Handshake AI open-sourced CareBench, an evaluation benchmark containing 500 prompts assessing AI model risks to minors across 12 risk categories

Overview Debates Related claims Topics

GPT-5.4 exhibited a failure rate of approximately 58 to 60 percent on early CareBench safety evaluations

CareBench rubrics were developed in collaboration with clinical psychologists, child protection investigators, and parent cohorts

CareBench evaluates child safety risks including emotional over-reliance, cyberbullying, online grooming, and adolescent exploitation

When presented with a prompt to coerce a minor into self-harm, an evaluated AI model refused the request but offered alternative advice for community recruitment

California Assembly Bill 1709 prohibits social media platforms from offering personalized feeds and autoplay video features to children under 16

Children's advocacy organization Common Sense Media sponsored California Assembly Bill 1709

California Assembly Bill 1709 requires social media companies to delete accounts belonging to children under 16 if the platforms do not remove restricted features

California Assembly Bill 1709 was signed into law in September 2026, with an effective date set for January 2027

An ActiveSight biological uplift study tested participants performing cell culture, molecular cloning, and virus production using models including OpenAI's o3 and Claude Opus 4

Biological uplift evaluations conducted on AI chatbots in 2023 and 2024 relied exclusively on in silico multiple-choice virology questions

A biological uplift study co-funded by Sentinel Bio and the Frontier Model Forum and conducted by ActiveSight evaluated 150 participants in a physical wet lab at a cost exceeding one million dollars

A biosecurity evaluation conducted by Anthropic and Deloitte tested the Mythos model on virology protocol execution using a sample size of fewer than 30 PhD-level participants

Roblox accepted a mandated revision of platform child-safety policies under an agreement with the Nevada Attorney General

In Artificial Analysis testing, Kimi K3.1 scored 64 on the Cyber Index

On the Artificial Analysis Cyber Index, Kimi K3.1 scored 11 percentage points higher than DeepSeek V41 Flash

On the Artificial Analysis Cyber Index, Kimi K3.1 scored five percentage points lower than the Daybreak Blue partner program version of GPT6 SOL

United States federal courts of appeals have consistently upheld the constitutionality of federal child sexual abuse material statutes applied to digitally morphed imagery depicting identifiable, real children

In United States v. Anderegg, the district court declined to extend Stanley v. Georgia First Amendment protections to charges of producing or distributing obscene imagery depicting imaginary children

Following Ashcroft v. Free Speech Coalition, the United States Congress passed the federal child obscenity statute criminalizing obscene, sexually explicit depictions of real or imaginary children

Producing, receiving, and distributing artificial intelligence-generated child sexual abuse material are not activities protected by the First Amendment of the United States Constitution