A teenager posts a perfectly innocent photo online.
asserted
teenager → post → photo
Before long, someone uses artificial intelligence to turn it into a sexually explicit fake and shares it with classmates.
asserted
someone → use → classmates
It's disturbing to think that something as simple as sharing a photo could lead to this kind of abuse.
uncertain
something → think → abuse
And for the child involved, the damage can last long after the fake picture gets taken down.
asserted
picture → involve → child
Now OpenAI is stepping up its efforts to tackle the problem.
asserted
OpenAI → step → problem
The company is introducing a new AI tool designed to detect suspected child sexual abuse material, including content created with artificial intelligence.
asserted
company → introduce → intelligence
It's also cracking down on attempts to find abusive material through ChatGPT, with new warnings and 24-hour suspensions.
asserted
It → crack → warnings
Here's a closer look at what's changing and what parents need to know.
asserted
parents → change → what
Our free CyberGuy LIVE class, 5 Ways to Travel Smarter and Save Money, has ended, but you can still watch the full replay.
asserted
you → travel → replay
Kurt "CyberGuy" Knutsson walks you through five practical ways to protect your phone and personal information, avoid common travel scams, find better deals, save money on bookings and use technology to make your travels easier and less stressful.
asserted
travels → walk → technology
No technical experience is needed.
asserted
experience → need → ?
How OpenAI's new child safety tool works
Finding child sexual abuse material online has become more complicated as AI-generated content spreads.
asserted
content → work → material
Many existing detection systems use digital fingerprints, called hashes, to recognize previously identified abusive files.
asserted
systems → exist → files
However, newly created material may have no matching fingerprint.
uncertain
material → create → fingerprint
OpenAI's new classifier examines visual signs of potential child sexual abuse material, also known as CSAM.
asserted
classifier → examine → CSAM
That allows it to flag suspected abuse even when the content has never been identified before.
asserted
content → allow → abuse
The system assigns scores to potentially abusive pictures so human reviewers can decide which cases need attention first.
asserted
cases → assign → attention
OpenAI says it developed and evaluated the technology with help from the National Center for Missing & Exploited Children (NCMEC) at missingkids.org/.
asserted
it → say → missingkids.org/.
The company is offering the classifier at no cost to vetted organizations through its API.
asserted
company → offer → API
That could give smaller child protection teams access to detection technology they otherwise couldn't afford.
uncertain
they → give → technology
Early testing shows fewer false alarms
OpenAI says its early testing produced encouraging results.
asserted
testing → show → results
In an evaluation involving pictures uploaded to its systems, the company compared the new classifier with its existing protections.
asserted
company → involve → protections
The findings suggest the new tool could flag 48% fewer pictures for review while detecting roughly the same amount of abusive material.
uncertain
tool → suggest → material
Pictures flagged by the new classifier were also about twice as likely to contain abuse.
asserted
Pictures → flag → abuse
Outside organizations reported additional results.
asserted
organizations → report → results
AI platform Fal said its reviewers confirmed 72% of high-confidence flags as previously unidentified CSAM.
asserted
reviewers → say → CSAM
Meanwhile, SafetyKit reported that the classifier detected 90% of cases involving drawings or animations that sexualize children in its evaluation.
asserted
that → report → evaluation
These results come from OpenAI and participating organizations.
asserted
results → come → OpenAI
Independent testing will help determine how reliably the technology performs across different platforms.
asserted
technology → help → platforms
ChatGPT adds warnings and 24-hour suspensions
OpenAI is also changing how ChatGPT responds when someone attempts to find child sexual abuse material.
asserted
someone → add → material
New warnings will interrupt those attempts and explain the harm caused by sexualizing children.
asserted
warnings → interrupt → children
The warnings specifically address AI-generated material, which can cause serious harm even when a picture has been digitally fabricated.
asserted
picture → address → harm
Along with displaying a warning, OpenAI says it will remove the flagged conversation and pause access to ChatGPT for 24 hours.
asserted
it → display → hours
The warning will also direct the person toward specialist support designed to help prevent harmful behavior.
asserted
warning → direct → behavior
Repeated violations may lead to longer restrictions or permanent account closure.
uncertain
violations → lead → restrictions
OpenAI says they could also result in reports to authorities.
uncertain
they → say → authorities
The new warnings focus on attempts to find abusive material.
asserted
warnings → focus → material
Under its existing policies, OpenAI says it bans accounts and reports attempts to generate or upload CSAM to NCMEC.
asserted
it → exist → NCMEC
The company developed the warnings with input from child protection organizations and researchers who study ways to interrupt harmful online behavior.
asserted
who → develop → behavior
However, whether these interventions prevent future abuse will require continued evaluation.
asserted
prevent → prevent → evaluation
…and 35 more, not listed.