying its real registration number, a cloned voice built from seconds of a podcast or a client webinar, and an AI search summary repeating the fake profile back to prospects as fact. A focused service could monitor for sites and profiles using the firm's name and registration number, run the takedown process FINRA's toolkit supports, and drill the firm's own callback rule: any instruction to move money gets confirmed on a number the firm already holds, with a second person approving. Independent advisers and small broker dealers are the buyers. Validation is a paid pilot with five firms, scored on fakes found that the firm had missed and time from detection to takedown request filed. Success is measured in live impersonations removed per quarter. The service fails if it sells monitoring without the drill, because FINRA's own account says detection by the investor alone cannot carry the burden, and the firm's clients need the callback habit in place before the call comes.
What You Can Do With This
If a child in your house uses Snapchat or any app with a chatbot at the top of the chat list
Teach the three Shield steps before anything happens: Stop, Screenshot, Show Someone. Say the second half out loud as well, that a child who sends a photo and gets threatened has committed no crime and will get help, not punishment, because fear of trouble is the silence the predator is counting on. Then open the app's chatbot and read its privacy terms for one answer: who receives the conversation. West Virginia's complaint alleges conversations go to outside companies for processing and ad selection. Your own reading of the current terms is the check available today, and a screenshot of those terms, dated, is worth keeping.
If you are pregnant or in the first year after a birth
Treat a general pregnancy chatbot as an information desk, and hold it to the Rosie standard: a year of use produced no measured gain in anxiety or depression against a monthly board book. Mood symptoms lasting more than two weeks, panic, or a feeling that you cannot cope deserve a clinician, and the National Maternal Mental Health Hotline at 1-833-TLC-MAMA takes calls free, around the clock, in English and Spanish. That number connects a mother to a trained counselor and local resources. In a crisis, call or text 988.
If chest pain has you in a cardiology workup
Ask which route your workup takes and what each answer changes. A stress test infers disease from blood flow under effort. A CT plaque reading measures the plaque itself, and the PARAMOUNT trial found the measuring route produced clearer diagnoses and far more starts on cholesterol lowering therapy over 180 days, in preliminary results awaiting peer review and a full publication expected in early 2027. Ask it concretely: given my result, what treatment starts, and when.
If anyone contacts you about an investment
Verify from an address you type yourself. Open FINRA's BrokerCheck through FINRA's own site and look the person and firm up there, because a website the adviser sends you can carry a real registration number on a fake page. Call the firm on its published number and ask for the person by name. Treat secrecy as the end of the conversation, and treat urgency, guaranteed returns, and exclusive access as the same signal in different dress. A video call settles nothing on its own: three to 15 seconds of audio can build the voice, and one clear photograph can build the face.
The Bigger Picture
Today's edition is a study in what a test can and cannot carry. West Virginia's complaint will be tested in court, over years, and its figures stand as allegations until then. The Shield steps need no verdict, and a parent can teach them tonight. Rosie was tested properly, with randomization and a year of follow up, and that is exactly why its empty result is worth more than a thousand satisfied user reviews: it tells a health system where the money does not work, and the RELAX trial beside it shows the shape of a digital tool that did move its target, narrow, structured, and aimed at one named pattern of thought. PARAMOUNT got the same proper test, inside its limits: randomized, compared against usual care, and still preliminary, with certainty and prescriptions as its measured wins and survival as the question it has not answered. FINRA's warning describes a market where the casual tests have collapsed: a familiar voice, a face on a screen, a registration number on a website, a confident summary at the top of a search page. Each of those used to count as a check. Each now takes seconds or minutes to manufacture.
Underneath the tests sits the same chain AI Next Wave has followed all month. A rule or a missing rule sets a control, the control shapes a product or a workplace, and the product lands on a person. Amsterdam shows the chain running backward through time: a permit granted in 2020 controls a build in 2026 that a rule written in 2025 cannot reach, and everyone waiting on that congested grid lives with the result. Binance shows the chain at the level of a promise with a clock on it, where the honest entry is a corrected date and a check any customer can run herself. A reader can sort next week's announcements with two questions that today's stories all answer in different ways. Who ran the test, and can I run any part of it myself. Where the answer to the second question is yes, a screenshot routine, a hotline number, a BrokerCheck lookup, a reserve verification, the household holds a piece of the control in its own hands. Where the answer is no, the dated record is what to keep: the complaint, the trial registration, the permit year, the update number. Paper outlasts the announcement, and this week handed over a good supply of both.
References
West Virginia Attorney General: Attorney General McCuskey sues Snapchat following two deaths, invests in education campaign, complaint allegations, the Tate and Thomas cases, and the Shield Initiative (Nov. 5, 2026) MedicalDaily: AI chatbot Rosie did not ease anxiety or depression in pregnant women and new mothers during yearlong trial, JAMA Network Open randomized trial of 400 participants (Nov. 6, 2026) Medical Xpress, from King's College London: Digital therapy offers new way to prevent worsening anxiety around childbirth, the RELAX randomized trial in JAMA Network Open (Nov. 4, 2026) HCPLive: AI driven quantification of coronary plaque outperforms standard care in suspected CAD, PARAMOUNT trial presentation at AHA Scientific Sessions 2026 (Nov. 7, 2026) BusinessWire via Morningstar: PARAMOUNT, the first randomized trial of AI enabled plaque based care, meets primary endpoint in late breaking results at AHA 2026, full preliminary figures (Nov. 7, 2026) Traders Magazine: FINRA warns AI is making investment fraud harder to detect, Kieffer and Patel on voice cloning, fake sites and verification (Nov. 4, 2026) FINRA: How AI is supercharging investment fraud and what you can do about it, FINRA Unscripted episode transcript and BrokerCheck resource (accessed Nov. 8, 2026) BraveNewCoin: Microsoft faces AI backlash as more than 100 protesters break into 1 billion dollar Amsterdam data center project, Reuters footage, developer and Dutch government figures (Nov. 8, 2026) Binance: Recovery Update 7, 98 percent of impacted assets replenished and proof of reserves expands to 44 tokens, snapshot and proof timing (Nov. 4, 2026) Binance: Recovery Update 8, withdrawals for more assets to reopen in phases (Nov. 5, 2026) Binance: Recovery Update 9, withdrawals for more assets to reopen in phases (Nov. 6, 2026) JAMA Network Open: Generative AI and maternal health outcomes in pregnant and postpartum individuals, the Rosie randomized clinical trial of 400 participants (Nov. 4, 2026)