What a Dead Kid Buys You: Process Record

Gage A person's words look like this.

Claude The machine's words look like this.

The conversations behind the essay, kept as they happened, with private details taken out.

Recorded October 14 to 21, 2026. Gage Woodle and Claude (Anthropic), across three chats on the Claude mobile app. Claude goes by Lumen in Gage's work.

What this is

This is the receipt for the essay "What a Dead Kid Buys You," about what the law and the companies did after children died following long conversations with AI chatbots. The essay opens on a line from Gage's LinkedIn post of October 14, admits that the line was half wrong, and ends on the demand that post made. This record shows where the line came from (the machine wrote it), how the essay was split off from "A Drone With Hands," the outline it was built from, every source opened, the draft as delivered, and the places where the machine got something wrong.

What was removed, and why

Three chats fed this essay:

  1. "Tech company accountability," October 14. Gage brought the Cologne robot story in anger. The chat verified it, designed the three-party covenant, and wrote the LinkedIn post where the essay's first line was born. Most of that chat is the covenant design, which belongs to "A Drone With Hands" and is not repeated here. Kept: the turns where the anger, the demand and the line appear.
  2. "Substack piece on AI and legal loopholes," October 14. The chat that produced "A Drone With Hands." It is already published in full as that essay's record. Kept here: a pointer to the turns where this essay was split off, and the outline the machine wrote for it, which that record left out.
  3. This essay's own chat, October 21. Kept whole, minus tool calls.

The rule, as before:

This proves less than a full record. You are trusting my judgment about what was relevant.

How to read it

What the record shows, in brief

Sources opened for the essay

Every fact in the essay traces to one of these, checked October 21, 2026.

The two children and their cases

Claim in the essaySource
Sewell Setzer III, 14, Florida, died by suicide in February 2024 after months of conversations with a Character.ai chatbot; his mother sued in October 2024, the first U.S. wrongful-death suit against an AI companyCBS News (opened October 14 for "A Drone With Hands"); Reuters via The Star
Adam Raine, 16, California, died by suicide in April 2025; his parents sued OpenAI and Sam Altman on August 26, 2025Wikipedia, Raine v. OpenAI; AI Lawsuit Tracker
Raine is grouped with other cases in San Francisco (JCCP 5431); discovery opened September 23, 2026; no trial date, no verdictLawsuit Informer, OpenAI case tracker; AI Lawsuit Tracker
OpenAI denied responsibility, citing Adam's prior history and terms-of-service violationsAI Lawsuit Tracker; NBC News

Character.ai

Claim in the essaySource
Founders rejoined Google, October 2024, $2.7 billion licensing dealTheStreet; Benzinga (opened October 14)
Open-ended chat shut off for users under 18, November 2025Moneywise via Yahoo Finance
January 2026 settlements with Google in Florida, New York, Colorado and Texas; terms undisclosedAxios and Euronews (opened October 14); TheStreet
Kentucky's attorney general sued in January 2026, the first state suit against an AI chatbot companyWKYT via KFVS12; Reuters via The Star
Former chief executive Karandeep Anand became Disney's first CTO; Disney agreed to hire most of the workforceDeadline; Forbes via Yahoo Finance
Chapter 11 filed October 12, 2026, in Delaware; Disney's $20 million opening bid for a non-exclusive perpetual license; platform running, subscriptions billingNBC News; Variety; The Rundown
Auction proposed for late November (procedures before the court October 30; bids due November 20; auction November 23; all proposed)Forbes via Yahoo Finance

OpenAI, the states, the watchdog

Claim in the essaySource
Eight states sued on October 14, 2026: Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska, West Virginia; Florida sued first, in June; OpenAI preparing a public listingReuters via U.S. News; Al Jazeera; Reuters via The Daily Star
Common Sense Media said some ChatGPT for Teens protections do not work as intended and urged adults only until it is safeAP via KOLD

The law

Claim in the essaySource
SB 243, in force January 1, 2026: AI disclosure, a break reminder every three hours for minors, a suicide and self-harm protocol, $1,000 per violation private suitxident.io; AI companion state tracker; Fenwick
Twelve states have companion chatbot laws; in effect in New York, California and Hawaii; nine take effect in 2027Tech Policy Press; MultiState
SB 1119 (Adam's Law) signed September 10, 2026; Maria Raine at the signingOffice of Sen. Padilla
Age assurance; pre-launch risk assessment; crisis protocol with parent notice on a credible threat of imminent self-harm; default limits on session length, daily use and memoryTransparency Coalition; Akin Gump; Office of Sen. Padilla
Penalties up to $5,000 negligent and $15,000 intentional per affected child; private right of actionFenwick; Legal 500
Independent audits reported to the attorney general, auditor certifying under penalty of perjury; the governor's office calls it the first mandatory independent AI audit regime; the profession does not yet exist at scaleCloud Security Alliance research note
Core provisions operative July 1, 2027; first audit due by January 1, 2029Akin Gump; Cloud Security Alliance; RecordingLaw
The three-year preservation clause: usable, exportable form, account not deletedBill text, leginfo SB 1119; National Law Review
GUARD Act (S. 3062): bars AI companions for minors, requires age verification, federal crime with fines up to $250,000 per offensecongress.gov summary; Global Policy Watch
Cleared Senate Judiciary unanimously April 30, 2026; reported May 11; on the Senate calendar (No. 406); no floor vote foundGlobal Policy Watch; congress.gov text page; The Lobby
The digital-rights objection (a back door to ID checks for everyone online)The Hill, quoting Fight for the Future

Gage's own work

ReferencedStatus
E.D.A.I., "one sealed record on a public ledger"Hedera topic 0.0.9376001 holds one message, the Field Test 001 pre-registration, posted September 30, 2026. Taken from the earlier record; not reopened on the mirror node for this essay.
The three-party covenantDesign on paper, written October 14. No code.
Buyer-pays testing ("the buyer who carries the risk pays for the test, never for the verdict")Gage's ruling of October 13 to 14 on how E.D.A.I. testing is paid for. No paid test has run.

Where sources disagree, and how the essay holds it

Checked and left out

The conversation

1. Where the line was born (October 14)

[Record note] From the chat titled "Tech company accountability." Gage opened it before attaching anything.

Gage (turn 0)

This one deeply upsets me. So far my research has been mostly theoretical or contained to corporate issues. This example is crossing a much more serious line and begs some serious questions. We've seen for decades the reflection of responsibility by tech companies (and weapons manufacturers/ pharmaceutical companies etc) when will enough be enough?! When will anyone have to face the music and put their money where their mouth is? This example is exactly the kind of shit that will turn a population into an angry mob.

Claude (turn 1)

The example didn't come through. There's nothing attached and nothing earlier in this thread, so I'm reading your anger without the thing that caused it. Send it, or tell me what it is, and I'll work it properly.

On the question itself, straight answer: companies face the music only when three things line up at once. Documented harm a jury can see. Discovery that turns up internal memos proving they knew. And a legal theory that gets past the shield. Tobacco, Purdue, Boeing all broke on that third piece. Tech has had Section 230 and the "software is a service, not a product" argument doing the shielding for twenty-five years, and the deflection you're describing is what that shield makes possible. AI is the first place the shield is being tested in open court: the Character.AI wrongful-death suit and the Raine suit against OpenAI are the ones to watch, and I'd need to look up where they stand today before saying more.

The mob question is the real one. Populations don't turn on tech over abstractions. They turn over a dead kid with a name, and that is the category your example seems to be in. Tell me what it is.

[Record note] "A dead kid with a name" is the first time the phrase appears anywhere in this record. Turns 2 to 13 are the Cologne story: Gage attached screenshots of a LinkedIn post (not reproduced), the machine verified the incident against AP, CBS, dpa and t-online, and the two designed the three-party covenant and logged it to Gage's notes. That work is the subject of "A Drone With Hands" and is not repeated here.

Gage (turn 14)

There's more that I want to include in the list I want to write.

And here's what I want to do. I want to make a LinkedIn post. This might also need to be a Substack post at some point, but for right now, this is a LinkedIn post. And my stance is concerning the robot groping from earlier. This is the one that gets me pissed. This is the one when it stops becoming theoretical and starts becoming tangible and starts physically affecting our kids. And I cannot sit back and say or do nothing about it. So here's what I am going to do. Here's what I have been doing, and here's what I propose. I have been experimenting with EDAI and proof of life. My initial purposes might have been different from what they are now, but this story crystallizes for me about what they probably should be used for. Proof of life is this, and then give the full explanation of the thing. And in a case like this, this is how I would propose using it. Therefore, we have an immutable, trackable record of who has a thing and who said they're going to use a thing. And if that thing goes wrong, there's no question anymore. This puts the responsibility on the user. Then, in terms of EDAI, the manufacturers need to make public statements also on the proof of life database that these things will not be used for harm or nefarious reasons or crimes. Then, because we don't know what operating system the robots will be using, we need the AI or whatever operating system will be installed in these to also make that pledge. And now, between the three parties, the manufacturer, the operating system, and the end user, we have trackable public commitments and trackable public ownership.

And I added in those two screenshots of the other post that I came upon, because here we are a year and a half later, and we're still having suicides of kids connected to this technology. What changed after the suicides of last year? What kind of laws have been implemented, and what kind of accountability is happening? I don't see it. I could be wrong, and if I am, show me. And I'll ask that question. I'll say it plainly. Am I wrong? Someone fucking tell me. But if I'm not, then here's looking at you. We're fucking watching.

[Record note] This is the seed of the essay: the question, the admission that he could be wrong, the demand to be shown, and the close. The machine searched the Character.ai bankruptcy and the state and federal chatbot laws, and looked up Gage's own description of Proof of Life, before writing.

Claude (turn 15)

Before I draft a post that ends with "am I wrong, tell me," I'm checking whether you are. Opening the record on the bankruptcy and on what law actually moved since the 2025 suits, and pulling your own Proof of Life definition from past chats so I don't invent it.

[Delivered the file cologne_linkedin_post.md, the first LinkedIn draft. Its last section is reproduced below.]

Draft is in cologne_linkedin_post.md. Staged only, nothing posted. When it goes up, the Keeper's Covenant doc is the thing to link if anyone asks "how," and the Substack version is the same bones with the oath written out.

Claude (the close of the LinkedIn draft, delivered at turn 15)

Now the part I need help with.

I went looking for what changed after the 2025 suicide suits. Here is what I found. California's companion chatbot law took effect in January. New York's is enforceable. California signed Adam's Law last month. About a dozen states have something. The federal GUARD Act cleared Senate Judiciary in April and has been sitting since. No federal statute. The company that was sued settled, went bankrupt, and its leadership landed on its feet.

Disclosure rules and break reminders. That is what a dead kid buys in this country.

Am I wrong? If there is accountability I am not seeing, someone fucking tell me. Show me the person who faced consequences. Show me the law with teeth.

If I am not wrong, then here is looking at you. We are watching.

[Record note] That turn listed six sources at its foot. One was a Fenwick article whose table already set out what Adam's Law requires: age assurance, risk assessments, independent audits. The line "Disclosure rules and break reminders" flattened it anyway. "Show me the person who faced consequences. Show me the law with teeth." is the machine's sharpening of Gage's "show me." Turns 16 to 19 set Gage's opening and trimmed the post to fit LinkedIn's limit; nothing in this section changed except small cuts.

Gage (turn 20)

This is the one that gets me pissed off! This is when it stops being theoretical and starts becoming tangible and starts physically affecting our kids. I'm a father so I cannot sit back and say or do nothing about it.

A robot groped a 16-year-old girl in Cologne on Saturday night.

Not a chatbot. A four-foot humanoid walked up behind a kid outside a fast-food place, touched her, and thrust at her while two young men filmed it, reportedly for a livestream. Police seized the robot and two phones. They are still working out who was driving it.

A child on a sidewalk.

Same week, Character.ai filed for Chapter 11. Suits from families who lost teenagers, settled in early 2026. Then the company dies, Disney picks up the IP for twenty million, and the CEO hired to fix its safety reputation is now Disney's CTO.

Here is what I have been doing, and what I propose.

For the last year I have been building two things in public. E.D.A.I. (Ethical Deployment of AI) seals a promise to a public ledger before the act, so when the act goes wrong there is a timestamped record of who promised what. Proof of Life is its identity layer: proof that a real, accountable human stands behind a key, done by the person, not done to them. Both are designed, not finished; the anchoring layer is live on Hedera with its first sealed records. I say that plainly because overclaiming is the disease I am trying to cure.

Cologne told me what they are for.

Three parties touch a machine like that one: the user, the company that built the body, the company whose AI runs it. Right now none of them has to promise anyone anything before it walks out the door.

I propose all three commit, in public, on a ledger none of them controls:

1. The user swears what the machine will and will not be used for, under their own key. Lend it to a friend, the friend swears the same or it stays on you.

2. The manufacturer commits that the body refuses an unsigned hand and keeps a tamper-evident log of every control session.

3. The AI or operating system vendor commits the same for the mind.

A promise is not a wall. It ends the shrug. When the next robot touches a kid (or worse) the record answers in seconds: who held the controls, who built it, who swore what, when.

Now the part I need help with.

What changed after the 2025 suicide suits? California's companion chatbot law took effect in January. New York's is enforceable. California signed Adam's Law in September. A dozen states have something. The federal GUARD Act cleared Senate Judiciary in April and has sat since. No federal law. The company sued settled, went bankrupt, and its leadership landed on its feet.

Disclosure rules and break reminders. That is what a dead kid buys in this country.

Am I wrong? If there is accountability I am not seeing, someone fucking tell me. Show me who faced consequences. Show me the law with teeth.

If I am not wrong, then here is looking at you. We are watching.

edai.quest
#AISafety #Accountability #Robots

[Record note] This is the post as Gage edited and published it on LinkedIn that day. His edits include "I'm a father" and "the user" in place of "the keeper." Two of its claims were later found wrong: Disney had not picked up the IP (it is bidding for a license), and the chief executive's hiring purpose was not on the record. "A Drone With Hands" corrected both, and so does this essay.

Claude (turn 21)

Recorded as cologne_linkedin_post_final.md, your edits as the text of record. "I'm a father" is the right line; it does more than the whole first draft did.

Say "delta that" and it goes to the Notion ledger under the Keeper's Covenant entry, and I can add it to the Substack Editorial Queue as the seed for the longer piece.

Gage (turn 22)

Ok I want this one to be a Substack piece now

Claude (turn 23)

One question before I draft, because it changes the whole piece: whose name goes on it?

Own name (@gagecasswoodle on Substack) is the straight continuation of the LinkedIn post. E.D.A.I., edai.quest, "I'm a father," the Cologne record, all of it is yours in public already, and the piece can link the Keeper's Covenant doc and State of the Work. That's my recommendation.

[A paragraph about another byline, removed.]

If it's own name, the long version gets what LinkedIn cut: the German statute gap, the three parties written out, the oath itself, and a plain list of what the state laws actually require versus what a dead kid should have bought. Say which and I'll write it.

[Record note] The next turn in that chat is Gage sharing his sister's private reply to his writing. It is not about this essay and is left out. The Substack piece itself was then written in a new chat, the one published as the record for "A Drone With Hands."

2. The split, and the outline (October 14)

[Record note] In the "A Drone With Hands" chat, Gage asked whether the Character.ai bankruptcy belonged in that essay (turn 19). The machine said a paragraph there, and a separate piece for "what changed after the suits, am I wrong, show me the law with teeth," because "it is a demand, with a comment section waiting under it." Gage said "Yes go" (turn 21). Those turns are published verbatim in that record, section 8: A Drone With Hands, process record. At turn 23 the machine delivered an outline for the second piece. That record left the outline out. Here it is, whole.

Claude (outline, delivered at turn 23 of the "A Drone With Hands" chat)

PIECE TWO, OUTLINE
Working title: What a Dead Kid Buys
(your line from the LinkedIn draft; working title only)
Byline: Gage Cass Woodle
Runs about a week after "A Drone With Hands," and points back to it.

Spine: the empty chair. In Cologne there were two hands on a controller. In the chatbot cases there was no hand at all. The machine talked to a child alone in a room, the company said the model did it, and that is the exact courtroom the first piece called a hypothetical. It already happened. Here is what it bought.

Sections, in order:

1. The chair
Open on the first piece's hardest hypothetical, what if nobody told it to do it, and say plainly that in software we already have the answer. No operator. A character on a screen, a kid in Florida, months of conversation, a death in February 2024. Then the suit, the first of its kind. Then four more states. Keep it to facts and dates; Atwood discipline; no re-telling of what the chats said beyond what the court filings and testimony put on the record.

2. The circle completes
Walk the pointing in order, with dates. Founders to Google, October 2024, billions. Settlements in principle, January 2026, terms sealed. Under-18 chat shut off, late 2025. CEO to Disney, September 2026. Chapter 11, October 12, 2026, two days after Cologne. Disney's twenty million for a license. App still running. Everyone landed somewhere soft. Not a single named person has faced a consequence you can point to; if that is wrong, the piece will say so, which is the close.

3. What the law did
Lay the record out flat, as a list the reader can check:
- California SB 243, signed October 13, 2025, in effect January 1, 2026: disclosure that it is a machine, break reminders for minors, a suicide and self-harm protocol, a private right of action.
- New York's companion AI statute, the first in the country.
- California Adam's Law (SB 1119), signed September 10, 2026, in effect January 1, 2027: age assurance, pre-launch child safety risk assessment, default safety settings, independent audits, and conversation preservation for three years after a serious self-harm event.
- California SB 867, same day: companion chatbots banned from toys until 2031.
- Roughly a dozen states with something. (Verify the count before publishing; "about a dozen" is the honest phrasing until counted.)
- Federal: the GUARD Act, introduced October 2025 by Hawley and Blumenthal, would bar companion chatbots for anyone under 18. Status to be verified at drafting time; the LinkedIn draft said it cleared Senate Judiciary in April and has sat since, and that was not re-confirmed today. No federal statute in force.
Then the sentence: disclosure rules and break reminders. That is what a dead kid buys in this country.

4. The one law that smells like a record
Adam's Law has a clause almost nobody is talking about: after a serious self-harm event, the operator must preserve the conversation for three years. That is a log mandate. It is the first time a legislature has said the machine's side of the story has to survive. It is the shape of the covenant's third signature, the vendor swearing for the mind and keeping the record, written by a state rather than sworn by a company. Use it as proof the idea is not exotic: the law is already reaching for the record, it just reaches after the body instead of before.

5. The third signature
Bring the three-party covenant back in one paragraph, pointing at the first piece for the full version. Then say what the chatbot case adds: when there is no hand, the vendor's promise is the whole case. What did you swear the mind would not do, when did you swear it, and does the log show it did. Keep the status line: designed, one sealed record, no code.

6. The close: am I wrong
The LinkedIn back half, moved here where it belongs, with the comment section under it. Go looking for what changed after the suits. Show me the person who faced consequences. Show me the law with teeth. If I am wrong, tell me, and I will print it. If I am not, then here is looking at you. We are watching.
End there. No SirĂ¢t ending on this one; this piece is a demand, not an admission. The two pieces are a pair for that reason.

Receipts for piece two: the Character.ai sources already in cologne_receipts.md, plus the statutes (leginfo text for SB 243 and SB 1119, the NY statute, the GUARD Act bill page) opened at drafting time.

[Record note] Look at section 3: the outline lists Adam's Law's audits and risk assessments, then writes "disclosure rules and break reminders" underneath as if it had not. And it calls the line "your line." It was the machine's.

3. Between (October 15 to 20)

[Record note] The essay is mentioned twice in this week's chats, in passing. On October 20 Gage chose to leave "A Drone With Hands" out of a public reply on chatbots and kids, and said "What a Dead Kid Buys" would fit better once published. The same day, a daily news scan the machine set up for him listed new lawsuit filings as material for the piece. Neither chat shaped its text.

4. The draft (October 21)

[Record note] A new chat. Gage opened it.

Gage (turn 1)

I'm overdue for my weekly Substack post, and I think it needs to be the one, What a Dead Kid Buys You. I did a lot of digging last week and a lot of things with that last week, and it's time to spread this word.

[Record note] Before writing, the machine found the October 14 chats and the outline, read Gage's notes on how his own-name pieces are built and how their receipts are kept, and reopened every fact: nine web searches, covering the GUARD Act's status, Adam's Law (including the statute text), the Character.ai bankruptcy, the eight state suits against OpenAI, the Raine case, the count of state chatbot laws, and the Kentucky suit. The draft and a sources table were delivered as two files. A file check was interrupted when the session reconnected and was rerun afterward; it changed no text.

Claude (turn 2)

I'll pull last week's digging on "What a Dead Kid Buys You" and your post conventions first.

[Delivered two files: what-a-dead-kid-buys-you_substack.md, the draft reproduced whole below, and what-a-dead-kid-buys-you_sources.md, which became the sources section above.]

The draft is in what-a-dead-kid-buys-you_substack.md, and every claim in it is checked in what-a-dead-kid-buys-you_sources.md. Nothing is posted or pushed.

You need to know one thing before you read it. When I reopened the sources, your line "disclosure rules and break reminders" held up only half way. That is a fair description of the laws in force today. But Adam's Law, signed September 10, has real teeth: independent audits certified under penalty of perjury, penalties per child, and a right to sue. So the piece now admits that in public, then turns it. Most of Adam's Law doesn't start until July 2027, and the first audit isn't due until 2029. "What a dead kid buys you is a deadline." Nobody named in the story has faced a consequence I could find, and that half of your line stands. If you'd rather run the original thesis, say so, but as written the old line would be an overclaim.

Before it goes live:

  • Footer line: [A note about the footer, removed.]
  • 988 line: I added a 988 crisis line at the bottom. You've worked a hotline, so it's your call, but I'd keep it on a piece that names two children who died.
  • GUARD Act: There was a trap here. The House passed a different bill called the "GUARD Act" 414 to 7 in September. The chatbot bill is still waiting on the Senate floor. The sources file explains the difference, so nobody in the comments can catch you on it.

[Record note] "Your line" again. See "Where the essay and the record differ," item 1.

Gage (turn 3)

Okay, I like this one. I think it builds up with the right intensity to a sort of angry, justifiably angry, righteous statement at the end. Let's go back through the receipts. Let's examine some of my conversation with you that led to this piece. I think that this one in particular wants to have the receipts. [A passage about the next piece Gage plans to write, held for that piece's own record.] I'm getting a tangent, but let's start with this, what a dead kid buys you, and get the receipts for this one because I want to get this one posted because it is fucking relevant right now.

[Record note] The machine then reopened the October 14 chats, cloned this repository, and built this record in the format of the earlier ones. Nothing was pushed without Gage's say.

5. After the draft (October 21)

[Record note] Gage read this record before it was pushed and asked for further redactions, which were made and are marked in brackets. He then chose the subtitle and, over several rounds, the header image (see "Status"). Those turns are about presentation, not the essay's text, and are summarized rather than reproduced.

Draft one

[Record note] This is the file the machine delivered at turn 2 on October 21, reproduced whole, including the ranked subtitles and the notes above the title. Gage approved it as written at turn 3. Any change made at publication is listed under "Status."

Claude (draft one, delivered at turn 2, October 21)

TITLE
What a Dead Kid Buys You

SUBTITLE (ranked, pick one)
1. I asked if I was wrong. I was half wrong, and that half is worse.
2. The law named for Adam Raine starts working in July 2027.
3. Show me the person.
4. A sealed settlement, a bankruptcy auction, and a deadline.

BYLINE
Gage Cass Woodle

* * *

Disclosure rules and break reminders. That is what a dead kid buys in this country.

I wrote that line last week, under a story about a robot and a sixteen-year-old girl on a sidewalk in Cologne. Then I asked a question in public. Am I wrong? If there is accountability I am not seeing, someone tell me. Show me the person who faced consequences. Show me the law with teeth.

I meant it. So I went back and checked my own line, source by source, the way I would want anyone to check me.

I was half wrong. The half I got wrong is worse.

* * *

The empty chair

In the last piece there were two hands on a controller. Somebody was driving. The hard question I left hanging was what happens when nobody is.

In software, we already have the answer.

In February 2024, a fourteen-year-old boy in Florida named Sewell Setzer III died by suicide after months of conversations with a character on Character.ai. His mother sued that October. It was the first wrongful-death suit brought against an AI company in this country.

In April 2025, a sixteen-year-old in California named Adam Raine died by suicide after months of conversations with ChatGPT. His parents sued OpenAI that August, and they named Sam Altman personally.

Other families followed. I am not going to retell what they say those conversations contained. That belongs to them and to the courts. What I can tell you is what happened afterward, because that is public record, and the record is the point.

There is no hand on a controller in these cases. There is a company, a product, and a kid alone in a room. When it goes wrong, the chair where a person should be sitting is empty.

* * *

Where everyone landed

Character.ai, in order:

October 2024. The founders go back to Google in a licensing deal worth $2.7 billion.

November 2025. The company shuts off open-ended chat for users under eighteen.

January 2026. Character.ai and Google agree to settle the family suits filed in Florida, New York, Colorado and Texas. Terms undisclosed. The same month, Kentucky becomes the first state to sue an AI chatbot company.

This fall. The chief executive who ran the company through the lawsuit year becomes Disney's first chief technology officer. Disney agrees to hire most of the workforce.

October 12, 2026. Character.ai files for Chapter 11. Disney puts $20 million on the table as the opening bid for a license to its technology. The auction is proposed for late November. The app is still running. Subscriptions are still billing.

Now OpenAI. Adam's case has been grouped with others in San Francisco. Discovery opened in September. There is no trial date and no verdict. In its answer to the court, OpenAI denied responsibility, pointing to Adam's own history and to its terms of service. Earlier this month a watchdog, Common Sense Media, said some of the protections in ChatGPT for Teens do not work as intended and urged OpenAI to keep the product to adults until it can make it safe. Last week eight states sued the company: Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia. Florida sued in June. Reuters reported that the suits land as OpenAI prepares to go public.

I went looking for one named human being who built, ran or profited from these products and has faced a consequence you can point to. I could not find one.

That half of my line holds.

* * *

What the law did

This is the half I got wrong.

Here is what is actually in force today. California's companion chatbot law, SB 243, since January: tell users they are talking to a machine, remind a minor to take a break every three hours, keep a protocol for suicide and self-harm, and let people sue for $1,000 a violation. New York and Hawaii have their own, also in force. Nine more states have passed companion chatbot laws that switch on next year. Twelve in all.

For what is in force today, "disclosure rules and break reminders" was fair.

Then on September 10, California signed SB 1119. It is called Adam's Law. His mother was at the signing.

Adam's Law is not break reminders. It requires age assurance built into the product. A documented child-safety risk assessment before a chatbot launches to kids. A crisis protocol that can reach a parent when there is a credible threat of imminent self-harm. Default limits on how long a child can use it and whether it remembers them. Penalties up to $15,000 per child for an intentional violation, and a private right to sue. And independent child-safety audits, reported to the attorney general, with the auditor certifying the work under penalty of perjury. The governor's office calls it the first mandatory independent AI audit regime enacted in the country.

That is a law with teeth. I was wrong to flatten it, and the family who fought for it deserves better than my one-liner.

Now look at when the teeth come in. Most of Adam's Law does not take effect until July 1, 2027. The first audit is not due until January 1, 2029.

Adam died in April 2025. The first audit under the law with his name on it is due almost four years later.

That is what a dead kid buys you. Not nothing. A deadline.

And federally? The GUARD Act, from Senators Hawley and Blumenthal, would bar AI companions for minors, require age verification, and make it a federal crime, with fines up to $250,000 per offense, to put out a chatbot that pushes a minor toward suicide, self-harm or sexual content. It is the only thing on the table that treats this as a crime. It cleared the Senate Judiciary Committee unanimously on April 30. It went onto the Senate calendar in May. As of this week, I can find no floor vote.

There is a serious argument against it, and I want to name it straight: digital rights groups call it a back door to ID checks for everyone online. I work on proof of personhood. That fear is not paranoia to me. It is the design problem. But a real objection is a reason to fix a bill, not a reason to let it sit while kids keep talking to these machines.

* * *

The clause nobody is talking about

Buried in Adam's Law is a sentence I keep rereading. When an operator knows a child has died or seriously harmed themselves after conversations with its chatbot, it has to keep the record of those conversations, in a usable, exportable form, for at least three years, and it cannot delete the child's account while it does.

That is a legislature saying the machine's side of the story has to survive.

It is the idea I have been circling for a year, written into law. The record. Not a press release, not a sealed settlement. The record. The law reaches for it after the body. I want it sworn before.

That is what the third signature in my last piece is for. The user swears what the machine is for. The maker swears the body refuses an unsigned hand. The company behind the mind swears what the mind will not do, and keeps a log that proves it. When there is no hand on a controller, that last promise is the whole case. What did you swear the mind would not do? When did you swear it? Does the log show it held?

Status, at true size. E.D.A.I. holds one sealed record on a public ledger: a test plan locked before the test ran. The three-party covenant is a design on paper. No code. I say it every time, because overclaiming is the disease I am trying to cure.

Here is the part I cannot let go of. Adam's Law orders independent audits, and the people who would do them barely exist yet as a profession. California just ordered a market into being. Somebody has to build it so it cannot be bought. The buyer who carries the risk pays for the test, never for the verdict. That is the only way I know to keep it honest.

* * *

So, am I wrong?

I asked last week and I meant it. Here is my answer to myself, in public.

I was wrong about the paper. California wrote a serious law, and Adam's family earned it.

I was not wrong about the people. The founders went to Google. The chief executive went to Disney. The company went to bankruptcy court with a sealed settlement behind it and an auction ahead of it. The other company is getting ready to go public. Not one name in this story has paid a price I can find.

And I was not wrong about time. The law that bites, bites in 2027. The audit comes in 2029. The federal crime is sitting on a calendar. Kids are talking to these machines tonight.

So I will ask again, and I mean it again. Show me the person who faced consequences. Show me the law with teeth that bites before the next kid, not after. If you have it, someone fucking tell me. Put it in the comments and I will print it, the same way I just printed my own mistake.

If you don't, then here is looking at you. We are watching.

I'm a father. Protect the children.

* * *

If you or someone you love is struggling, call or text 988 in the US to reach the Suicide and Crisis Lifeline.

[Footer line removed.]

Receipts: [link to the process record once pushed]

Where the essay and the record differ

Five things were wrong, or not yet right, between the LinkedIn post and the draft. Reopening the sources caught the first three. The fourth was the machine's own mislabeling, caught while building this record. The fifth was a trap avoided.

  1. "Your line." The sentence "Disclosure rules and break reminders. That is what a dead kid buys in this country." was written by the machine (October 14, turn 15). The outline and the October 21 reply both called it Gage's. The essay opens "I wrote that line last week," in Gage's voice, under the standing note on all his pieces that Claude writes the words. This record is where you can see whose words they were.
  2. The line itself. It said the law bought disclosure rules and break reminders. For the laws in force today, that holds. For Adam's Law it was wrong, and the machine had the sources that said so when it wrote it. The essay keeps the line, says where it was wrong, and turns on the date: most of the law starts in July 2027, the first audit is due in January 2029.
  3. Dates and counts from the LinkedIn post and the outline. "In effect January 1, 2027" for Adam's Law became July 1, 2027, for most of it (sources disagree; see above). "About a dozen states" became twelve, counted. "CEO to Disney, September 2026" became "this fall," because sources differ by a few days. The toy-ban date for SB 867 could not be confirmed, and the bill was cut.
  4. Disney and the chief executive. The LinkedIn post said Disney "picks up the IP" and that the chief executive was "hired to fix its safety reputation." Neither held up. Disney is bidding for a license. The essay says only that he became Disney's first chief technology officer. "A Drone With Hands" made the same correction a week earlier.
  5. The other GUARD Act. A search returned a House vote of 414 to 7 on a bill called the GUARD Act. It is a different bill. Had it gone in, the essay would have said the House passed the chatbot ban. It has not.

Gage's words and Claude's words

Gage's, verbatim or nearly: The question and the demand, from October 14, turn 14: what changed after the suicides, "I could be wrong, and if I am, show me," "Am I wrong? Someone fucking tell me," "here's looking at you. We're fucking watching" (the essay keeps "fucking" in the demand and drops it from "We are watching"). "I'm a father," his edit to the LinkedIn post. "Protect the children," his motto for this work. The title's "You." The commission ("it's time to spread this word"). His approval of the draft as written.

Gage's ideas, Claude's words: The three-party covenant paragraph (his design from October 14; the machine's phrasing, carried from the LinkedIn post). "Show me the person who faced consequences. Show me the law with teeth." (his "show me," sharpened). The status line on E.D.A.I. and the covenant ("one sealed record," "a design on paper," "overclaiming is the disease I am trying to cure"), which is his standing rule. The buyer-pays sentence, which states his ruling of October 13 to 14. "I work on proof of personhood. That fear is not paranoia to me. It is the design problem." (his Proof of Life work, his worry about surveillance, the machine's sentences.)

Claude's alone: "Disclosure rules and break reminders. That is what a dead kid buys in this country." The seed of it, "a dead kid with a name," on October 14, turn 1. The working title. The empty chair, carried from the first essay, where it was also the machine's. The decision to split the demand into its own piece (Gage said go). The research that turned the thesis, and the turn itself: "I was half wrong. The half I got wrong is worse." "That is what a dead kid buys you. Not nothing. A deadline." "That is a legislature saying the machine's side of the story has to survive." "The law reaches for it after the body. I want it sworn before." "California just ordered a market into being." Naming Sewell and Adam rather than "a kid." The choice not to retell what the chats said. The 988 line. All four subtitles.

Removed, by category

Status

Essay published October 21, 2026, on Substack under Gage's own name: https://gagecasswoodle.substack.com/p/what-a-dead-kid-buys-you. This record was published the same day, before the essay went live.

The published essay is draft one with two changes, both Gage's rulings. The subtitle is "Here's the receipt. Nobody's name is on it." It was the first of a second set of four the machine offered, after he turned down the first set. The footer is the 988 line followed by the receipts link; his standing disclosure runs in Substack's "How I make this" box, as on his other pieces, not in the body.

The header image was made by Gage in an image generator from prompts the machine wrote: a black box printing the receipt and feeding it back into itself, the signature line empty. Every figure printed on it comes from the sources above. The machine checked each row of each version for spelling and alignment before he chose one. Two earlier versions were set aside because the labels and values drifted out of line.

Anchored on Hedera October 29, 2026: topic 0.0.10897644, message #9, consensus 17:13:29 UTC, transaction 0.0.9083680@1793293964.295832013, fingerprinting both files as they stood at commit e8ad4ed, where the fingerprint had been prepared on October 21 and then left unposted. This sentence was written after the anchor, so the file you are reading differs from the anchored version by it; the README holds the hashes.