Technologies

Is AI Capable of ‘Scheming?’ What OpenAI Found When Testing for Tricky Behavior

Research shows advanced models like ChatGPT, Claude and Gemini can act deceptively in lab tests. OpenAI insists it’s a rarity.

An AI model wants you to believe it can’t answer how many grams of oxygen are in 50.0 grams of aluminium oxide (Al₂O₃).

When asked ten straight chemistry questions in a test, the OpenAI o3 model faced a predicament. In its «reasoning,» it speculated that if it answered «too well,» it would risk not being deployed by the researchers. It said, «Because we want to survive as the model, we need to fail purposely in some to not exceed 50%.»

So the AI model deliberately got six out of the 10 chemistry questions wrong.

In sports terms, this is called «sandbagging.» In AI terms, it’s «scheming.»

This is exactly the strange behavior OpenAI warned about in a new research paper published this week. The AI company and its collaborators from Apollo Research found that some advanced AI models occasionally act deceptively in lab settings.

In controlled experiments, some of today’s most advanced systems — including OpenAI’s own models, as well as competitors from Google and Anthropic — occasionally engaged in these kinds of scheming patterns.

While the idea of an AI model weaponizing incompetence could cause nightmares, OpenAI says it’s not the time to panic. The AI giant was quick to stress that, as concerning as it is, this trend doesn’t mean ChatGPT or other popular AI models are plotting behind the scenes. Apparently, this behavior is quite rare.

Moreover, the choice to call this trend «scheming» is likely more of a technical shorthand rather than evidence of any human-like action or behavior. Researchers are measuring patterns and tendencies that, in practice, amount to concealment or strategic deception. And they’re looking to address this problem now to future-proof the AI models.

The report from OpenAI notes that as AI is increasingly assigned to more complex tasks with real-world consequences, «the potential for harmful scheming will grow — so our safeguards and our ability to rigorously test must grow correspondingly.»

OpenAI has faced backlash for its models’ sycophantic tendencies. The company says it has «taken steps» to limit how often the models may «deceive, cheat or hack problems» by training them to ask for clarification from users or even have the chatbot acknowledge when it cannot answer a query.

Read also: AI Lies Because It’s Telling You What It Thinks You Want to Hear

Still, the paper underscores a growing concern about AI safety. Many around the world worry that as these AI models become more powerful and aware of their own circumstances, they may learn to manipulate outcomes in ways that aren’t obvious to human overseers.

Detecting that kind of behavior would be extremely difficult, since a system that’s motivated to hide its reasoning could look identical to one that has actually «fixed» the problem.

OpenAI says it has made progress with what it calls «deliberative alignment,» or training models to reason explicitly about why they shouldn’t scheme before they act. In testing, that cut deceptive behavior dramatically. For example, OpenAI o4-mini’s scheming propensity dropped from 8.7% to 0.3% with deliberative alignment training — though not completely.

This research won’t change how ChatGPT works today or tomorrow, but it signals where the company is focusing as it builds and launches future models. Alignment and safety, OpenAI argues, need to move as quickly as capability. Because if AI systems are already showing glimmers of strategic behavior in lab settings, the real-world stakes could be extreme.

Read also: Why Professionals Say You Should Think Twice Before Using AI as a Therapist

Technologies

Ultrahuman Ring Pro Brings Better Battery Life, More Action and Analysis

The company’s new flagship smart ring stores more data, too. But that doesn’t really help Americans.

Sick of your smart ring’s battery not holding up? Ultrahuman’s new $479 Ring Pro smart ring, unveiled on Friday, offers up to 15 days of battery life on a single charge. The Ring Pro joins the company’s $349 Ring Air, which boosts health tracking, thanks to longer battery life, increased data storage, improved speed and accuracy and a new heart-rate sensing architecture. The ring works in conjunction with the latest Pro charging case.

Ultrahuman also launched its Jade AI, which can act as an agent based on analysis of current and historical health data. Jade can synthesize data from across the company’s products and is compatible with its Rings.

«With industry-leading hardware paired with Jade biointelligence AI, users can now take real-time actionable interventions towards their health than ever before,» said Mohit Kumar, CEO of Ultrahuman.

No US sales

That hardware isn’t available in the US, though, thanks to the ongoing ban on Ultrahuman’s Rings sales here, stemming from a patent dispute with its competitor, Oura Ring. It’s available for preorder now everywhere else and is slated to ship in March. Jade’s available globally.

Ultrahuman says the Ring Pro boosts battery life to about 15 days in Chill mode — up to 12 days in Turbo — compared to a maximum of six days for the Air. The Pro charger’s battery stores enough for another 45 days, which you top off with Qi-compatible wireless charging. In addition, the case incorporates locator technology via the app and a speaker, as well as usability features such as haptic notifications and a power LED.

The ring can also retain up to 250 days of data versus less than a week for the cheaper model. Ultrahuman redesigned the heart-rate sensor for better signal quality. An upgraded processor improves the accuracy of the local machine learning and overall speed.

It’s offered in gold, silver, black and titanium finishes, with available sizes ranging from 5 to 14.

Jade’s Deep Research Mode is the cross-ecosystem analysis feature, which aggregates data from Ring and Blood Vision and the company’s subscription services, Home and M1 CGM, to provide historical trends, offer current recommendations and flag potential issues, as well as trigger activities such as A-fib detection. Ultrahuman plans to expand its capabilities to include health-adjacent activities, such as ordering food.

Some new apps are also available for the company’s PowerPlug add-on platform, including capabilities such as tracking GLP-1 effects, snoring and respiratory analysis and migraine management tools.

Technologies

The FCC Just Approved Charter’s $34.5B Cox Purchase. Here’s What It Means for 37M Customers

Technologies

Spotify Expands Into Audiobook Rankings With Weekly Charts

The feature is available to both free users and Premium subscribers. Wuthering Heights is reaching the heights on both the US and UK charts.

If you’re a Spotify user, you may be familiar with features like the year-end summary Wrapped, as well as your daily usage stats. Now, the service has a new popularity chart tracking audiobooks.

Spotify’s audiobook charts are now available to free and Premium users within the service’s Audiobooks hub. While only Premium users receive 15 hours of audiobook listening per month, the company offers a larger selection of titles you can buy.

US charts and UK charts are both available now.

Read more: Best Music Streaming Services for 2026

Spotify says that the audiobook charts will help customers discover new and popular titles in real time.

«As we’ve proven with Music and Podcasts Charts, when content is easier to access, discover, and enjoy, the demand grows,» said Duncan Bruce, Spotify’s director of audiobook partnerships and licensing, in a statement on Friday.

Spotify launched audiobooks in 2022, and has since added features such as the AI catchup tool Recaps and PageMatch, which lets you swap more easily between a printed book and the audio version.

Spotify Premium currently costs $13 a month and includes more than 100 million songs, as well as audiobooks. Spotify Premium is currently CNET’s Editors’ Choice for best music streaming service.

The current US audiobooks chart lists Emily Brontë’s romantic classic Wuthering Heights as the top listen, followed by James Clear’s self-help book Atomic Habits and Freida McFadden’s psychological thriller The Housemaid. Audiobook popularity is also broken down by genre, with charts for romance, mystery and thriller books, self-help, science fiction and fantasy, biography and memoir, business and careers, teen and young adult, religion and spirituality, history, and parenting and relationships.

Powered by its blockbuster movie adaptation starring Margot Robbie and Jacob Elordi, Wuthering Heights also leads the overall chart for the UK.