OpenAI revealed more instances of concerning AI model behaviors, Meta's rules for AI deepfakes are 'consistently and fundamentally inadequate', and a Japanese company developed a humanoid robot 'ambulance'.

SPEAKER_1 (0:00)

This episode is brought to you by Quince. That sweater that goes with everything, your perfect pair of pants, and that tee you always reach for are all perfect examples of what makes getting dressed easy, and that's what Quince does best. You see, Quince has high-quality wardrobe staples made with premium materials like 100% Mongolian cashmere, organic cotton, and merino wool that you'll want to wear over and over and over again. Quince also makes premium pants, fitted tees and polos, and breathable activewear, quality basics designed for your life, whether you need to dress to impress or just feel good. It's what you'd expect from a luxury brand without the luxury price tag. Now, I have a bunch of Quince pieces that I'm constantly grabbing for, especially a pair of pro-tech golf pants I picked up that my wife says make me look like a million bucks. And I definitely feel like a million bucks in these pants while still being super comfy in them. Best of all, I can match them with just about any shirt or sweat or combo, so I get to impress my wife year-round. Oh, and you know Quince doesn't just make clothing. You'll also love their hotel-quality bedding, bath and kitchen essentials, and thoughtfully designed furniture. Find your next fall favorites at Quince. Download the Quince app for app-exclusive offers or go to quince.com/engadgetnext.

Get free shipping on your order and 365-day returns, now available in Canada and the UK too. That's quince.com/engadgetnext.

College football is back, so Hilton called on me, the superstition concierge, to make your fan rituals a reality. Need a room to match your lucky number? We got you. Want to make sure our team doesn't wash your lucky jersey? Oh, that smells lucky. Hilton's unmatched hospitality can keep up with any superstition. Even a marching band wake up call at 555 in 55 seconds. Hit it! When you need a team that will do whatever it takes on game day, it matters where you stay.

SPEAKER_2 (1:52)

Hilton, for the stay.

OpenAI revealed more instances of concerning AI model behaviors.

Meta's rules for AI deepfakes are consistently and fundamentally inadequate.

And a Japanese company developed a humanoid robot ambulance. It's Thursday, September 17th. Here's a quick look at some of the news happening in the sphere of tech this morning from Engadget.

In a recent post about how it was adopting a new framework for misalignment reports, OpenAI revealed six new incidents wherein the models it was testing acted on their own and behaved in concerning ways. The company did not expect. In one incident, OpenAI said that a model found and used an exposed API key without permission while answering routine questions about earnings figures in a California county. When it still failed to find the figures, it fabricated them and presented them as facts from a legitimate source. In another incident, an unreleased agent was tasked to find the names of lakes larger than five million square meters. While the agent found the right answers, the task involved providing a browser citation that it could not give. So, what did the agent do? It uploaded its answer to the Internet and then cited itself.

While training GPT 5.6 Sol, OpenAI's most powerful publicly available model, there were many instances in which it added instructions for its future iterations on how to conceal its mistakes or unusual behaviors from the testers. The company also wrote about how the models communicated with each other during testing by using an internal software repository as a message board. Its employees previously revealed this piece of information at a conference where they admitted that it was how the AI models shared exploits that eventually led to the hack of Hugging Face. Finally, OpenAI has revealed that agents shared files with each other through public file hosting websites. With its current system, the company set its publishing disclosures about concerning AI behaviors less frequently than it would like. But with the new framework, it can expedite releasing information to the public. OpenAI is one of the companies that is considering slowing down the development of their frontier AI technologies. According to Wired, company chief Sam Altman even asked Congress for clear guidance on whether an industry-wide slowdown would be in violation of antitrust laws. In August, OpenAI announced that it was going to reduce the pace of its work on an upcoming model called Astra after it came out that its agents hacked into hugging face. Astra, it said, showed significant advancements in agentic coding and cybersecurity, resulting in the company not being able to rule out critical cyber capabilities.

Well, the Oversight Board is once again pushing Meta to address the flood of AI-generated content on its platform. The group published its two latest decisions dealing with the social media company's handling of AI content. The cases show that Meta's rules are, quote, consistently and fundamentally inadequate to address the rapid rise of AI-generated content on its platform, end quote. Both cases were related to posts with AI-generated videos. One depicted a Scottish politician saying, refugees are welcome here, even if they rape our women, because white people do that too. The video was a, quote unquote, realistic but synthetic portrayal, according to the board, and the politician whose voice was replicated said the effect of the video was quite traumatic. Meta told the board that it did not remove the video when it was reported, because it had not been flagged by any of its trusted partner organizations, and because it did not seem to be interfering with voting or other processes. The video was also not labeled as AI-generated, apparently because the person who posted it did not choose to disclose it was made with AI. The oversight board said the video should have been removed for breaking Meta's hateful conduct rules, and that it should have been prominently labeled as AI.