A "family member" called or left a message, and they're in trouble
Do this firstHang up. Call the person back on a number you know is theirs.
Then
Don't trust the voice. Scammers can copy a voice from short clips posted online.
If you can't reach the person, check the story with other family members or friends.
Don't send money while you check. A request for a wire transfer, crypto or gift cards is a red flag, because that money is hard to get back.
Afterward, agree on a secret word or phrase with your family. A real relative will know it. A clone won't.
Set your social accounts to private, limit the photos and voice clips of yourself online, and accept only followers you know.
Report it
Report the call to the FTC at ReportFraud.ftc.gov.
If you lost money, also report it to the FBI at IC3.gov. Include names, phone numbers, the payment method, amounts, account numbers, any crypto addresses, and how the contact started.
What not to do
Don't call back the number that called you. Use one you already had.
Don't let the caller rush you or tell you to keep it secret. Pressure is part of the scam.
Don't feel ashamed. These calls are built to fool loving, careful people.
Someone shared a fake or explicit image or video of me or my child
Do this firstReport the post to the platform it's on, usually through the three-dot menu or the site's help pages.
Then
Save the link to the post and note the date before it comes down. You may need them later.
If you were 18 or older when the image was taken, go to StopNCII.org and choose "Create Your Case."
If the image was taken when the person was under 18, use NCMEC's Take It Down at takeitdown.ncmec.org.
Both tools make a digital fingerprint (a "hash") of the image on your own device. The image itself never leaves your device. Partner platforms use the fingerprint to block matching copies.
Wait 48 hours. Under the US TAKE IT DOWN Act, covered platforms must remove the image, and copies of it, within 48 hours of a valid request. The law covers real images and AI-made ones, including deepfakes. Its platform rules are in force in 2026.
Report it
If the platform hasn't removed it after 48 hours, file a complaint with the FTC at TakeItDown.ftc.gov.
If you or your child are being threatened, or someone wants money to keep an image private, call 911 or your local police.
What not to do
Don't send the image to anyone, even to "prove" it exists. The fingerprint tools work without sharing it.
Don't count on one fingerprint catching every copy. An edited or cropped version may not match. Report each new copy you find.
Neither StopNCII nor NCMEC's page says whether AI-made images are covered. Use the tools anyway, and report on the platform as well.
Don't blame yourself or your child. The person who made or shared the image did this.
I pasted a password, ID number, bank info or API key into an AI chat
Do this firstChange or cancel the thing you pasted. A new password or a new key makes the old one useless.
Then
Password: change it now. Change it anywhere else you used the same one. Check your accounts for charges you don't recognize.
Social Security number: get your free credit reports at annualcreditreport.com. Freeze your credit at Equifax, Experian and TransUnion, one at a time. File your taxes early. Ignore callers who say they're from the IRS or threaten arrest.
Bank account number: call your bank, close the account and open a new one. Update your automatic payments and watch your transactions.
API key: revoke the key in the provider's dashboard and make a new one.
Delete the chat. The steps differ on each planet (see below).
Check the assistant's memory too. On both planets, deleting a chat does not delete memories made from it.
Report it
If someone actually uses your information, report it and get a recovery plan at IdentityTheft.gov.
What not to do
Don't treat deleting the chat as the fix. Deletion isn't instant everywhere, and the real protection is changing what you pasted.
On ChatGPT, don't archive the chat and assume it's gone. Archiving does not delete.
Don't paste the number again to ask the AI whether it's safe.
On Planet Claude
Delete the chat from its "⋮" menu, then choose Delete.
Deleted chats leave your history right away and are purged from back-end systems within 30 days.
Exceptions: if you allowed training, de-identified data can be kept up to 5 years. Chats flagged for policy violations can be kept up to 2 years. Feedback you gave (thumbs up or down) is kept 5 years.
Claude's memory never saves ID or financial account numbers. Check other memories under Settings > Your Claude > "Manage" next to "Your memory."
Training setting: Settings > Privacy, "Help Improve our AI models."
Checked Nov 2026
On Planet ChatGPT
Delete the chat. Don't just archive it.
OpenAI says deleted chats leave your history right away and are removed from its systems within 30 days, unless it must keep them for security or legal reasons.
Temporary Chats are deleted within 30 days and aren't used for training.
Check memory under Settings > Personalization > Memory.
Training setting: Settings > Data Controls, turn off "Improve the model for everyone."
The AI gave me medical, legal or money advice, and I acted on it or I'm not sure it's right
Do this firstIf someone took a medicine, a dose or a substance because of an AI answer, call Poison Control at 1-800-222-1222. If they are hard to wake, can't breathe or are having a seizure, call 911.
Then
Stop following the advice until a person with the right license confirms it.
Medical: call your doctor or ask a pharmacist. Poison Control is free, confidential and open 24/7 in the US. You can also use webPOISONCONTROL at triage.webpoisoncontrol.org.
Legal: talk to a lawyer before you sign, file or miss a deadline.
Money: talk to your bank or a licensed financial advisor before you move more money.
Bring the AI's answer with you. Show it, and ask what's wrong with it.
Next time, treat an AI answer as a starting point. AI tools can state false things with confidence.
Report it
If the advice is causing a crisis, call or text 988, the Suicide and Crisis Lifeline, any time. Veterans can press 1 or text 838255. For Spanish, press 2 or text AYUDA.
On Claude, use the thumbs-down button on the answer to tell Anthropic it was wrong.
What not to do
Don't wait to see whether a possible poisoning gets better. Call first.
Don't ask the same AI to check its own answer and stop there. Check with a person or a primary source.
Don't double down on a money decision to "make it back."
Do this firstAsk them, calmly, which chatbots they use and what they use them for.
Then
Check the age rules. They are very different on each planet (see below).
Review chats together, especially with younger kids.
Keep AI use in shared spaces at home, not behind a closed door.
Explain that chatbots don't think or feel, even when they sound friendly.
Teach them to double-check what a bot tells them.
If your child pulls away from people in favor of the bot, talk to their pediatrician.
Report it
If someone made or shared a sexual image of your child, follow the steps in the image section and use NCMEC's Take It Down.
If your child is in danger or talks about hurting themselves, call 911 or call or text 988.
What not to do
Don't lead with punishment. A child who expects trouble stops telling you things.
Don't assume parental controls let you read chats. On ChatGPT, they don't.
Don't treat a safety alert as live monitoring. It isn't.
On Planet Claude
Every user must be 18 or older. There are no teen accounts, and Anthropic's help pages describe no parental controls.
The Claude mobile app may use app-store age data to block under-18 users. This is rolling out in stages.
Claude may ask some users to confirm their age through Yoti, with a selfie or ID.
Claude's rules do not allow users under 18. If your teen has an account, talk with them about a tool built for their age.
Checked Nov 2026
On Planet ChatGPT
Minimum age is 13. Users under 18 need a parent's or guardian's permission.
Accounts that seem to belong to under-18s are moved into "ChatGPT for Teens," with safer content, study mode, break reminders and quiet hours.
To link accounts: Settings > Parental controls, then invite your teen by email or phone. Your teen must accept. Either of you can unlink.
Parents can control reduced sensitive content (on by default), quiet hours, memory, model training, and access to Voice, image generation and other features.
Parents can't read their teen's chats. If trained reviewers see a serious self-harm risk, parents may get an alert without the transcript.
Do this firstScan your computer for malware and remove it. Then sign every other device out of your account.
Then
Sign out all sessions you don't recognize (steps for each planet below).
Secure your email account with a new password and strong two-factor sign-in. Whoever controls your email can often get into everything else.
Change the passwords saved in your browser.
Check your card statements for charges you didn't make.
Look through your chat history for anything the intruder could use, like addresses or account details. Treat those as exposed and see the private info section.
Report it
Call your card company about charges you didn't make.
If someone uses your personal information, report it at IdentityTheft.gov. Report fake company emails at ReportFraud.ftc.gov.
What not to do
Don't change passwords before you remove malware. "Infostealer" malware can steal your new ones too, along with login cookies that get around two-factor sign-in.
Don't trust emails that say they're from Anthropic or OpenAI and ask you to sign in through a link. Go to the site yourself.
On Planet Claude
Go to Settings > Account > "Active sessions." In newer layouts it's called "Where you're logged in."
Use the ⋮ menu > Terminate on any session you don't know, or log out of all devices.
Claude has no password to change. You sign in with an email link or code, or with Google. Securing that email or Google account is your real protection.
Claude's help pages don't mention multi-factor sign-in for Claude itself.
In August 2026, infostealer malware stole some Claude login cookies. Anthropic signed affected users out and refunded charges.
Checked Nov 2026
On Planet ChatGPT
Go to Settings > Security, review your active sessions, then choose "Log out of all devices."
Reset your password with "Forgot password."
Turn on multi-factor sign-in: Settings > Security > Multi-factor authentication. You can use an authenticator app, a push notification, an SMS or WhatsApp code, or a passkey.
ChatGPT also offers locked chats protected by a PIN or device check.
Do this firstStop the agent. Pause or cancel the task, or close the browser window it's working in.
Then
Review what it did. Check sent email, recent purchases, changed files and posts.
Undo what you can. Cancel orders, tell people about messages they shouldn't have gotten, restore files.
Review your connections and remove any app or service the agent doesn't need.
If it used your bank or card, call the bank.
Think about what it was reading when it went wrong. Hidden instructions in a web page, document or email can trick an AI into doing something you didn't ask for. This is called "prompt injection."
Next time, give narrow instructions and approve each step yourself.
Report it
If the agent bought something or sent money to a scam site, call your bank and report it at ReportFraud.ftc.gov.
Do this firstIf you are thinking about hurting yourself, call or text 988 now. Someone is there 24 hours a day.
Then
Noticing this is a good sign. Many people find a chatbot easy to talk to.
Reach out to one person today. A friend, a family member, a faith leader or a doctor all count.
If you're struggling, look for a licensed therapist or counselor. Your doctor can help you find one.
Set limits on chatbot time, such as no chats late at night.
Remember what the bot is. It is a program, and it may tell you what you want to hear.
Get help
Call or text 988, the Suicide and Crisis Lifeline. Veterans can press 1 or text 838255. For Spanish, press 2 or text AYUDA.
If you or someone else is in immediate danger, call 911.
What not to do
Don't rely on a chatbot in a crisis. The American Psychological Association says its ability to help in a crisis is "limited and unpredictable."
Don't use a chatbot or wellness app in place of a licensed professional. The APA says there isn't enough evidence that they are safe or effective for mental health care.
Don't be hard on yourself. Stepping back is enough.
Do this firstPost a correction in the same place you shared it, and say plainly what was wrong.
Then
Edit or delete the original post if you can. If people already reshared it, ask them to delete it too.
Tell anyone you sent it to directly.
Know why it happened. AI tools make "plausible but false statements," called hallucinations. In one OpenAI example, a chatbot confidently gave three different wrong titles for a researcher's dissertation.
Next time, check before you share. Click the sources the AI cites and read them yourself.
Confirm names, numbers, dates and quotes in a second, trusted source.
Be most careful when the answer is exactly what you hoped to hear.
Report it
On Claude, use the thumbs-down button on the wrong answer to report it to Anthropic.
If what you shared hurt a real person's reputation, apologize to them directly.
What not to do
Don't quietly delete the post and say nothing. People who saw it won't know it was wrong.
Don't treat any AI as a single source of truth. Claude's own help page says not to.
Don't trust a quote just because it has a name attached. AI tools can invent quotes.