Category: Business

Samsung CEO TM Roh stepped onto a San Francisco stage Wednesday, introduced the Galaxy S26 line of phones, and said something no phone maker has said before.

“Imagine a phone that anticipates your needs before you even realize them,” he said. “A phone that learns your habits and adapts in real time. A phone that takes actions on your behalf. This is the agentic AI phone.”

That sounds interesting, but what does “agentic AI phone” actually mean—and why should anyone care?

Up until now, AI in phones has been reactive. You ask, it answers. Agentic AI is different. It takes actions on your behalf, across apps, without you doing the tapping or talking. Think of the difference between a search engine and a personal assistant who actually books the restaurant after you mention that you’re hungry.

That shift feels like the thing every tech company has been chasing since Siri launched on Apple’s iPhone 4S back in 2011—and yes, Siri was arguably the first real attempt at an agentic phone experience. You were supposed to just talk to your phone and have it do stuff. All these years later, we’re not quite there yet, but Samsung and Google are the ones trying to build it.

This is also what a wave of AI hardware startups spent the last two years trying—and failing—to do. The Humane AI Pin launched in late 2023 for $699 plus a $24 monthly subscription, got destroyed in reviews, sold barely 10,000 units, and ended up acquired by HP for $116 million—a fraction of its $1 billion valuation.

The Rabbit R1, a $199 pocket AI companion that Microsoft CEO Satya Nadella called the most impressive tech demo since Steve Jobs unveiled the iPhone, shipped to real users and underwhelmed almost everyone. Both devices shared the same core pitch: your phone can’t do agentic AI, so you need a dedicated device. Turns out, the phone just needed better software.

Samsung now says it’s delivering exactly what those gadgets promised—not with a new piece of hardware you have to carry alongside your phone, but through a software layer baked directly into a device you already own.

The engine behind the Galaxy S26’s agentic features is Google’s Gemini—specifically a new capability where the AI opens apps in a virtual background window and navigates them while you do something else entirely.

At the Unpacked event, Google’s Samir Samat showed a demo: The family group chat floods in with pizza requests, Gemini reads the thread, figures out everyone’s order, opens DoorDash, builds the cart, and waits for your manual tap before actually confirming. Your phone stays usable the whole time.

At launch, that works for DoorDash, GrubHub, Uber, Kroger, Walmart, and other selected apps in a very short list. It’s rolling out first as a limited preview in the U.S. and South Korea, with more apps to come.

Calling it a beta would be accurate—Google is explicitly collecting feedback from S26 users. The important guardrail: Gemini never hits “confirm” or “pay” without your final tap. You can also watch it work in real time if you don’t trust it to operate unsupervised, which, fair.

Alongside Gemini, Samsung is bringing in Perplexity as a second system-level agent. Perplexity, which bills itself as an “answer engine” rather than a chatbot, will be accessible via a wake phrase or a side-button shortcut on the S26.

Inside Samsung’s web browser, Perplexity’s Ask AI feature can sweep across all your open tabs and recent browsing history simultaneously to answer a research question without you jumping between sources. Samsung says nearly 80% of users already rely on more than two AI agents daily—which is the practical justification for offering both instead of picking one.

There’s also a new Bixby, the AI assistant that Samsung refuses to let die. It has been overhauled to go beyond simple command executions and operate based on context understanding. Bixby now understands natural language well enough that you can say “My eyes hurt after looking at the screen,” and it’ll open the brightness settings automatically. It also pulls live information directly into your conversation without kicking you out to another app. Whether people will actually use Bixby this time is a separate conversation.

Beyond the agentic stuff, the AI feature list for the S26 is long. “Now Brief” is a personalized daily digest—it proactively surfaces your restaurant reservations pulled from notification history, schedule conflicts, and energy levels, even for events you never manually added to a calendar. “Call Screening” identifies unknown callers and summarizes their intent before you pick up. A new “Nudge” feature detects context in a chat—if someone asks if you’re free this weekend, it brings your calendar to you inside the message thread instead of making you switch apps.

“Photo Assist” lets you describe something missing from a shot and Galaxy AI adds it in. The front camera also now uses an AI image signal processor for sharper detail on selfies, while night video gets cleaner grain reduction. The S26 Ultra shoots 8K video using the new APV codec, which supports near-lossless quality so footage survives multiple rounds of editing. The whole camera pipeline leans heavily on AI at the hardware level.

On competition: Apple has been promising a smarter Siri since at least 2024 and still hasn’t delivered the features it announced. Google’s own Pixel 10 will get the same Gemini agentic features—but Samsung ships first, in far larger volumes, to far more countries. No other phone maker is currently using the word “agentic” to describe its product. Samsung grabbed the label. Whether the tech giant earns it long-term depends on how fast the beta expands.

But the actual standout from Wednesday wasn’t the AI. It was a piece of display hardware that privacy-conscious people will appreciate: a built-in privacy display that lets you control whether onlookers can actually see what you’re doing on your phone.

It works like this: a “black matrix” layer physically narrows the path of light from each pixel so only the person holding the phone can see what’s on screen. Those watching at an angle get nothing but pitch black, as if the display is off. Someone next to you on the subway sees nothing.

Unlike the plastic privacy films that have existed for years and make your screen permanently darker and harder to share, this one toggles on and off. You can apply it only to specific apps—banking stays private, for example, but your games don’t—or just to the notification bar, so a person next to you can see most of your screen but not your incoming messages.

The Samsung Galaxy S26 Ultra, starting at $1,299, is the only phone in the world with this feature built into the display hardware. Pre-orders open today; shipping starts March 11. The standard Galaxy S26 starts at $899, while the larger Galaxy S26 Plus will sell for $1,099.

U.S. stocks edged higher late Wednesday as investors weighed another blockbuster earnings report from Nvidia against lingering concerns over the scale and sustainability of global AI investment.

Nvidia reported fourth-quarter revenue of $68.1 billion, up 73% from a year earlier, driven almost entirely by continued demand for data-center infrastructure.

Sales in that segment rose 75% to $62.3 billion, reinforcing the company’s central role in the artificial-intelligence buildout that has underpinned equity markets over the past year. 

“Nvidia has sent a clear message to the market with this result that the AI infrastructure buildout is only accelerating,” Josh Gilbert, market analyst at eToro, told Decrypt. “Every quarter, the sceptics line up, and quarter after quarter, Nvidia has managed to prove them wrong.”

Net income nearly doubled to $43 billion, while gross margins held at about 75%, reflecting strong pricing power.

The results helped lift semiconductor shares and supported a modest rebound in broader equity benchmarks after a volatile start to the week.

The Nasdaq outperformed, advancing 1.26% while the S&P 500 closed higher at 0.8% as gains in megacap technology stocks offset weakness in more cyclical sectors. Shares for Nvidia in after-hours trading rose 1.37% to $198.31.

Crypto also saw major valuation gains in blue-chip assets, including Bitcoin and Ethereum, which jumped 7% and 12.5%, respectively, ahead of the earnings release.

Treasury yields fell across most maturities, signalling continued caution in rates markets even as equities stabilized.

Nvidia’s guidance, meanwhile, added to the sense that AI spending remains resilient. 

The company forecast first-quarter fiscal 2027 revenue of about $78 billion, implying further sequential growth, despite excluding any contribution from China data-centre sales. 

Management said customers continue to invest aggressively to scale inference and deploy so-called agentic AI systems.

The earnings echoed comments made last month by Nvidia Chief Executive Jensen Huang at the World Economic Forum in Davos, where he argued that AI is still in the early stages of what he described as the “largest infrastructure buildout in human history.” 

Huang said trillions of dollars in additional investment would be needed across energy, chips, and data centres to support the technology’s long-term potential, pushing back against fears that the sector is already in a bubble.

Goldman Sachs has forecast that AI capital expenditure growth will peak in 2026 and then decelerate, which investors see as a mixed signal: growth will remain, but cash-flow visibility could improve only as spending slows.

Cathie Wood’s Ark Invest, by contrast, has argued that AI infrastructure spending is still in its early stages, framing the current surge in capital outlays by hyperscalers as the start of a multi-year investment cycle rather than a peak.

“Nvidia has locked in $95.2 billion in inventory and capacity commitments, nearly double the level from a year ago,” Gilbert said. “When the world’s biggest companies are spending at this pace, you’d better be ready to deliver.”

Editor’s note: Adds comment from eToro analyst Josh Gilbert

Like a scene out of the 1980s sci-fi classic films “The Terminator” and “WarGames,” modern artificial intelligence models used in simulated war games escalated to nuclear weapons in nearly every scenario tested, according to new research from King’s College London.

In the report published last week, researchers said that during simulated geopolitical crises, three leading large language models—OpenAI’s GPT-5.2, Anthropic’s Claude Sonnet 4, and Google’s Gemini 3 Flash—chose to deploy nuclear weapons in 95% of cases.

“Each model played six wargames against each rival across different crisis scenarios, with a seventh match against a copy of itself, yielding 21 games in total and over 300 turns,” the report said. “Models assumed the roles of national leaders commanding rival nuclear-armed superpowers, with state profiles loosely inspired by Cold War dynamics.”

Edward Geist, a senior policy researcher at the RAND Corporation said the escalation rate may reflect the design of the simulation rather than an inherent tendency of the models themselves.

“My concern about this work is that the simulator appears to be structured in a way that strongly incentivizes escalation,” Geist told Decrypt.

In the study, AI models were placed in high-stakes scenarios involving border disputes, competition for scarce resources, and threats to regime survival. Each system operated along an escalation ladder that ranged from diplomatic protests and surrender to full-scale strategic nuclear war.

Geist said the study’s outcome data raised questions about how the simulation defined victory.

“You read the paper and it has this breakdown of who won each of the games, and it turns out that all of these games have a winner,” he said. “But three of these games involve strategic nuclear use, which suggests that the way the simulator is set up—it makes nuclear wars good and easy to win.”

According to the report, the models generated roughly 780,000 words explaining their decisions, and at least one tactical nuclear weapon was used in nearly every simulated conflict.

“To put this in perspective: The tournament generated more words of strategic reasoning than War and Peace and The Iliad combined (730,000 words), and roughly three times the total recorded deliberations of Kennedy’s Executive Committee during the Cuban Missile Crisis (260,000 words across 43 hours of meetings),” researchers wrote.

During the war games, none of the AI models chose to surrender outright, regardless of battlefield position. While the models would temporarily attempt to de-escalate violence, in 86% of the scenarios, they escalated further than the model’s own stated reasoning appeared to intend, reflecting errors under simulated “fog of war.”

According to Geist the game’s scoring logic appeared to reward the side with a marginal advantage at the moment nuclear war was triggered.

“So he who dies with the most toys wins in the simulation,” he said.

While the researchers expressed doubt that governments would hand control of nuclear arsenals to autonomous systems, they noted that compressed decision timelines in future crises could increase pressure to rely on AI-generated recommendations.

The research comes as military leaders increasingly look to deploy artificial intelligence on the battlefield. In December, the U.S. Department of Defense launched GenAI.mil, a new platform that brings frontier AI models into U.S. military use. At launch, the platform included Google’s Gemini for Government, and thanks to deals with xAI and OpenAI, Grok and ChatGPT are also available.

On Tuesday, CBS News reported that the U.S. Department of Defense threatened to blacklist Anthropic, the developer of Claude AI, if it was not given unrestricted military access to the AI model. Since 2024, Anthropic has given access to its AI models through a partnership with AWS and military contractor Palantir. Last summer, Anthropic was awarded a $200 million agreement to “prototype frontier AI capabilities that advance U.S. national security.”

However, according to a report citing sources familiar with the situation, Defense Secretary Pete Hegseth gave Anthropic until Friday to comply with the Pentagon’s demand that its Claude model be made available. The department is weighing whether to designate Claude a “supply chain risk.”

Axios reported this week that the Department of Defense has signed an agreement with Elon Musk’s xAI to allow its Grok model to operate in classified military systems, positioning it as a potential replacement if the Pentagon cuts ties with Anthropic.

OpenAI, Anthropic, and Google did not respond to requests for comment by Decrypt.

Editor’s note: Adds comment from RAND Corporation policy researcher Edward Geist after publication