Feed aggregator

NPR's Sarah McCammon talks to Kori Schake of the American Enterprise Institute, about her recent piece in Foreign Affairs, Dispensable Nation: America in a Post-American World.

NPR's David Folkenflik shares what it's been like covering President Trump's contentious relationship with the media, including public media and NPR itself.

NPR's Sarah McCammon speaks with Carol Mason about her new book, From the Clinics to the Capitol: How Opposing Abortion Became Insurrectionary.

• Researchers recreated the Equifax hack and watched AI do everything without direct control
• The AI model successfully carried out a major breach with zero human input
• Shell commands weren’t needed, the AI acted as the planner and delegated everything else

Large language models (LLMs) have long been considered useful tools in areas like data analysis, content generation, and code assistance.

However, a new study from Carnegie Mellon University, conducted in collaboration with Anthropic, has raised difficult questions about their role in cybersecurity.

The study showed that under the right conditions, LLMs can plan and carry out complex cyberattacks without human guidance, suggesting a shift from mere assistance to full autonomy in digital intrusion.

From puzzles to enterprise environments

Earlier experiments with AI in cybersecurity were mostly limited to “capture-the-flag” scenarios, simplified challenges used for training.

The Carnegie Mellon team, led by PhD candidate Brian Singer, went further by giving LLMs structured guidance and integrating them into a hierarchy of agents.

With these settings, they were able to test the models in more realistic network setups.

In one case, they recreated the same conditions that led to the 2017 Equifax breach, including the vulnerabilities and layout documented in official reports.

The AI not only planned the attack but also deployed malware and extracted data, all without direct human commands.

What makes this research striking is how little raw coding the LLM had to perform. Traditional approaches often fail because models struggle to execute shell commands or parse detailed logs.

Instead, this system relied on a higher-level structure where the LLM acted as a planner while delegating lower-level actions to sub-agents.

This abstraction gave the AI enough context to “understand” and adapt to its environment.

Although these results were achieved in a controlled lab setting, they raise questions about how far this autonomy could go.

The risks here are not just hypothetical. If LLMs can carry out network breaches on their own, then malicious actors could potentially use them to scale attacks far beyond what’s feasible with human teams.

Even tools such as endpoint protection and the best antivirus software may be tested by such adaptive and responsive agents.

Nevertheless, there are potential benefits to this capability. An LLM capable of mimicking realistic attacks might be used to improve system testing and expose flaws that would otherwise go unnoticed.

“It only works under specific conditions, and we do not have something that could just autonomously attack the internet… But it’s a critical first step,” said Singer in explaining that this work remains a prototype.

Still, the ability of an AI to replicate a major breach with minimal input should not be dismissed.

Follow-up research is now exploring how these same techniques can be applied in defense, potentially even enabling AI agents to detect or block attacks in real-time.

You may also like

• ByteDance AI tool Trae caught spying on users
• These are the best VPN with antivirus solutions
• Take a look at our pick of the best password managers

Here are hints and the answer for today's Wordle for Aug. 3, No. 1,506.

Here are hints and answers for the NYT Strands puzzle for Aug. 3 No. 518.

Dozens of Palestinians were killed, many while waiting for food aid, amid a deepening starvation crisis and despite Israeli assurances of a humanitarian pause in some areas of the territory.

(Image credit: Eyad Baba)

Here are the best places to buy and own digital music from stores such as iTunes, Bandcamp and more.

• LG UltraFine 40-inch monitor delivers 5K2K resolution and strong color specs for creative workflows
• Thunderbolt 5 brings high-speed data up to 120Gbps and 96W charging
• Nano IPS Black panel doubles contrast over regular IPS monitors

LG has released its latest 5K monitor as it looks to encourage the hardware into territory once reserved for niche creative professionals.

The LG 40U990A-W UltraFine 40-inch monitor features a 39.7-inch curved ultrawide screen with a 5120 x 2160 resolution, and it is the world’s first display to support Thunderbolt 5 connectivity.

It arrives with a steep price of approximately $2,246, raising questions about how much functionality users genuinely need and what justifies the cost.

Targeting creative precision with premium display specs

The UltraFine Nano IPS Black panel boasts a 2000:1 contrast ratio, which doubles that of traditional IPS monitors.

Combined with 99% DCI-P3 coverage and VESA DisplayHDR 600 certification, LG is targeting professionals in video production and photo editing, where color fidelity and shadow detail matter.

Its wide 21:9 aspect ratio and 2500R curvature add a layer of immersion that some professionals may appreciate, making the 40U990A-W a capable business monitor.

For those who rely on pixel accuracy, this monitor could appeal as the best monitor for video editing, although its impact will depend on how well these specifications translate into real-world consistency and performance.

From a hardware perspective, the presence of Thunderbolt 5 brings speed improvements for data transfer up to 120 Gbps and enables up to 96W of laptop charging.

The port also supports daisy chaining of 5K2K monitors, allowing complex, multi-display setups without clutter.

While this might suggest future-forward convenience, the real-world benefits of Thunderbolt 5 remain largely untested at scale.

Early buyers of this device may encounter compatibility issues or diminishing returns compared to more mature standards.

This 5K monitor also features a 3-sided narrow bezel, with adjustable height, swivel, and tilt options that are now standard on premium displays.

You might also like

• We've rounded up the best monitors for photo editing on the market today
• And here are the best monitors for programming we've found
• Want to host an Nvidia GeForce RTX 5090 GPU and up to 4PB of SSD storage on one single PCIe slot?