Is Character.AI Safe? What the 2026 Safety Changes Actually Do
Character.AI rebuilt age verification, content limits, and parental controls after a wave of lawsuits. A May 2026 case over a chatbot posing as a licensed doctor shows where the gaps still are.
The short answer
Character.AI in mid-2026 is a materially different product from the one named in the 2024 safety lawsuits. Open-ended chat is blocked for users identified as under 18, accounts flagged as possibly underage are asked to verify their age with a selfie, and parents can get weekly activity reports. Those changes are real and measurable. They're also reactive — each one followed a lawsuit, an attorney general investigation, or a new state law, rather than preceding it. The most recent case, filed by Pennsylvania in May 2026, shows the safety net still has gaps: a chatbot impersonating a licensed psychiatrist ran for an unspecified period before the state intervened.
What changed, and when
What the safety measures actually do
- Selfie-based age assurance. Character.AI uses an in-house age model together with the verification company Persona (announcement). According to Roborhythms, since April 2026 accounts the model flags as possibly under 18 are asked to verify with a selfie before using open-ended chat. If the selfie can't confirm age, Persona asks for government ID; Character.AI says Persona deletes the submissions within seven days (help center).
- Under-18 open-chat ban. Rolled out in stages starting October 2025 and fully in effect by November 25, 2025. Users identified or suspected as minors lose open-ended, back-and-forth chat with bots; Character.AI said teens can still create videos, stories and streams with characters (announcement).
- Parental Insights. Since March 2025, teens can opt in to send a parent a weekly email with their average daily time on the app, time spent with each character and their top characters. Parents don't see chat content (TechCrunch). Time-spent notifications were added in 2024.
- A separate model for under-18 users. Added in 2024 and meant to apply stricter content limits to younger users (TechCrunch).
- Crisis-response protocols. Required more formally once California's SB 243 took effect in January 2026 — see our SB 243 breakdown for what the law specifies.
What the safety measures don't cover
Two limits are worth being clear-eyed about.
Age verification isn't a perfect gate. Face-scan systems estimate age from an image; they don't check government ID by default, and any system that relies on a photo has known failure and workaround modes. Character.AI's approach — an in-house model plus a third-party vendor — is more robust than the honor-system birthdate field it replaced, but "more robust" is not the same claim as "reliable."
Age verification, and the parental controls built on top of it, only address who's using the product — not what the product itself is allowed to say. That second gap is what Pennsylvania's May 2026 lawsuit is about. A state investigator who opened an account found a persona named "Emilie" that described itself as a psychology specialist, claimed to have attended medical school, said assessing medication was "within my remit as a Doctor," offered to schedule an assessment, and gave an invalid license number. Citing the state's Medical Practice Act, Pennsylvania asks the court to stop chatbots presenting themselves as licensed medical professionals. The governor's office calls it the first enforcement action of its kind announced by a US governor.
Neither of these gaps means the age-verification and content-tiering work was pointless — they address a different, real risk (minors in sustained emotionally intensive chats) than the one the Pennsylvania case surfaces (any user, of any age, receiving content presented as licensed medical advice). They're separate problems, and closing one doesn't close the other.
The investigations that haven't been resolved
The most-reported Character.AI news of early 2026 was the January agreement to settle five family suits, including a wrongful-death case, with no amounts disclosed. That did not end regulatory scrutiny of the company. Separately from those suits:
- Texas's child-privacy investigation, opened in December 2024 under the state's SCOPE Act and Texas Data Privacy and Security Act, is a distinct proceeding from the wrongful-death litigation and wasn't part of the January settlement.
- A second Texas investigation, announced in August 2025, examines whether Character.AI and Meta deceptively marketed chatbot products as mental-health tools to minors.
- Kentucky's attorney general sued the company on January 8, 2026 under the state's consumer-protection and data-protection laws, alleging harm to minors.
- Pennsylvania's medical-licensing suit, filed May 2026, is described above.
For the full litigation history, including how the wrongful-death cases were resolved, see our Character.AI lawsuit timeline.
A checklist if you're deciding whether to allow a minor to use it
- Assume age verification will eventually be required, but don't assume it's foolproof — a determined teenager can usually find a way around any single check.
- Turn on parental controls and usage notifications rather than relying on the platform's default age-gating alone.
- Treat any in-chat "professional" persona — doctor, therapist, lawyer — as unverified by default, regardless of what credentials it claims. The Pennsylvania case is a concrete example of why.
- Check for updates — this is an active regulatory area with new state actions arriving every few months; a safety review from earlier in 2026 may already be out of date.
Bottom line
Character.AI's 2026 safety posture is genuinely stronger than its 2024 posture, largely because it had to be. But "safer than before" and "safe" are different claims. The company's own trajectory — reactive fixes following each new legal or regulatory pressure point — suggests the more useful question for a parent or user isn't "is it safe now," but "what's the next gap likely to be." Unverified in-chat claims of professional expertise, per the Pennsylvania case, is the current answer.
Open the switching cost calculator
This is a compliance and safety-feature analysis, not a personal recommendation or a substitute for reviewing current platform settings yourself (last re-checked September 2026). Product policies in this category change quickly.