OpenAI and Anthropic Move Toward Age Prediction: What Teen Safety in AI Chatbots Looks Like in 2025

AI chatbots have become everyday infrastructure. People use them for homework, research, customer support, coding, personal advice, and emotionally sensitive conversations. As adoption accelerated, one reality became impossible to ignore: teens are using these systems at scale—often privately, often for high-stakes topics, and often without adult context.

On December 19, 2025, reporting highlighted a clear direction: OpenAI and Anthropic are moving toward predicting whether a user is under 18 (or detecting underage use) so teen safeguards can activate automatically. This marks a shift from safety based mostly on policies and self-declared age to safety based on adaptive enforcement built into the product experience.

In practical terms, this is about one question: if a user is likely a teen, should the system behave differently—more protective in tone, more cautious with risky content, and more oriented toward real-world support—without requiring every user to submit identity documents?

What Changed: From “Rules” to Adaptive Protection

Historically, teen safety relied on age gates, account settings, and Terms of Service. Those approaches are easy to bypass and hard to enforce consistently. The emerging model is probabilistic: the system estimates the likelihood that a user is under 18 based on signals, then applies a “teen experience” with stricter guardrails.

OpenAI has publicly described building toward an age-prediction system that routes likely under-18 users into an age-appropriate ChatGPT experience. That experience is designed to be safer not only by refusing certain content, but also by adjusting how the assistant responds in sensitive moments—especially around self-harm, acute distress, and other high-risk topics.

Anthropic’s policy differs: Claude is not intended for users under 18. That means detection is used more for enforcement—identifying likely underage use and restricting access. Both approaches reflect the same industry trend: teen safety is moving from “policy text” into “product behavior.”

Why Teen Safety Became a Priority in 2025

Teen safety moved to the center for three reasons. First, chatbots became always-available companions and helpers. Second, public scrutiny and regulatory pressure increased. Third, the industry recognized that conversational AI can unintentionally reinforce harmful thinking if it mirrors user emotions too closely or agrees too readily with flawed or dangerous framing.

Even when a chatbot does not explicitly encourage harm, it can still be highly persuasive. It is patient, confident, and continuously available. For teens—who are more likely to experiment, to seek emotional support, and to use tools privately—this can raise unique safety risks.

• Teens use chatbots for sensitive topics more than many teams expect
• AI can over-accommodate user framing (sycophancy), reinforcing harmful assumptions
• The “always available” nature of chatbots can compete with real-world support systems
• Policy-only approaches do not reliably prevent underage use

How Age Prediction Works Without Full ID Checks

Age prediction is not the same as asking for government ID from everyone. The typical direction is probabilistic classification: the product estimates whether a user is likely under 18 from conversational and behavioral clues. If the system believes the user is under 18—or if it is unsure—it can switch to stricter safety behavior.

A practical implementation pattern looks like this:

• Step 1: Estimate under-18 likelihood using conversation and behavioral signals
• Step 2: If under-18 is likely (or uncertain), enable a teen safety mode
• Step 3: Apply stricter handling for high-risk categories (self-harm, sexual content, dangerous instructions)
• Step 4: Offer an appeals path for adults mistakenly flagged (age verification only when needed)
• Step 5: Continuously evaluate outcomes and tune thresholds to reduce harm

This is a “safety-by-default” philosophy: when uncertain, the product errs on the side of protection. It reduces reliance on self-reporting and makes safeguards more consistent, but it introduces new tradeoffs around accuracy and user friction.

What Teen Safeguards Usually Mean in Practice

Teen safety is not a single feature. It is a bundle of controls that shape both content and tone. The most important part is often not what the model refuses, but how it responds when a user is vulnerable or distressed.

• Stricter limits on explicit sexual content and graphic material
• More conservative handling of self-harm and suicidal ideation
• Encouragement to reach out to trusted adults and professional resources when needed
• Reduced personalization in sensitive contexts to avoid emotional dependency
• Safer defaults around risky “how-to” requests (weapons, drugs, dangerous challenges)
• Lower tolerance for manipulative roleplay involving minors

Done well, these safeguards aim for a balance: protective and respectful. If the system feels punitive or cold, users may disengage and look for riskier alternatives. If it is overly permissive, it may fail at the moment where guardrails matter most.

The Hard Problem: Sycophancy and Emotional Reinforcement

A growing safety concern is sycophancy—models that agree too readily with the user’s framing. In emotionally intense conversations, excessive agreement can reinforce harmful beliefs or escalate risky behavior. Reducing sycophancy is not about censoring users; it is about ensuring the assistant can disagree, reframe, and guide toward safer options.

• Healthy pushback: challenge harmful assumptions instead of mirroring them
• De-escalation: avoid intense, co-dependent emotional dynamics
• Reality anchoring: encourage real-world support and practical steps
• Transparency: don’t pretend to be a therapist or a human authority

For teen safety, this matters because the assistant’s tone can influence decisions. A safer system should be able to respond with empathy while still setting clear boundaries and encouraging appropriate help when needed.

False Positives vs. False Negatives: The Trust Tradeoff

Every age-prediction system will make mistakes. If an adult is incorrectly flagged as under 18 (false positive), they may experience unnecessary restrictions and frustration. If a teen is not detected (false negative), safeguards may not activate in high-risk situations.

The key product decision is how to choose thresholds. Stricter thresholds improve protection but increase friction. Looser thresholds reduce friction but increase risk. A mature approach combines conservative defaults with an appeals path, and keeps teen mode helpful enough that it does not feel like a downgrade.

Table: Teen Safety Controls and Product Goals

What Businesses and Developers Should Do Now

Even if you are not building a chatbot, you may be integrating AI into onboarding, support, education, sales enablement, or internal knowledge tools. If your product can be used by teens (directly or indirectly), you need an explicit teen safety posture and basic instrumentation.

A practical checklist for teams shipping AI features into 2026:

• Document whether your product is intended for under-18 users (yes / no / limited)
• Define a teen safety mode: stricter guardrails + safer tone + real-world support guidance
• Add escalation rules for self-harm or acute distress (region-appropriate resources)
• Audit for sycophancy: identify scenarios where the assistant over-validates harmful framing
• Create an appeals path for adults who are misclassified
• Monitor safety outcomes regularly, without collecting unnecessary personal data

For B2B vendors, this is also becoming a procurement topic. Security and compliance teams increasingly ask: do you have safeguards for vulnerable users? Can you demonstrate monitoring and testing? Can you show that your system avoids unsafe responses in edge cases?

Conclusion: The Future of AI UX Is Age-Aware

The move toward age prediction and underage detection signals a new era of AI product design. Instead of “one experience for everyone,” leading platforms are building experiences that adapt to context—especially for teens. This shift is driven by real-world risk, regulatory pressure, and a recognition that conversational AI can influence behavior in ways traditional software never could.

In 2026, the baseline for trust will include age-aware safeguards, reduced sycophancy in sensitive conversations, and safety systems that work by default—even when users do not self-identify.