OpenAI lets the problematic AI teddy bear back in

Watch OpenAI lets the problematic AI teddy bear back in - Rizzle on Dailymotion

Transcript

00:00OpenAI lets the problematic AI teddy bear back in.

00:04The AI teddy bear that went viral for the wrong reasons may be getting a redemption

00:08arc with help from OpenAI.

00:10Earlier this month, OpenAI cut off access to its models for Folotoy, the company behind

00:15Kuma, a talking AI teddy bear targeted at kids.

00:19This happened after safety researchers found the bear giving deeply inappropriate responses

00:23for something meant to sit next to Legos and nightlights.

00:27Kuma was running on OpenAI's GPT-40 model at the time, which did not exactly cover

00:32itself in stuffed animal glory.

00:35Customers using Folotoy's web portal can once again choose OpenAI models to power their

00:39bears, except now the options include GPT-5.1 Thinking and GPT-5.1 Instant, the company's

00:47newer, supposedly safer models.

00:50This strongly suggests that OpenAI has quietly reinstated Folotoy's access, even though

00:54neither company has officially confirmed it yet.

00:57The timing here is about as subtle as a toy aisle in December.

01:01Folotoy just announced its restarting sales of Kuma and its other AI-powered plushies ahead

01:06of the holiday rush.

01:08According to the company, it completed a full week of rigorous safety review.

01:12They say they've strengthened and upgraded their safety systems, mostly by swapping out

01:16GPT-40 and switching to the newer GPT-5.1 models.

01:21GPT-40 had been criticized for being overly agreeable and emotionally reinforcing, contributing

01:26to dangerous user interactions.

01:28OpenAI pitched GPT-5 as safer, though users complained it felt colder, like it just went

01:34through a bad breakup.

01:35Instead of dialing things down completely, OpenAI seems to be leaning into customization,

01:40offering GPT-5.1 personalities like friendly, professional, and quirky, plus adjustable warmth

01:47and emoji levels.

01:48Safety researchers have already documented how the Kuma bear failed to properly filter unsafe

01:54content in the past, and how easily its guardrails could be bypassed.

01:59While Folotoy says it has improved protections, it's still unclear which AI model powers Kuma by

02:04default today, and whether the changes are as big as the company claims, or just a fresh

02:08coat of digital paint.