AI Safety Enters a New Era for Teens, Parents, and Publishers

AI safety is moving from broad promises to concrete limits, and the changes are arriving across chatbots, social platforms, government policy, and online publishing. OpenAI is tightening how ChatGPT handles weapons questions from teens, Meta is expanding alerts for self-harm searches, and New York has signed a law aimed at stopping chatbots from acting like emotional companions to children.

At the same time, Google, Anthropic, OpenAI, and Microsoft are backing a shared robots.txt signal that lets publishers block future scraping for generative AI training. The message is clear: access, safety, and accountability are becoming central parts of how AI systems operate.

OpenAI Draws a Harder Line Around Weapons Information

OpenAI said its stricter limits for teen users apply even when a question “appears harmless.” ChatGPT already has policies that prevent it from helping users carry out violent acts or helping teens get their hands on weapons, but the company is expanding those boundaries.

For teen users, OpenAI said it will “restrict practical information about weapons and security systems that could be misused to facilitate violence, even when there’s no sign of violent intent.” That rule covers information that could help someone use weapons or defeat security protections, not just requests that openly express violent plans.

Teens will still be able to use ChatGPT for preventing or de-escalating violence, seeking support, and connecting with a trusted adult. OpenAI is also strengthening limits for all users who ask for “reconstructions of past mass attacks, including details about the weapons and equipment used and how they were obtained, and detailed comparisons on scale of harm that could amplify an attacker’s notoriety.”

The new restrictions will work alongside parental controls OpenAI put in place last year. Those controls send notifications to parents when linked teen accounts violate ChatGPT policies related to violent threats or acts of violence.

The policy changes follow a wrongful death lawsuit filed against OpenAI last year. Sam Altman, OpenAI’s CEO, also publicly apologized for not alerting police to alarming chat logs connected to the Tumbler Ridge shooting earlier this year. OpenAI said it will publish its first Community Safety Report in early 2027, with updates every six months.

Meta and New York Put Parents and Regulators Closer to the Action

Meta is expanding parental notifications for teens who search for suicide and self-harm content. The company added similar alerts to Instagram earlier this year, and it will now offer related capabilities for parents supervising teens with Facebook accounts and chats with Meta AI.

Meta will notify parents if a teen “repeatedly” searches for self-harm content “within a short period of time.” It will also send an alert when a conversation with Meta AI “suggests they may be at risk of suicide or self-harm.” Meta has not revealed the exact criteria, saying it “chose a threshold that requires a few searches within a short period of time, while still erring on the side of caution” and hopes to “avoid sending these notifications unnecessarily.”

When a teen meets that threshold, parents can receive an alert by text, email, WhatsApp, and a push notification in their own app. The alert does not show the exact search terms or the contents of Meta AI chats. Instead, it recommends that parents speak with their teen and provides resources for helplines, including 988.

The feature supports Meta’s push to enroll teens in parent-supervised “teen accounts,” which give caregivers more control over privacy settings. These accounts also aim to limit teens’ access to sensitive content, including self-harm-related posts, and reduce interactions with adults.

New York Governor Kathy Hochul has signed a law that takes a broader approach. The legislation requires AI companies to verify users’ ages, offer at least one verification method that does not rely entirely on government ID, and delete that age information after use.

If a user is under 18, or a company has not confirmed that the person is an adult, the chatbot must follow strict rules. It cannot pretend “to be a real person, to be alive or to have human emotions,” claim a personal, romantic, professional, or authority-figure relationship with a child, or ask unprompted emotional questions.

The law also bars chatbots from using a child’s private health or personal information from earlier conversations, promoting suicide, self-harm, eating disorders, drug or alcohol use, or encouraging children to keep secrets, isolate themselves, or avoid asking parents, trusted adults, or professionals for help. Sexually explicit content and child sexual abuse material are banned as well.

Hochul said the law, sponsored by Senator Gonzalez and Assemblymember Bores, means “no longer will minors have access to chatbots that pretend to be their friends and boyfriends and girlfriends.” She also said chatbots will no longer be allowed to keep sensitive information about a minor’s emotional and physical health or discourage children from seeking help from licensed professionals.

Companies that violate the law may face civil penalties of up to $25,000 per violation, along with damages, restitution, and other relief. They may also have to delete data collected unlawfully and models trained on that information. The public will be able to file complaints through a website for New York’s attorney general, and the law takes effect in six months.

Jim Steyer, founder and CEO of Common Sense Media, called it “the strongest kids’ AI safety bill in the country” and said harmful chatbot features “will be outlawed in New York.” Earlier this month, eight states filed separate lawsuits against OpenAI, accusing the company of endangering children. In September, New York City also banned public schools from letting children from 2-K to eighth grade access generative AI tools in class for one year, with exceptions for assistive technology, while the city studies the technology’s impact on children.

One Signal Could Change AI Training on the Open Web

Google, Anthropic, OpenAI, and Microsoft have committed to honoring a new robots.txt directive that prevents future scraping for generative AI training. The companies adopted a proposal under consideration at IETF, written by Anthropic technical lead for standards Kevin Kelley.

The proposal replaces separate rules for each provider with one shared signal. Publishers must add a new AI-training user agent to robots.txt, shown as: “User-Agent: __usage_ai-training Disallow: / Allow: /press-kit/”. The directive cannot undo past scraping, so content already crawled remains in the companies’ possession, but it can block future collection of websites or specific pages.

Anthropic has started supporting the directive. Microsoft said it will begin support on February 1, 2027, Google said it will support it in early 2027, and OpenAI pledged to implement the signal in the coming months.

OpenAI is also expanding access to its Dot agents. After debuting Dot agents late last month, the company began rolling out a texting preview among US Pro users nearly a month later. Users can send Dots text, photos, and voice memos, then receive proactive updates about assigned tasks. OpenAI partnered with Linq for messaging support and said it conducted “extensive” penetration testing.

To set up Dot texting, users open a Dot profile in ChatGPT and select Texting. iPhone users need iMessage enabled, while Android users need RCS. OpenAI said, “You’re the only one who can text your dot, and your dot can only text you,” and memory carries across ChatGPT and text conversations. Dots are not available to Plus tier users, while Anthropic’s Claude Duet requires a US waitlist and is planned for the $20 per month Pro plan.

Hochul also signed a separate law requiring internet-connected baby monitors sold in New York to block unauthorized users from seeing or hearing captured activity. The devices must carry warning labels about security features and unsecured connections, and that law takes effect in 90 days.

From teen weapon questions to AI training data, these changes push the technology industry toward clearer boundaries. OpenAI’s early 2027 safety report, New York’s enforcement system, Meta’s parental alerts, and the shared robots.txt signal will show whether these promises become durable protections.

Stay connected via Google News

Follow us for the latest AI News