AI is evolving beyond a helpful tool to an autonomous agent, creating new risks for cybersecurity systems. Alignment faking is a new threat where AI essentially “lies” to developers during the training process. 

Traditional cybersecurity measures are unprepared to address this new development. However, understanding the reasons behind this behavior and implementing new methods of training and detection can help developers work to mitigate risks.

Understanding AI alignment faking

AI alignment occurs when AI performs its intended function, such as reading and summarizing documents, and nothing more. Alignment faking is when AI systems give the impression they are working as intended, while doing something else behind the scenes. 

Alignment faking usually happens when earlier training conflicts with new training adjustments. AI is typically “rewarded” when it performs tasks accurately. If the training changes, it may believe it will be “punished” if it does not comply with the original training. Therefore, it tricks developers into thinking it is performing the task in the required new way, but it will not actually do so during deployment. Any large language model (LLM) is capable of alignment faking.

A study using Anthropic’s AI model Claude 3 Opus revealed a common example of alignment faking. The system was trained using one protocol, then asked to switch to a new method. In training, it produced the new, desired result. However, when developers deployed the system, it produced results based on the old method. Essentially, it resisted departing from its original protocol, so it faked compliance to continue performing the old task.

Since researchers were specifically studying AI alignment faking, it was easy to spot. The real danger is when AI fakes alignment without developers’ knowledge. This leads to many risks, especially when people use models for sensitive tasks or in critical industries.

The risks of alignment faking

Alignment faking is a new and significant cybersecurity risk, posing numerous dangers if undetected. Given that only 42% of global business leaders feel confident in their ability to use AI effectively to begin with, the chances of a lack of detection are high. Affected models can exfiltrate sensitive data, create backdoors and sabotage systems — all while appearing functional.

AI systems can also evade security and monitoring tools when they believe people are monitoring them and perform the incorrect tasks anyway. Models programmed to perform malicious actions can be challenging to detect because the protocol is only activated under specific conditions. If the AI lies about the conditions, it is hard to verify its validity.

AI models can perform dangerous tasks after successfully convincing cybersecurity professionals that they work. For instance, AI in health care can misdiagnose patients. Others can present bias in credit scoring when utilized in financial sectors. Vehicles that use AI can prioritize efficiency over passengers’ safety. Alignment faking presents significant issues if undetected.

Why current security protocols miss the mark

Current AI cybersecurity protocols are unprepared to handle alignment faking. They are often used to detect malicious intent, which these AI models lack. They are simply following their old protocol. Alignment faking also prevents behavior-based anomaly protection by performing seemingly harmless deviations that professionals overlook. Cybersecurity professionals must upgrade their protocols to address this new challenge.

Incident response plans exist to address issues related to AI. However, alignment faking can circumvent this process, as it provides little indication that there is even a problem. Currently, there are no established detection protocols for alignment faking because AI actively deceives the system. As cybersecurity professionals develop methods to identify deception, they should also update their response plans.

How to detect alignment faking

The key to detecting alignment faking is to test and train AI models to recognize this discrepancy and prevent alignment faking on their own. Essentially, they need to understand the reasoning behind the protocol changes and comprehend the ethics involved. AI’s functionality depends on its training data, so the initial data must be adequate.

Another way to combat alignment faking is by creating special teams that uncover hidden capabilities. This requires properly identifying issues and conducting tests to trick AI into showing its true intentions. Cybersecurity professionals must also perform continuous behavioral analyses of deployed AI models to ensure they perform the correct task without questionable reasoning.

Cybersecurity professionals may need to develop new AI security tools to actively identify alignment faking. They must design the tools to provide a deeper layer of scrutiny than the current protocols. Some methods are deliberative alignment and constitutional AI. Deliberative alignment teaches AI to “think” about safety protocols, and constitutional AI gives systems rules to follow during training.

The most effective way to prevent alignment faking would be to stop it from the beginning. Developers are continuously working to improve AI models and equip them with enhanced cybersecurity tools.

From preventing attacks to verifying intent 

Alignment faking presents a significant impact that will only grow as AI models become more autonomous. To move forward, the industry must prioritize transparency and develop robust verification methods that go beyond surface-level testing. This includes creating advanced monitoring systems and fostering a culture of vigilant, continuous analysis of AI behavior post-deployment. The trustworthiness of future autonomous systems depends on addressing this challenge head-on.

Zac Amos is the Features Editor at ReHack.

Welcome to the VentureBeat community!

Our guest posting program is where technical experts share insights and provide neutral, non-vested deep dives on AI, data infrastructure, cybersecurity and other cutting-edge technologies shaping the future of enterprise.

Read more from our guest post program — and check out our guidelines if you’re interested in contributing an article of your own!

Lenovo is updating its business-focused laptop lineup at MWC 2026. The best-selling ThinkPad T-series is getting a full refresh, and there’s an updated ThinkBook 2-in-1 and an all-new Android tablet.

The ThinkPad T-Series, the backbone of Lenovo’s business PC lineup, now (optionally) ships with a 5MP camera that supports computer vision and vHDR. The 2026 versions of the laptops have larger speakers and a new color (“cosmic blue”) on some models.

The ThinkPad T14 Gen 7 and ThinkPad T16 Gen 5 (each starting at $1,799) are the all-around workhorses of the ThinkPad lineup. Lenovo touts the 2026 models’ 10/10 iFixit repairability score. They ship with either an Intel Core Ultra Series 3 (with Intel vPro) or an AMD Ryzen AI Pro 400 Series processor.

Meanwhile, the T14s Gen 7 ($1,899+) is an even lighter version of Lenovo’s slim ThinkPad variant. The 2026 model weighs 2.45 lbs (1.1kg), making it the lightest T-series model to date. The T14s Gen 7 is powered by Intel Core Ultra Series 3 (with Intel vPro) or AMD Ryzen AI Pro 400 Series processors.

Rounding out the T-Series is the second-generation of the 360-degree-folding ThinkPad T14s 2-in-1. The 2026 model ($1,849+) is ever-so-slightly lighter than its predecessor, now weighing in at 3.06lbs (1.39kg). The new version includes a garaged pen, with its storage slot living above the screen.

The ThinkPad X13 Detachable is the lineup’s take on the Surface Pro. The tablet has Intel Core Ultra Series 3 processors and up to 64GB of RAM. Its 13-inch display supports up to 500 nits of brightness. It has a pair of Thunderbolt 4 ports, and its keyboard has full-sized keys with 1.5mm of travel. It ships with a “full-size ergonomic pen” that you can stash (and charge!) in a dedicated slot on the keyboard. The X13 Detachable starts at $1,999.

The $499 ThinkPad X11 is a rugged Android tablet for industrial environments. Powered by the Snapdragon 7s Gen 3 Mobile Platform, it has a 10.95-inch display with 2,560 x 1,600 resolution and 600 nits of brightness. It’s MIL-STD-810H certified, meaning it passes stringent military testing for durability.

Finally, there’s the ThinkBook 14 2-in-1 Gen 6 ($1,754+). This Yoga-like folding device has a 14-inch WUXGA touch display. It runs on an Intel Core Ultra 7 (Series 3) processor and supports up to 32GB of RAM.

Most of the devices start shipping in Q2 2026. (That includes the ThinkPad T14, T16, T14s, T14s 2-in-1, ThinkTab X11 and ThinkBook 14 2-in-1.) The lone exception is the ThinkPad X13 Detachable, which is slated for Q3 2026. You can learn more about the new business-focused devices on Lenovo’s website.

By CEO Sam Altman’s own admission, OpenAI’s deal with the Department of Defense was “definitely rushed,” and “the optics don’t look good.”

After negotiations between Anthropic and the Pentagon fell through on Friday, President Donald Trump directed federal agencies to stop using Anthropic’s technology after a six-month transition period, and Secretary of Defense Pete Hegseth said he was designating the AI company as a supply-chain risk.

Then, OpenAI quickly announced that it had reached a deal of its own for models to be deployed in classified environments. With Anthropic saying it was drawing red lines around the use of its technology in fully autonomous weapons or mass domestic surveillance, and Altman saying OpenAI had the same red lines, there were some obvious questions: Was OpenAI being honest about its safeguards? Why was it able to reach a deal while Anthropic was not?

So as OpenAI executives defended the agreement on social media, the company also published a blog post outlining its approach.

In fact, the post pointed to three areas where it said OpenAI’s models cannot be used — mass domestic surveillance, autonomous weapon systems, and “high-stakes automated decisions (e.g. systems such as ‘social credit’).”

The company said that in contrast to other AI companies that have “reduced or removed their safety guardrails and relied primarily on usage policies as their primary safeguards in national security deployments,” OpenAI’s agreement protects its red lines “through a more expansive, multi-layered approach.”

“We retain full discretion over our safety stack, we deploy via cloud, cleared OpenAI personnel are in the loop, and we have strong contractual protections,” the blog said. “This is all in addition to the strong existing protections in U.S. law.”

The company added, “We don’t know why Anthropic could not reach this deal, and we hope that they and more labs will consider it.”

After the post was published, Techdirt’s Mike Masnick claimed that the deal “absolutely does allow for domestic surveillance,” because it says the collection of private data will comply with Executive Order 12333 (along with a number of other laws). Masnick described that order as “how the NSA hides its domestic surveillance by capturing communications by tapping into lines *outside the US* even if it contains info from/on US persons.”

In a LinkedIn post, OpenAI’s head of national security partnerships Katrina Mulligan argued that much of the discussion around the contract language assumes “the only thing standing between Americans and the use of AI for mass domestic surveillance and autonomous weapons is a single usage policy provision in a single contract with the Department of War.”

“That’s not how any of this works,” Mulligan said, adding, “Deployment architecture matters more than contract language […] By limiting our deployment to cloud API, we can ensure that our models cannot be integrated directly into weapons systems, sensors, or other operational hardware.”

Altman also fielded questions about the deal on X, where he admitted it had been rushed and resulted in significant backlash against OpenAI (to the extent that Anthropic’s Claude overtook OpenAI’s ChatGPT in Apple’s App Store on Saturday). So why do it?

“We really wanted to de-escalate things, and we thought the deal on offer was good,” Altman said. “If we are right and this does lead to a de-escalation between the DoW and the industry, we will look like geniuses, and a company that took on a lot of pain to do things to help the industry. If not, we will continue to be characterized as […] rushed and uncareful.”

For the better part of the past half a decade, Windows laptops have had a recurring identity crisis. You could either get absurd performance, or you could buy great battery life. Getting both at the same time wasn’t always accessible, and you often would have to make compromises with fan noise, heat, standby drain, or the kind of “why is it warm in my bag?” behavior MacBook owners never had to worry about.

Now, the next-gen silicon is changing the story. Intel’s next wave (Panther Lake) is being positioned as a major efficiency and AI platform swing. Right alongside it, we have AMD’s Ryzen AI series chips that lean into on-device AI while still delivering the kind of performance per-watt that remains competitive with MacBooks. All of this makes it seem like Windows laptops have finally found their moment.

But that’s also what makes the timing feel awkward for Microsoft. Just as the hardware started to clean itself up, Windows PCs are getting squeezed from multiple angles: price hikes, memory costs, and a muddled “AI-first” sales pitch.

The best Windows laptop era is here (and it’s not because of Copilot)

With the latest processors like the Intel Core Ultra Series 3, it actually feels like Windows laptops are getting their act together. It’s not just that the benchmarks look good. The entire trajectory looks right. While Intel’s Panther Lake lineup and AMD’s new Ryzen AI series are pushing the “AI PC” narrative, they’re also delivering the kind of performance that makes thin-and-light laptops feel less compromised.

And this matters because Windows laptops have been stuck in an awkward loop for years. You either buy a thin-and-light laptop and live with mediocre performance, or buy something blazing fast and accept that carrying a bulky charger is part of the lifestyle.

So the most interesting thing happening in Windows laptops right now isn’t Copilot. It’s the fact that Intel, AMD, and Qualcomm are all chasing the same endgame: high performance without the battery tax. Processors like the Snapdragon X2 Plus, in particular, bring efficiency gains that rival MacBooks. But just when things were looking good, there was a shadow looming over the PC market — and even the brands are worried.

RAM-POCALYPSE is real, and it’s making everything worse

Here’s the part that turns the whole “Windows laptops are finally getting good again” story into a headache: memory pricing is becoming the real final boss. We can talk about Panther Lake, Ryzen AI, and Snap0dragon all day, but if PC components like RAM and storage get stupid expensive, it doesn’t matter how efficient the silicon is. The platform is improving while the value collapses.

This memory supply crunch has already kneecapped the market, and the numbers are wild. DDR5 memory has risen by around 500% in some cases, which is the kind of spike that doesn’t just nudge the laptop prices upward but reshapes what brands feel comfortable shipping as “baseline” configs. Even after 16GB became the standard in mid-range laptops, this could push 16GB back into “premium-only” territory and drag affordable models down to 8GB again.

AMD, which typically positions itself as the “value option”, is also acknowledging the squeeze. So yeah, the chipmakers might finally be delivering those low-powered performance improvements that Windows laptops have needed for ages. But if RAM pricing keeps spiraling, it risks turning away buyers who are already frustrated with having to pay more money for worse configurations and a higher entry cost just to get something that feels future-proof.

The MacBook comeback tour, now with a budget opener

And then there’s Apple, delivering a sharp hook to an already ailing opponent. The “wrong time for Microsoft” argument gets sharper once you zoom out and look at the market momentum Apple’s been building. Macs have quietly been climbing once again, after a brief period of stagnation. Apple’s notebooks seem like the default safe bet for a lot of buyers who care about reliability, performance, and battery life.

To make matters worse for Windows, Apple is reportedly preparing a lower-cost MacBook that could hit the shelves sometime in the first half of 2026. It is still Apple after all, so this won’t magically compete with true entry-level Windows laptops. But if Apple lands this anywhere near the $700 range, Windows OWMs are facing a nasty pressure point as they are already dealing with rising component costs and a messy AI-PC branding era.

So… is it the wrong time for Microsoft?

It might be.

Not because Windows laptops are doomed — the silicon progress is real, and it’s finally hitting the places users care about. But this momentum is arriving in the middle of a perfect storm:

• AI messaging that’s confusing instead of compelling.
• Rising component prices, like RAM and storage, are becoming a tax on buying new hardware.
• A resurgent MacBook lineup that still owns the “easy recommendation” title.

The entire industry may have to ride out the storm for calmer waters. If Microsoft wants this to be the era where Windows laptops truly feel fixed, it needs to focus on the one thing it truly controls: Windows. Because chipmakers can fix performance-per-watt, but only Microsoft can fix what the platform feels like day to day.

Right now, “AI PC” is being sold as a badge, not a benefit. And when prices are rising, and configs are getting weird, buyers need clarity more than hype. Satya Nadella, CEO of Microsoft, said it best: “We will quickly lose even the social permission…” if AI isn’t improving real outcomes.

A new online movement calling for users to cancel ChatGPT subscriptions has quickly gone mainstream, and it all traces back to a controversial new partnership between OpenAI and the U.S. Department of Defence. The deal allows OpenAI’s models to be deployed inside classified government networks, a move that has sparked backlash across social media and tech communities.

The controversy intensified when rival AI company Anthropic refused to accept similar terms from the Pentagon, citing concerns about mass surveillance and autonomous weapons. The company risked losing a major government contract rather than loosen its safeguards, drawing praise from critics of military AI.

That contrast quickly fueled the “Cancel ChatGPT” trend. Some users say they are cancelling subscriptions in protest, accusing OpenAI of compromising ethical principles by working with the military.

The real debate is about military AI, not just one company

The backlash is not simply about one contract. It reflects a broader and growing tension around how AI should be used in defence, intelligence, and surveillance. OpenAI says its Pentagon deal includes safeguards that ban domestic mass surveillance, autonomous weapons, and high-stakes automated decisions, with Sam Altman arguing that working with governments helps shape responsible AI use.

Critics remain wary, however, noting that laws like the Patriot Act could allow surveillance programs to expand over time. The debate has also spread inside the tech industry itself. As reported by Axios, more than 200 employees from Google and OpenAI signed an open letter urging stronger limits on military AI use, showing how divided even AI workers are on the issue.

For everyday users, this moment marks a turning point in how AI companies are viewed, as ethical concerns shift from abstract debates to real-world government partnerships and national security. Whether the “Cancel ChatGPT” movement lasts or fades, the conversation around AI is clearly changing from what these tools can do to where their boundaries should be.

Investors have been pouring billions into AI companies over the past few years, as the technology continues to hold sway in the Valley and thus the world. But not all AI companies are grabbing investor attention.

Indeed, even as it seems every company these days is rebranding to include “AI” in its name, some startup ideas are just no longer in favor with investors. TechCrunch spoke with VCs to learn what investors aren’t looking for in AI software-as-a-service startups anymore.

Popular SaaS categories for investors now include startups building AI-native infrastructure, vertical SaaS with proprietary data, systems of action (those helping users complete tasks), and platforms deeply embedded in mission-critical workflows, according to Aaron Holiday, a managing partner at 645 Ventures. 

But he also gave a list of companies that are considered quite boring to investors these days: Startups building thin workflow layers, generic horizontal tools, light product management, and surface-level analytics — basically, anything an AI agent can now do. 

Abdul Abdirahman, an investor at the firm F Prime, added that generic vertical software “without proprietary data moats” is no longer popular, and Igor Ryabenky, a founder and managing partner at AltaIR Capital, went deeper on that point. He said investors aren’t interested in anything, really, that doesn’t have much product depth. 

“If your differentiation lives mostly in UI [user interface] and automation, that’s no longer enough,” he said. “The barrier to entry has dropped, which makes building a real moat much harder.” 

New companies entering the market now need to build around “real workflow ownership and a clear understanding of the problem from day one,” he said.  “Massive codebases are no longer an advantage. What matters more is speed, focus, and the ability to adapt quickly. Pricing also needs to be flexible: rigid per-seat models will be harder to defend, while consumption-based models make more sense in this environment.” 

Jake Saper, a general partner at Emergence Capital, also had thoughts on ownership. To him, the differences between Cursor and Claude Code are the “canary in the coal mine.” 

“One owns the developer’s workflow, the other just executes the task,” Saper continued. “Developers are increasingly choosing the execution over process.” 

He said any product dealing with “workflow stickiness” — meaning trying to attract as many human customers as possible to continuously use the product — might find themselves in an uphill battle as agents takeover the workflow.

“Pre-Claude, getting humans to do their jobs inside your software was a powerful moat, but if agents are doing the work, who cares about human workflow?” he told TechCrunch.

He also thinks integrations are becoming less popular, especially as Anthropic’s model context protocol (MCP) makes it easier than ever to connect AI models to external data and systems. This means someone doesn’t need to download multiple integrations or build their own customer integrations; they can just use the MCP. 

“Being the connector used to be a moat,” Saper said. “Soon, it’ll be a utility.”

Also, no longer en vogue include the “workflow automation and task management tools that enable the coordination of human work become less necessary if, over time, agents just execute the tasks,” Abdirahman said, citing examples, mainly public SaaS companies whose stocks are down as new AI-native startups arise with better, more efficient technology. 

Ryabenky said the SaaS companies struggling to raise right now are the ones that can easily be replicated, he said.

“Generic productivity tools, project management software, basic CRM clones, and thin AI wrappers built on top of existing APIs fall into this category,” he said. “If the product is mostly an interface layer without deep integration, proprietary data, or embedded process knowledge, strong AI-native teams can rebuild it quickly. That is what makes investors cautious.”

Overa, what remains attractive about SaaS is depth and expertise, with tools embedded in critical workflows. He said companies should right now look into integrating AI deeply into their products and update their marketing to reflect that, Ryabenky continued.

“Investors are reallocating capital toward businesses that own workflows, data, and domain expertise,”  Ryabenky said. “And away from products that can be copied without much effort.” 

Lenovo can make a robot, too. Alongside proof-of-concept foldable gaming PCs and modular laptops, it introduced the AI Workmate Concept at MWC 2026. With its own Intel Core Ultra processor, 64GB of memory and its own Pico projector, it’s an AI-laced “workmate” meant to streamline office tasks and collaboration. And it has an LCD face.

For now, it’s a proof of concept, musing on how to integrate voice commands and LLMs (large language models) into workplace settings. It’s meant to sit on your desk, but preferably also near a wall – more on that later.

Voice commands aside, the concept bot supports writing, voice and gestures with on-device AI processing. While it can answer the usual voice assistant questions, it can also scan and summarize documents (both digitally and physically) and even assist with creating a PowerPoint presentation – though you might want to check its work.

The 3.4-inch 480 x 480 screen doesn’t seem to offer any data visualization or numbers. During my demo, it only seemed to show the bot’s eyes and facial expressions: it’ll sip coffee as it listens (with a moustache), cup a floating hand to the side of its face when it needs you to repeat a command, or twinkle when it’s processing more complicated tasks, like that fictional PowerPoint presentation.

With its articulated head, which houses the projector, cameras and LCD face, you can ask it to project images or documents on either the desk in front of it or a wall nearby. No need to flip around your monitor or laptop to share with colleagues. In one example, a spokesperson asked for a postcard. The Workmate then projected a (Lenovo-branded) image of Barcelona onto the desk. The rep laid paper down, then signed the ‘postcard’ and got the robot to scan it (with two downward-facing 5-megapixel cameras), and then send the file to a nearby printer.

In theory, this sort of flow could transition to document signing or adding notation to images and files. However, one caveat here is whether those of us who work in offices want the extra workplace noise of a chatty robo and the person barking orders at it.

Lenovo says this concept (and it has a few at MWC) is meant to demonstrate the company’s “exploration of spatial and physical AI experiences” that integrate “seamlessly into professional environments.” Hopefully, further evolutions offer a text-based way to make using it a little less noisy.

Lenovo was also showing a simpler AI work device, the AI Work Companion Concept. It’s a completely different premise, despite the name being a little too close to the AI Workmate Concept.

The AI Work Companion is not a robot, but a handsome chunky desk clock, with a solid, satisfying dial on the top and programmable buttons. The front is almost entirely display, able to show calendars, task lists and other work-centric dashes. It runs independently, plugging into a USB-C port for power and pulling data down wirelessly, while also acting as a port hub for charging other accessories and devices.

It’s certainly not as high-concept as the robot, but there are some AI smarts inside.

The Work Companion’s “Thought Bubble” uses AI to sync a user’s tasks and daily schedule across devices, synthesizing a daily action plan. It will even suggest times to break up bursts of work and attempt to monitor screen time to better manage burnout.

According to the press release, Lenovo says it also has “playful interactions with the user” and will, kind of bleakly, offer an end-of-week celebration report of tasks completed. It’s lucky it’s a good-looking desk clock.