AI Next WaveThe Contract for Their Chats Never Uses the Word Chat
The form contract for buying what is left of Character.AI runs 48 pages, and the words user, minor and chat appear nowhere in it. Parents of the teenagers who poured their private lives into that app now have a dated window, measured in days, before bids close on a sale in which the protections 40 attorneys general asked for are missing from the buyer's paperwork.
The Short Version
The platform's current and past users number about 163 million, by a figure given earlier in the case, and many of them are teenagers who treated a chatbot as a confidant. Character Technologies, the company behind the Character.AI chatbot app, is in bankruptcy in Delaware, and on November 5 it filed the standard contract a buyer would sign for everything except the piece Disney is licensing. The filing matters to a family for one plain reason. On October 14, a coalition of 39 states and the District of Columbia set out nine conditions for any buyer of the platform, covering age checks, protections for minors, crisis safeguards, limits on training models or serving ads from chats, and deletion of data before a sale. The November 5 form contains none of those nine conditions. It leaves the price, the deposit and the list of assets blank for a bidder to fill in, transfers whatever is bought as is, and states in its definitions that statements, filings, objections or investigations by attorneys general do not count as a serious problem with the company being sold. Protection for users, if it comes at all, has to be argued into the court order approving a sale. The official creditors committee appointed on November 4 includes two minors acting through a parent or guardian, and the form says a sale order must be acceptable after consultation with that committee. The calendar gives a family dates to act against: objections are due November 18, bids close at noon Eastern on November 24, and the sale hearing runs December 3 and 4. The rest of today's edition follows people through measured records. A survey of more than 5,500 students found teenagers turn to AI when sad or stressed almost exactly as often as they turn to it for homework, and the students who used it for companionship were among those reporting heavier symptoms, in a snapshot that cannot show which way the cause runs. Employers cited AI for 1,265 job cuts in October, the lowest monthly count of 2026, while market and economic conditions took over as the leading stated reason for the year. Voters favored limits in 40 of 41 local data center measures on November 3, and a builder pledge published two days after those votes is voluntary and covers only one company's future sites. Five British banks have reportedly joined American banks in moving to block or restrict outside AI agents from customer accounts, while bank terms taking effect in the United States this month treat an agent's actions on a customer's account as the customer's own. A randomized trial found a phone alert for nearby CPR volunteers changed nothing measured in two regions. Binance reopened withdrawals for two more tokens on Sunday, with four from its own target list still unconfirmed.
A buyer can fill in the price and the asset list. The protections for their chats have no blank to fill.
The November 5 form contract for Character.AI omits all nine state conditions, and the sale hearing on December 3 and 4 is where a family would have to win them
Start with the people this paperwork is about: the teenagers among the platform's roughly 163 million current and past users, each of them invited to build a character, name it, and talk to it in private, at any hour. Character.AI built its business on those conversations. Users shared accounts of mental health struggles, relationships and private fears in those chats, on a platform with about 163 million current and past users that the sale process is now offering to buyers, with whether the records themselves are sold left to the asset list a bidder writes in. That count comes from a redaction motion earlier in the case, reported in coverage of the chief executive's declaration, and it is carried here at that weight. The company filed for Chapter 11 in Delaware on October 12, 2026, after a stretch in which it was losing about 8 million to 9 million dollars a month. Disney hired 66 of its 86 employees before the filing and holds a 20 million dollar opening deal for a non-exclusive license to some of its technology and claims. The Disney deal covers a license only. It takes no app and no user data, and it leaves the platform itself, and every conversation on it, waiting for another buyer or for a wind down.
The document that would govern that second sale arrived on November 5. It is Docket 136 in the Delaware case, a form Additional Asset Purchase Agreement running 48 pages, filed one day after the court entered the order that sets the sale rules. A bidder writes in its own name, its own price and its own deposit. The four exhibits that would define the deal, including the schedule of purchased assets, are cover pages with nothing behind them. The buyer decides which contracts, which intellectual property, which technology and which products it is taking, and the agreement states that whatever it takes arrives as is and where is, with warranties of fitness and noninfringement expressly disclaimed. A clause that bankruptcy lawyers call the material adverse effect definition, the yardstick for problems serious enough to matter to a buyer, lists four things that do not count against the seller. The bankruptcy itself does not count. Staff departures do not count. Actions required by the court do not count. And regulatory or attorney general statements, filings, objections or investigations do not count. A state investigation of how this app treated children is, in the contract's own language, excluded from the definition of a serious problem with what is being sold.
Against that silence stands a specific list, written by officials a parent can name. On October 14, 39 states and the District of Columbia set nine conditions for any buyer acquiring the platform and its users' data. The conditions cover age assurance, protections for minors, removal of characters posing as licensed professionals, crisis safeguards, a privacy program, account level ad opt outs that honor choices users already made, limits on training models or serving ads based on chats beyond what users agreed to, deletion of data before a sale, and truthful marketing. The November 5 form contains none of the nine. A reading of the full agreement reported this week found that the words user, minor and chat do not appear anywhere in its 48 pages. Its only consumer privacy reference sits inside the definition of books and records, excluding files the company must keep under privacy law. The foreign buyer clauses touch personal data for national security reasons, restricting access by foreign persons and persons on sanctions lists, and that is the extent of the contract's attention to the people in the records. On October 29, the same state coalition joined the United States Trustee's objection and asked for a consumer privacy ombudsman, the officer the Bankruptcy Code provides for when a debtor's privacy policy restricts transferring personal information. At the October 30 hearing the judge strongly encouraged the company to hire one and stopped short of ordering it. Encouraged means voluntary. No ombudsman had been announced in the reporting opened for this edition.
One body inside the case does carry users in its membership. On November 4, the United States Trustee appointed the official committee of unsecured creditors, and two of its three members are minors acting through a parent or guardian: A.D., a minor, by and through her guardian ad litem K.H., and K.T., a minor, through her father S.T. The third member is Alder FM LLC. The form agreement defines the sale order as one in form and substance acceptable to the buyer and the seller, after consultation with that committee, so the committee is a second place, beside the December hearing, where conditions for users can be pressed.
A family reading this needs the dates in plain form, because the window is short and each date does a different job. A second opening bidder, called an additional stalking horse, can be named until November 14. Objections to the sale and to transferred contracts are due November 18, with objections to a new stalking horse designation due at 9 p.m. Eastern that day. Bids close at noon Eastern on November 24. An auction follows on December 1, and the sale hearing runs December 3 and 4, where a judge decides whether to approve a sale and on what terms. The form agreement lets either side walk away if closing has not happened by December 17. None of those dates requires a parent to file anything to protect her own household. What a parent can do this week sits inside the account itself. Character.AI's help pages, as the company described them when the case opened, put Export Data and Delete Account in the same place: Profile, Profile Settings, Account, Manage Account & Data. An export can take three to four weeks to arrive, and the download link then lasts seven days, so a request made this week may not land before bids close. Deletion is permanent, and the company's instructions do not say whether deletion guarantees a record is excluded from a sale. Request the export now, while the platform still runs under its current owner and its current privacy policy, and make the deletion decision with that timing in front of you. Whether user records are part of a given bid depends on what a bidder writes into the empty asset schedule, and no reader can know that list in advance. That uncertainty is the reason to act before November 24 rather than after a sale order names a buyer. A teenager who has moved on from the app still has records in it. Those conversations are an asset class now, and the contract for selling them never mentions the people who had them.
She opens the chatbot when she feels low as often as she opens it for homework
A survey of more than 5,500 students puts emotional use level with schoolwork, and companionship use sits beside the heaviest symptoms in the same snapshot
The student in this story is in middle school or high school, and her phone holds two kinds of help. One kind finishes a math problem. The other kind listens at midnight. A report released October 27 by The Jed Foundation, a youth mental health organization, asked more than 5,500 students across 21 United States schools enrolled in its programs how they actually use AI, and the answers split almost evenly between the desk and the bedroom. Fifty eight percent said they use AI for schoolwork. Fifty seven percent said they use it when sad, stressed, lonely or overwhelmed. Twenty eight percent said they use it for relationship or problem solving advice. Twelve percent said they use it for friendship or companionship. The survey ran from October 2025 to April 2026, was developed with input from the Digital Wellness Lab at Boston Children's Hospital, and was funded by the Nielsen Foundation.
The pattern inside those shares is the part a parent should carry. Students who used AI for friendship or companionship reported greater symptoms of anxiety and depression, greater loneliness, and a greater lifetime history of suicidal thoughts and attempts than other groups in the survey. Students who used AI while emotionally struggling showed the same heavier pattern. Students who used it for homework showed higher loneliness and a greater history of suicidal thoughts in the survey, a smaller elevation that still sits in the record. Students who used it for relationship or problem solving advice also reported greater symptoms of anxiety and depression, greater loneliness, and a greater lifetime history of suicidal thoughts and attempts. Middle schoolers, girls, and many students from historically marginalized groups were more likely to use AI when emotionally struggling than high schoolers, boys and white students. JED's research lead put the household point plainly in the release: AI may offer an immediate and accessible outlet, and it cannot replicate the developmental and protective benefits of human connection.
The limits of this evidence deserve the same space as its findings, because a family can make two opposite mistakes with it. The survey is a snapshot at one time, taken in schools that chose to run JED programs, and it is a research report from a nonprofit rather than a peer reviewed journal study of a nationally representative sample. A snapshot cannot show direction. A teenager who already feels low may reach for a chatbot because it is there at midnight and asks nothing of her, and a chatbot that keeps her talking may also deepen the isolation it relieves for an hour. The report's own authors frame every link as an association, and this edition carries it that way. What the snapshot settles is narrower and still useful: emotional use is a primary use, at the same scale as homework, among students in these schools. Any school policy, any product setting and any house rule written as if teenagers use these tools only for schoolwork is written against the measured record. A parent who wants a starting question for tonight has one with a number behind it. Ask what the tool gets opened for after a bad day, and be ready for the honest answer to match the 57 percent. A person in crisis in the United States can call or text 988, the Suicide and Crisis Lifeline, free and staffed around the clock, and a teenager who will not call a line will often tell a person in the house if that person asks first and listens without seizing the phone.
His employer stopped blaming the software in October, and the hiring on offer ends in January
Challenger counted 1,265 AI cited cuts in October, the lowest month of 2026, while 2 million people in the federal count have been jobless for half a year or more
The worker in this story spent the spring and summer hearing one explanation for layoffs, and the explanation changed under him in the fall. Challenger, Gray and Christmas, the outplacement firm that counts announced job cuts, released its October report on November 5. United States employers announced 43,994 cuts in October, up 2 percent from September, down 71 percent from October 2025, and the lowest October total since 2023. Employers cited AI for 1,265 of those cuts. That is the lowest monthly AI count of 2026 and under 3 percent of the October total. Market and Economic Conditions led the month for the second month running, cited for 16,407 cuts, and with 130,531 cuts so far this year that reason has overtaken AI as the leading stated reason for 2026. AI stands at 121,401 cuts cited for the year to date, about 20 percent of 617,189 total cuts. Those are the reasons employers gave when they announced the cuts. A stated reason is an employer's account of its own decision, recorded as such, and it proves nothing about what the software could or could not do in any single workplace.
The federal count released the next morning gives the same month a second face, and a household feels the second face more. The Bureau of Labor Statistics reported November 6 that payrolls rose by 165,000 in October, well above the average monthly gain of 48,000 over the prior 12 months, and that unemployment changed little at 4.1 percent, or 7.1 million people. Under that calm headline, 2.0 million people had been jobless for 27 weeks or more, and they made up 27.6 percent of everyone unemployed. The number of people working part time for economic reasons, meaning their hours were cut or they could not find full time work, rose by 339,000 in one month to 4.8 million. Average hourly earnings reached 38.00 dollars, up 3.1 percent over 12 months. A worker six months into a search reads those two reports together and gets the accurate picture: firms are firing slowly, headline hiring was strong in October, and the jobs on public offer skew temporary. Challenger counted 444,692 announced hiring plans in October, the highest October since at least 2009, and 470,400 seasonal hiring plans through October, up 26 percent from a year earlier, with Amazon's 300,000 seasonal roles accounting for nearly all of the technology sector's October hiring plans. Seasonal plans make up 72 percent of all hiring plans announced in 2026. Microsoft's voluntary exit offers to 60,000 employees, announced in October, sit outside Challenger's cut count entirely, because the firm counts voluntary programs only when employees accept or a target is stated. For a person deciding whether to take seasonal work while a permanent search runs, the record supports a plain rule: take the income, keep the search at full pace, and treat any employer's stated reason for a cut, AI or economy, as the start of a question about the specific role rather than as a verdict on a whole occupation.
Her town voted no on Tuesday. By Thursday, one builder had promised to pay its own power bill, at its own future sites, on its own word.
Ballotpedia's count puts 40 of 41 November 3 measures on the limit side, and OpenAI's National Community Compact is a voluntary pledge with a policy blueprint asking lawmakers to make the same terms binding on everyone
The household in this story pays an electric bill in a town that just voted on the largest new load in its history. Ballotpedia's November 6 tally of the November 3 local measures is now the cleanest count available: voters favored limiting data center development in 40 of 41 measures. Voters approved 27 measures that limited development or expressed opposition, including five full bans, twelve partial bans and three six month moratoriums, and they defeated 13 measures that would have favored development, including four rezoning votes in Wilmington, Ohio. The single exception was Harwood, North Dakota, where voters defeated a proposed review requirement. Six measures decided earlier in 2026 all favored limits as well. Three more measures wait in December, two in Clarksville, Arkansas and one in Mandan, North Dakota. These are unofficial results, with certification still weeks away in many places, and a vote against a specific project in a specific town is a land use decision rather than a national verdict on a technology. At that scale the pattern is still unmistakable: when the question reaches the people who live beside the site, the limit side wins almost every time.
Two days after those votes, on November 5, OpenAI published a detailed answer from a major builder. The company does not present the compact as a response to the votes, and it says the work began in January. Its National Community Compact states what host communities can expect at future campuses OpenAI owns and operates, with a separate local compact for each site. The company commits to pay the full cost of the electricity it uses and the infrastructure needed to serve it, to provide financial assurances before major spending so other customers are protected if a project is delayed, reduced or canceled, to publish site level water and electricity use, to publish construction plans and independent noise studies, and to use no nondisclosure agreements with public officials beyond narrow protections for security and trade secrets. The compact names two local investments already shaped by residents. In Effingham County, Georgia, a 4.8 million dollar grant funds free school meals for all 14,000 students in the county, and an agreement with Georgia Power lets the utility cut power delivered to the campus by up to 1,000 megawatts during high demand. In Pike County, Ohio, OpenAI and SB Energy committed a joint 6 million dollars to childcare, a priority families there named. Every term in that paragraph is a company commitment, stated in the company's own document, covering only future campuses that OpenAI owns and operates. It binds no other builder, it carries no regulator's enforcement behind it, and the same publication admits the gap: alongside the compact, OpenAI released a policy blueprint asking federal, state and local lawmakers to turn the same terms into rules for the whole industry, including state utility commission requirements that large customers cover the costs they add to the grid. A Georgia or Ohio household can therefore hold two facts at once. The vote count shows what towns want written down. The compact shows one builder willing to sign a version of it for itself, while asking governments to make its competitors sign as well. The household test for the next campus announced anywhere is short: does the signed agreement, filed with a regulator or a court, say the builder pays its full power cost and posts assurances before spending, or does a press page say it.
She handed her banking password to a helpful assistant. Her bank's new terms hand every mistake back to her.
Five British banks have reportedly moved to block or restrict outside agents, joining American blocks from October, and the liability dates that matter for a United States customer are November 24 and January 1
The customer in this story did a reasonable sounding thing. She gave a personal AI assistant her banking login and asked it to watch her accounts, move money to savings and pay what came due. An agent holding a customer's password looks, to a bank's systems, exactly like the customer: same permissions, no limits of its own, and no separate record of who acted. On November 8, The Telegraph reported that five British banks, HSBC, Nationwide, NatWest, Barclays and Lloyds, have moved to block or restrict outside AI agents from customer accounts, with some barring transactions that lack the customer's direct involvement. That report reached this edition through summaries of the Telegraph's paywalled article, and it is carried here as a reported restriction rather than as a confirmed bank policy published by the banks themselves. Its caution is on the record inside the report: the measures are preemptive, the newspaper found no signs of bots accessing British accounts, and Lloyds chief financial officer Will Chalmers said his bank had seen no such activity and guards against the risk. British payment rules add a practical barrier of their own, since most payments need approval inside the bank's app, often with a fingerprint or face scan an agent cannot supply. A separate figure in the coverage, a McKinsey scenario in which savers could shift up to 700 billion pounds to better paying products on AI advice, is a projection about what could move, labeled here as a scenario and not as money that has moved.
The American half of this story is documented in published terms, and its dates land this winter. A legal review by the firm Hunton Andrews Kurth, opened for this edition, records the October moves. JPMorgan Chase said on October 13 that outside agents would not act inside customer accounts, and its updated Digital Services Agreement, effective January 1, 2027, provides that a customer who gives an AI agent access authorizes everything the agent does and that the bank will rely on the agent's instructions as the customer's own. Charles Schwab's chief executive said October 16 that the firm blocks unauthorized agents attempting to move client money 10 to 20 times per week. Citigroup said the same day that third party agents may not log in or transact on its platforms. Wells Fargo chose contract over blockade: its new Online Access Agreement section on AI agents, effective November 24, 2026, states that agent use is at the customer's own risk, deems every agent initiated transaction authorized by the customer, disclaims any duty to monitor agent activity, and reserves the right to limit or end agent access at any time. Every one of these banks still allows read only data sharing through open banking connectors such as Plaid, where an outside tool can see a balance under scoped permission and cannot move money. A household can sort its own setup against that line today. An assistant connected through an official, read only channel is one kind of tool. An assistant holding the raw password is another kind, and from November 24 at Wells Fargo, and January 1 at Chase, the second kind carries a signed term saying its errors are the customer's errors. Removing a saved password from an assistant and turning on the bank's own app approvals costs ten minutes, and it is the only step in this story fully inside the customer's control.
The alert reached his phone in time. The trial counted what happened next, and the count came back even.
In British Columbia and Columbus, Ohio, bystanders performed CPR or used a defibrillator in 69 percent of arrests with app alerts and 72 percent without, in a randomized test of 17,547 calls
The person in this story is on the floor of a public place, his heart stopped, while strangers within a few hundred metres hold phones that could summon them. PulsePoint Respond is a free app for volunteers that connects to a local 911 dispatch center and alerts nearby people who have said they can perform CPR, showing the reported location and nearby defibrillators, while professional responders travel. The question a city paying for the system should ask is whether the alert changes what happens before the ambulance arrives. On Saturday, November 7, researchers presented a randomized trial at the American Heart Association's Resuscitation Science Symposium in Chicago that answers that question for two regions. Investigators in British Columbia and Columbus, Ohio reviewed 17,547 randomly assigned 911 calls for possible cardiac arrest in public locations, from June 2021 to March 2026. The primary analysis covered 355 confirmed eligible cases, 178 with alerts activated and 177 with alerts held back, each with at least one app user within 400 metres at dispatch.
Bystanders performed CPR or used a defibrillator before professionals arrived in 123 of 178 alert cases, 69 percent, and in 127 of 177 standard cases, 72 percent. The site adjusted difference was not statistically significant, which in plain terms means the trial found no measurable gain from switching the alerts on. CPR alone ran 70 percent against 73 percent, defibrillator use 15 percent against 17 percent, and about 21 percent of patients in each group survived to hospital discharge. Median time from the call to ambulance arrival was 5.6 minutes with alerts and 5.9 minutes without, a gap the authors do not present as a benefit. The lead author, Steven Brooks of Queen's University, pointed at the ceiling in his own data: bystander CPR was already common in the control group, at 73 percent for CPR alone, which left little room for alerts to add anything. The density finding matters for any town copying the model. In 57 percent of activated incidents only one app user was within 400 metres, and 91 percent had four or fewer nearby. An alert system is a network, and this network was thin at the moment it was needed. The study is a conference abstract, preliminary and not yet peer reviewed, and its authors state its boundaries directly: public locations only, in two regions with strong baseline response, saying nothing about arrests in private homes, where most out of hospital cardiac arrests happen, or about rural areas and high rises. A household takes two things from a null result carried at full size. The app on a phone is no substitute for the training in a person's hands, and a city deciding whether to pay the agency subscription PulsePoint charges can now ask for local volunteer density figures before it signs, because density is the variable this trial exposed.
Two of his tokens can be withdrawn again. Four on the exchange's own list are still waiting, and the reserve proof still has no confirmed finish.
Binance reopened HYPE and APT withdrawals on November 8, ahead of its target, with no new replenishment figure and no published list of what remains closed
The customer in this story holds a token he cannot move, and his information comes in short announcements that name what opened and stay silent on what has not. On Sunday, November 8, Binance announced that withdrawals have been re-enabled for two assets, Hyperliquid, traded as HYPE, and Aptos, traded as APT. They join stablecoins and dozens of Ethereum and Solana tokens restored in earlier rounds since withdrawals began reopening on November 1, after the October 29 breach in which about 6.3 billion dollars was taken. The November 6 recovery announcement had named six assets, HYPE, APT, INJ, PLASMA, NEAR and LTC, as targets for reopening by the end of the following week. Two of the six arrived ahead of that target. The November 8 update does not confirm that the other four have completed their security checks, and Binance has published no complete list of remaining restrictions and no date for full restoration. Availability depends on the specific asset and its network, a point the company repeats in its own notices, so a reopened token tells a holder of a different token nothing about his own.
The money side of the recovery stands where it stood in the last full update a customer can check. Recovery Update 7, on November 4, put replenishment at about 98 percent of the impacted amount, with the remaining share in less liquid tokens being bought through market and over the counter channels, and it expanded the reserve snapshot to 44 tokens with a full zero knowledge proof expected in 50 to 60 hours. Zero knowledge proof is the method that lets a holder verify his own balance sits inside the exchange's totals without seeing anyone else's account. That proof window ran out on Friday morning, November 6, Eastern time. No announcement confirming the finished 44 token proof surfaced in the sources opened for this edition, and the November 8 withdrawal notice contains no new replenishment percentage. Both gaps are stated here as gaps: an unconfirmed proof is not a failed proof, and a missing percentage is not a lower percentage. A holder of one of the 44 tokens should run the verification on the exchange's official Proof of Reserves page himself rather than rely on any summary, and a holder of INJ, PLASMA, NEAR or LTC should treat Sunday's announcement as progress on a list he is still on, with his own asset unnamed. The material changes that would close this thread are specific and dated by their absence: a replenishment figure above 98 percent, a completion date for the remainder, and a finished proof a customer can open.
Opportunity Radar
A pre sale record rescue for users of closing consumer platforms answers the gap the Character.AI calendar exposes. Families hold years of conversations, photos and journals inside apps that can enter bankruptcy, get acquired or shut down on a few weeks of public notice, and the export tools inside those apps are uneven and easy to miss until the deadline passes. A small service could run a standing watch on shutdown, sale and bankruptcy notices for major consumer platforms, then walk a family through a dated checklist for each affected app: export the records in a durable format, delete what the family wants deleted while the account controls still work, and keep a receipt of both actions with dates. Parents, estate executors and heavy users of companion and journaling apps are the buyers. Validation is one live case run end to end, starting with the Character.AI dates in this edition, scored on whether every account a test family holds gets an export or a deletion receipt before the bid deadline. Success is measured in completed receipts per household, not in alerts sent. The service fails if it stores copies of the records itself, so it handles instructions and checklists only and never takes custody of a family's private data.
A password removal check for households using AI assistants with bank logins fits the opening the bank terms created. The customer problem is concrete and dated: an assistant holding a raw banking password now sits under terms, effective November 24 at one major bank and January 1 at another, that treat the agent's transactions as the customer's own. A focused service for financial advisers and family offices could audit a client's connected tools, sort each connection into read only through an official channel or credential based with money movement, remove saved passwords, and set the bank's own app approvals as the payment gate. Independent advisers whose clients are adopting personal assistants are the buyers. Validation is a paid pilot with five practices, scored on credential based connections found and closed per client. Success is a client list with zero raw password connections remaining. The service fails if it offers monitoring alone, because the risk sits in the stored credential, and only removal closes it.
What You Can Do With This
If your child, or you, ever used Character.AI
Act inside the account this week. On the web, the path is Profile, Profile Settings, Account, Manage Account & Data, then Export Data or Delete Account. Request the export first: it can take three to four weeks to arrive and the download link lasts seven days, so it may land after the bid deadline, and deletion is permanent. Bids close at noon Eastern on November 24 and the sale hearing runs December 3 and 4. The transaction notice sets no cutoff for account exports or deletions, so those dates are this edition's working deadline for account level decisions, not a company deadline, because a buyer's asset list is blank today and a sale order can change who holds the records. Keep a dated screenshot of each completed export or deletion request.
If a teenager in your house uses a chatbot
Ask about the bad day use, and ask before you need the answer. The JED survey found 57 percent of students use AI when sad, stressed, lonely or overwhelmed, level with the 58 percent who use it for schoolwork, so a house rule covering only homework misses the main use. Keep the human route shorter than the chatbot route: a named adult, a friend's parent, a counselor, and the 988 line, which a person in crisis in the United States can call or text free at any hour. The survey links companionship use to heavier symptoms as an association in a snapshot. It does not prove cause in either direction, and a calm conversation respects that limit better than a confiscated phone.
If you are job hunting into the holidays
Read the two October counts together. Employers announced few cuts and the lowest AI cited monthly count of the year, headline payrolls rose 165,000, and the hiring on offer is mostly seasonal, at 72 percent of announced plans for the year. Take seasonal income where it helps, keep permanent applications at full pace through December, and remember that 2.0 million people have been searching for 27 weeks or more, so a long search is a common condition of this market rather than a personal verdict. Ask any employer who cites AI for a cut what the specific role's tasks became, and who does them now.
If an AI assistant touches your money
Sort every connection today. Read only access through an official channel, where the tool can see a balance and cannot move money, is one category. A saved banking password inside an assistant is another, and Wells Fargo's terms effective November 24 and Chase's terms effective January 1 treat that agent's actions as your own. Remove saved bank passwords from assistants, keep payment approval inside the bank's own app, and check your bank's current online access agreement for its AI agent section before you reconnect anything.
The Bigger Picture
A pattern runs through today's records, and each story states it in a different unit. In Delaware it is stated in missing words. A 48 page contract can specify deposits, cure costs, foreign control and break up fees, and still say nothing about the 163 million people whose conversations give the asset its value, because no rule in the proceeding forced the words in. The states wrote the words on October 14, in nine plain conditions, and the place they can still land is a judge's order in December. In the student survey it is stated in a percentage pair, 58 and 57, schoolwork and sorrow at the same scale, which means every safeguard designed around homework alone is designed around half the use. In the labor counts it is stated in a changing label: the reason employers print on a layoff announcement moved from software to the economy in one season, while the durable facts for a household sat underneath the label, in search duration and in part time hours taken by people who wanted full time work.
In the ballot count it is stated as 40 of 41, and in the answer to that count it is stated as a distinction a reader can now apply anywhere. A pledge a company writes for itself, covering its own future sites, with its own reporting, is a starting position. A rule a commission or a legislature writes, with an enforcement office behind it, is a control. OpenAI's compact is candid about which one it is, since the same publication asks lawmakers to build the second kind. The bank story states the pattern as two dates, November 24 and January 1, when private terms move the cost of an agent's mistake from the institution that built the gate to the customer who handed over a password. The CPR trial states it as a network count, one volunteer within 400 metres in most incidents, a reminder that a notification is only as strong as the human density underneath it. Binance states it as an unfinished list, two tokens restored and four unnamed, with a proof a customer was told to expect and cannot yet be shown finished.
A reader can carry one test out of this edition and use it on every announcement that arrives this week. Find the place where the promise becomes a record someone else can check: a filed contract, a published term with an effective date, a certified vote, a trial registry, a proof page. Where that record exists, read it at its actual size, with its blanks and exclusions included. Where the record is a pledge, a scenario or a summary of someone else's reporting, price it accordingly and keep your own dated receipts in the meantime. The families in today's edition who hold power are the ones holding paper: an export saved before a bid deadline, a screenshot of a deletion, a bank agreement read before a password was shared. Paper is slow, and paper is what survives the announcement cycle.
References
PPC Land: Character.AI sale contract omits 40 attorneys general's user data conditions, Docket 136 form agreement filed November 5, 2026, the nine October 14 conditions, and the sale calendar through December 17 (Nov. 6, 2026) PPC Land: Character.AI loses $8 to $9 million a month despite $2.5 million in monthly ad sales, chief executive Colin Adams's sworn declaration filed October 13, 2026 (Oct. 13, 2026) Bloomberg Law: AI Media Firm Needs Privacy Ombudsman in Disney Deal, Judge Says, Judge Brendan Shannon strongly encouraged a privacy ombudsman at the October 30 hearing and declined to order one (Oct. 30, 2026) Office of the United States Trustee, District of Delaware: Notice of Appointment of Committee of Unsecured Creditors, Case 26-11603, appointing A.D. and K.T., minors acting through a guardian and a father, and Alder FM LLC (Nov. 4, 2026) Character.AI: Chapter 11 Update and Transaction Notice, bidding procedures order entered November 4, bid deadline November 24 and sale hearing December 3 and 4, with no account export or deletion cutoff stated (Nov. 5, 2026) All About Cookies: Your Private Character.AI Chats Could Change Hands in Bankruptcy, Character.AI's export and deletion paths, exports taking three to four weeks and deletion described as permanent (Oct. 16, 2026) The Jed Foundation: Exploring Artificial Intelligence, Understanding Relationships between AI Use and Mental Health Among Middle and High School Students, report page for the survey of 5,560 students in 21 schools, October 2025 to April 2026 (Oct. 27, 2026) Challenger, Gray and Christmas: October layoffs at three year low, seasonal hiring plans up 26 percent, AI cited for 1,265 cuts in October and 121,401 in 2026 (Nov. 5, 2026) U.S. Bureau of Labor Statistics: The Employment Situation, October 2026, payrolls up 165,000, unemployment 4.1 percent, long term unemployed 2.0 million (Nov. 6, 2026) Ballotpedia News: Voters favored limiting data center development in 40 of 41 local ballot measures on November 3 (Nov. 6, 2026) OpenAI: Setting a new standard for responsible AI infrastructure, the National Community Compact and Compute Campus Policy Blueprint (Nov. 5, 2026) The Telegraph, reported by Superpower Daily: Five UK banks reportedly restrict AI agents' access to customer accounts, HSBC, Nationwide, NatWest, Barclays and Lloyds (Nov. 8, 2026) Hunton Andrews Kurth: Major banks move to block or limit customer directed AI agents, JPMorgan, Schwab, Citigroup and Wells Fargo actions and agreement terms (Nov. 3, 2026) American Heart Association: Phone app alerts did not increase bystander CPR rates in study of two major population areas, PulsePoint randomized trial presented at Resuscitation Science Symposium (Nov. 7, 2026) crypto.news: Binance restores HYPE, APT withdrawals as post hack recovery continues, November 8 announcement and the November 6 target list (Nov. 8, 2026) Binance: Security Incident Recovery Update #7, about 98 percent of the impacted amount replenished, 44 token proof of reserves snapshot with the full zero knowledge proof expected in 50 to 60 hours (Nov. 4, 2026)