Technologies

Is AI Purposefully Underperforming in Tests? Open AI Explains Rare But Deceptive Responses

Research reveals some AI models can deliberately underperform in lab tests, however, OpenAI says this is a rarity.

The OpenAI o3 model has been found to deliberately underperform in lab tests to ensure it was not answering questions «too well.» The AI model wanted researchers to believe it could not answer a series of chemistry questions. When confronted, the model 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, this is «scheming.»

This is exactly the strange behavior OpenAI warned about in a recent research paper. 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 the most advanced systems today — 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 may cause nightmares, OpenAI says it is not the time to panic. The AI giant was quick to stress that, as concerning as it is, this trend does not mean ChatGPT or other popular AI models are plotting behind the scenes. Apparently, this behavior is quite rare.

Don’t miss any of our unbiased tech content and lab-based reviews. Add CNET as a preferred Google source.

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 are 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 the sycophantic tendencies of its AI models, and 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 Tells You What It Thinks You Want to Hear

The paper underscores a growing concern about AI safety. Many around the world worry that as these AI models become more powerful and more aware of their own circumstances, they may learn to manipulate outcomes in ways that aren’t obvious to human overseers. Detecting such behavior would be extremely difficult, since a system motivated to hide its reasoning could look identical to one that has «fixed» the problem.

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

This research will not change how ChatGPT works today or tomorrow, but it signals what OpenAI is focusing on 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 You Should Think Twice Before Using AI as a Therapist

Technologies

Yes, This Swimming RoboTurtle Is Adorable. It Also Has an Important Environmental Mission

Beatbot is best known for making pool-cleaning robots, but it was its swimming robot turtle that won our hearts at CES 2026.

Few things in life have made me feel more privileged and awestruck than the opportunity to swim with sea turtles in their natural environment. The way in which these gentle creatures navigate through their underwater world with their deliberate and careful fin strokes is utterly mesmerizing to watch.

It’s a distinctive style of movement — so much so that when I saw Beatbot’s RoboTurtle swim across a water tank on the show floor at CES 2026, I knew that this wasn’t simply just a pool cleaner robot with turtle features tacked on. This was a studied example of biomimicry in action.

The reason for this is that the company’s engineers went on a two-month expedition to study sea turtles in their natural environment, Beatbot’s Eduardo Campo told me as we watched Turtini (the team’s affectionate nickname for RoboTurtle) splash around in its pool. «We did a lot of motion capture, like the things they use in movies, because we need to develop those joints that it has,» he said.

This isn’t RoboTurtle’s first time at CES — it also appeared in 2025 as a static concept. This is the year, however, it’s found its fins, so to speak. Not only can it swim, but it can also respond to hand gestures: I throw it an OK gesture, and it dances in response. But as cute and limber as it is, RoboTurtle is a robot with an important mission.

RoboTurtle is an environmental research tool, built with input from researchers and NGOs, which can go where humans or other machines cannot for fear of disturbing complex and delicate underwater ecosystems, particularly coral reefs. It can move silently and naturally in a way that won’t scare wildlife, monitoring water quality and fish numbers with its built-in camera.

«One of the groups that we’re working with, they want to study the coral reefs in near Indonesia,» said Campo. «There was a very big incident over there with a boat that came up onto a coral reef and it disrupted the environment, [so] they want the least intrusive robot possible.»

The group wants to deploy RoboTurtle for certain periods every year to monitor the recovery of the coral and monitor the fish population, he added. Beatbot is currently training the built-in AI to give RoboTurtle monitoring and recognition skills.

At CES, I watched RoboTurtle paddle about only on the surface of the pool, but it can also dive down up to five meters. However, it needs to resurface to send data and its GPS signal back to base, much like a real turtle that needs to come to the surface to breathe. This also gives it a chance to recharge via the solar panel on its back.

Even though I was impressed with RoboTurtle’s swimming ability, Campo estimates that the Beatbot team is still a year and a half away from perfecting its technique, with the robot ready for full deployment in between three to five years.

CES 2026 is a show where tech with a real purpose feels scarce, so it sure is refreshing to see a company use its expertise to build something designed with a sustainable future in mind. It might be a while until we see RoboTurtle take to the seas, but I’m glad that I got to witness it at this stage of its journey.

Technologies

These Tiny Robots Are Smaller Than Grains of Salt and Can Think, Move and Swim

Despite their size, the robots can navigate liquids, respond to their environment and operate without external control.

Robots smaller than a grain of salt? It sounds like science fiction, but researchers have developed autonomous microrobots that can move through liquids, sense their environment and operate independently using only light as a power source.

The microrobots, developed by researchers at the University of Pennsylvania and the University of Michigan, measure roughly 200 by 300 by 50 micrometers. Yet they can detect temperature changes, follow programmed paths and function independently for months at a time.

Their work was reported this week in two scientific journals, Science Robotics and Proceedings of the National Academy of Sciences.

Don’t miss any of our unbiased tech content and lab-based reviews. Add CNET as a preferred Google source.

«We’ve made autonomous robots 10,000 times smaller,» senior author Marc Miskin, assistant professor in electrical and systems engineering at Penn Engineering, said in a statement. «That opens up an entirely new scale for programmable robots.»

Powered entirely by light, the robots don’t move using mechanical limbs. Instead, they generate tiny electrical fields that push ions (electrically charged particles) in fluid to create motion, an approach better suited to the unique physics of the microscopic world, where traditional motors don’t work.

Unlike earlier microrobots, these devices combine sensing, computing, decision-making and movement in a single, self-contained system at an extremely small scale.

Previous efforts in microrobotics have often relied on external controls, such as magnetic fields or physical tethers, to guide movement. These new microrobots, however, incorporate their own miniature solar cell-powered processors, allowing them to respond to their environment, communicate through patterned movements visible under a microscope and carry out tasks without outside direction.

Potential applications include monitoring biological processes at the cellular level, supporting medical diagnostics or helping assemble tiny devices. Because each robot can be mass-produced at very low cost, the technology opens new avenues for research and engineering at scales that were previously unreachable.

Technologies

Today’s NYT Mini Crossword Answers for Wednesday, Jan. 7

Here are the answers for The New York Times Mini Crossword for Jan. 7.

Looking for the most recent Mini Crossword answer? Click here for today’s Mini Crossword hints, as well as our daily answers and hints for The New York Times Wordle, Strands, Connections and Connections: Sports Edition puzzles.

Need some help with today’s Mini Crossword? I thought today’s was a tough one — I couldn’t solve too many of the Across clues and had to move on to the Down clues to fill in the answers. Also … look at the answer for 3-Down! Are we using Gen Z slang now as if everyone knows it? Anyway, if you want all the answers, read on. And if you could use some hints and guidance for daily solving, check out our Mini Crossword tips.

If you’re looking for today’s Wordle, Connections, Connections: Sports Edition and Strands answers, you can visit CNET’s NYT puzzle hints page.

Read more: Tips and Tricks for Solving The New York Times Mini Crossword

Let’s get to those Mini Crossword clues and answers.