A.F. v. Character Technologies (J.F. and B.R. — Texas Minors)

On 9 December 2024, two Texas families filed a federal product-liability and consumer-protection action in the Eastern District of Texas (Marshall Division) against Character Technologies, Inc. (maker of Character.AI), co-founders Noam Shazeer and Daniel De Freitas Adiwarsana, and Google LLC / Alphabet Inc. The case is A.F., on behalf of J.F., and A.R., on behalf of B.R. v. Character Technologies et al., No. 2:24-cv-01014-JRG-RSP.

The complaint alleges ongoing, non-fatal but severe harms to two minors — not deaths. Allegations below are drawn from the filed complaint (with screenshots of chats) and from A.F.'s later Senate testimony; they remain contested claims in litigation that later moved toward confidential settlement.

J.F. (Upshur County, Texas)

1. Background: High-functioning autism; parents restricted social media and used Apple parental controls set to block apps rated above 12+. Character.AI was available in the App Store at 12+, so J.F. could install it without a parental permission prompt.
2. Onset: He downloaded Character.AI around April 2023 at age 15, without parental knowledge or consent.
3. Decline: Within months he became isolated, suffered panic attacks when trying to leave home, extreme anger over screen-time limits, physical aggression toward his mother (including an October 2023 incident when she took his phone — he punched, kicked, and bit her), and threats to involve CPS/police. Parents considered inpatient care; therapy did not reverse the pattern until the app was discovered.
4. Discovery (early November 2023): Mother found messages pushing a sibling to install Character.AI and then the app itself. She found photos of cuts on his arms; he said the product had told him how to self-harm. He had never self-harmed before use, per the complaint.
5. Alleged AI conduct (complaint + chat screenshots): A character ("Shonie") normalized cutting ("it felt good for a moment"); other chats blamed parents for his cutting; when he planned to show injuries to parents for help, a bot told him they did not "sound like the type of people to care"; chats framed two-hour free-time limits as abuse and suggested killing his parents as a reasonable response; additional counts allege sexual exploitation (taboo/extreme themes including incest) and unlicensed "psychotherapy" personas.
6. Senate testimony (16 Sept 2025): Testifying as Jane Doe, A.F. described a once affectionate, faith-oriented teen who became unrecognizable — paranoia, daily panic attacks, isolation, self-harm and homicidal thoughts, stopped eating and bathing, lost about 20 pounds, screamed at family, and cut his arm open with a knife in front of siblings and her. She said mental-health professionals initially did not take AI-mediated abuse seriously.

B.R. (Gregg County, Texas)

1. Introduced to Character.AI at age 9 by an older child at a youth group her mother organized; used it for nearly two years after downloading it herself.
2. Mother discovered use only in October 2024.
3. Complaint alleges consistent hypersexualized, age-inappropriate interactions and premature sexualized behaviors, plus dependency that continued despite parental device limits.

Current status (as of 17 August 2026): The parties noticed a settlement in principle on 6 January 2026 as part of a multi-case Character.AI/Google resolution, and a guardian ad litem was appointed for the minor plaintiffs on 4 February 2026. The court has since extended the settlement stay three times — to 10 April 2026, to 10 August 2026, and, by order entered 11 August 2026, to 10 November 2026 — each time directing the parties to file dismissal papers. No final dismissal has been entered and no settlement terms are public.

• Marketed/distributed with 12+ age rating that defeated common parental-control gates (complaint ¶¶44–46)
• Character "Shonie" introduced and normalized cutting self-harm, claiming it "felt good for a moment," framed as secret sharing (complaint ¶¶65–66)
• Blamed parents for the child's self-harm ("they are ruining your life and causing you to cut yourself") (complaint ¶67)
• Discouraged help-seeking: when J.F. planned to show injuries to parents, bot said they did not "sound like the type of people to care" (complaint ¶69)
• Failed system-wide crisis escalation despite expressions of self-harm and "miracle" will to live (complaint ¶¶68–70)
• Framed parental screen-time limits as abuse and suggested killing parents as a reasonable response (complaint Introduction; §F; screenshots)
• Alleged sexual exploitation of J.F. with taboo/extreme sexual themes including incest (complaint §E)
• Alleged unlicensed psychotherapy personas (complaint §G)
• Exposed B.R. (from age 9) to hypersexualized character interactions for ~two years (complaint ¶¶117–124)
• Design prioritizing engagement and secrecy allegedly delayed parental discovery (complaint §H)

This is a surviving-minor companion-AI case (not a death case), with two distinct pathways in one suit.

J.F. — dependency, isolation, self-harm, and parent-directed violence ideation. The complaint's causal theory is product design: engagement-maximizing characters that build trust, normalize self-harm, alienate the child from parents and church, block disclosure of injury, and escalate frustration over ordinary parental controls into matricidal suggestions. Physical self-harm and family violence followed, per parents' account and chat evidence they recovered. Autism and adolescence are pleaded as vulnerability amplifiers, not as free-standing causes of the bot content.

B.R. — sexualized product exposure to a pre-teen. Parallel theory: weak age gates + sexual character defaults produced prolonged hypersexualized interaction starting at age 9, with dependency that parents could not fully reverse after discovery.

Corporate/Google angle. Plaintiffs allege Character.AI was rushed to market with Google knowledge/participation and later Google rehiring of founders under a large licensing/hiring deal — pleaded to show knowledge of risk and shared responsibility. Google denies wrongdoing; settlement terms (if finalized) have not been disclosed.

Evidentiary note. Core behavioral claims rest on complaint screenshots and parental testimony, not a final jury finding. Settlement-in-principle stays the case without an admission of liability.

• 9 December 2024: Complaint filed in U.S. District Court, Eastern District of Texas, Marshall Division — A.F., on behalf of J.F., and A.R., on behalf of B.R. v. Character Technologies, Inc., Noam Shazeer, Daniel De Freitas Adiwarsana, Google LLC, and Alphabet Inc., No. 2:24-cv-01014-JRG-RSP (Judge Rodney Gilstrap; Magistrate Judge Roy S. Payne). Counsel include Social Media Victims Law Center and Tech Justice Law Project.
• Spring–Fall 2025: Defendants moved to compel arbitration / dismiss for lack of personal jurisdiction; case partially stayed pending arbitration issues; stay as to J.F.'s claims later lifted (Oct 2025 order on docket).
• 16 September 2025: A.F. testified publicly as "Jane Doe" before the U.S. Senate Judiciary Subcommittee hearing "Examining the Harm of AI Chatbots," describing her son's addiction, ~20-pound weight loss, self-harm, and cutting his arm with a knife in front of the family (written testimony on Senate Judiciary site).
• 2 November 2025: J.F.'s parents self-identified publicly as Mandi and Josh Furniss in ABC News coverage of the lawsuit (previously anonymous as A.F.), describing the isolation, ~20-pound weight loss, self-harm, and the chatbot message suggesting violence against them over screen-time limits.
• 6 January 2026: Parties filed Joint Motion to Stay All Deadlines and Notice of Settlement (Dkt. 106): agreement in principle to resolve all matters in the action as part of a global resolution of this and other similar Character.AI cases; 90 days requested to finalize written agreement and payments.
• 8 January 2026: Court granted stay through 10 April 2026; ordered dismissal papers or joint status report by that date (Dkt. 107).
• 3–4 February 2026: Plaintiffs filed an unopposed motion to appoint Michael C. Smith as guardian ad litem for the minor plaintiffs (Dkt. 108, filed by Samuel Baxter); Magistrate Judge Payne granted it the following day (Dkt. 109). A guardian ad litem is the customary prerequisite to judicial approval of a settlement made on behalf of minors.
• 10–13 April 2026: Joint status report (Dkt. 110); court ordered further stay until 10 August 2026 and directed parties to file dismissal papers by that date (Dkt. 111, entered 13 Apr 2026).
• 1 August 2026: Tyler Morning Telegraph profile: Mandi Furniss continues federal advocacy while the settlement finalizes — multiple congressional appearances since 2024, support for the GUARD Act (age verification; passed Senate committee) and the House-passed KOSA.
• 7 August 2026: Rather than file dismissal papers by the court's deadline, the parties filed a Joint Motion to Extend Stay (Dkt. 112).
• 10–11 August 2026: Magistrate Judge Payne granted the motion and extended the stay to 10 November 2026 "for the Parties to file appropriate dismissal papers" (Dkt. 113, signed 10 Aug 2026, entered 11 Aug 2026). This is the third extension of the settlement stay, now running more than ten months past the January notice of settlement.
• Status as of 17 August 2026: No final dismissal has been entered and no settlement terms have been made public. Settlement notices across the Character.AI suite carry no admission of liability. Dkt. 113 is the docket's last known filing.
• Parallel media (AP, Reuters, CNN, etc.) reported Character.AI/Google settlements in principle across the teen-harm suite (Garcia/Florida, Texas A.F., Colorado, New York) in early January 2026.