As you’ve no doubt noticed, there’s a lot to read on the internet. Vast numbers of new articles are published every day. There’s still quite a lot of it not being written by AI, too.

Trying to keep up with everything can feel exhausting, and that’s where read-it-later apps come in. These apps will queue up content that catches your eye as you go to and fro across the web, and keep these articles, videos, blog posts, and social media updates bookmarked until you get a chance to catch up with them. When it’s time to dig in, the apps present the articles in a clean, uncluttered layout.

It’s a neat idea, but it needs to be well implemented for it to work—otherwise you can spend more time compiling and organizing your reading than actually catching up on your queue.

Instapaper is one of the oldest read-it-later apps, having launched back in 2008. It’s also still one of the best options, even more so now that it’s been given a major revamp. Here’s what’s new, and why it might be worth your while diving back into your Instapaper account (or signing up for the first time).

Instapaper Basics

Instapaper offers a clean, distraction-free reading environment.Photograph: David Nield

Instapaper is available for the web, Android, and iOS, and reflows web articles into a tidily arranged, distraction-free format that doesn’t feature any ads, menus, or other onscreen furniture. You just get the text and the images, so reading the articles you’ve saved is a more pleasurable experience than it is on the web.

You can customize the Instapaper interface in a variety of ways. The apps let you change the font style, size, and spacing, and adjust the background color of the “paper” that the text and images are placed on.

Everything is neatly arranged on the screen, too. You get estimates for how long each article will take to read (as well as the option to see summaries of the text). You can use folders and tags to categorize saved posts in a variety of ways. Also, everything syncs seamlessly across all the devices you’re using Instapaper on.

Getting articles into Instapaper—a crucial part of the process—is straightforward. On mobile you can pass web links into the app through the standard share tools, while on desktop you can either paste the URLs straight into Instapaper, or use the extensions that are available for the most popular browsers.

Why it matters: A vest tested during NASA’s Artemis I mission could reduce astronauts’ radiation exposure during major solar storms, according to a new study. The findings show how targeted shielding could support crew safety as NASA prepares for longer missions around and eventually on the moon.

The AstroRad vest flew around the moon in 2022 on Zohar, one of two female manikins aboard NASA’s Orion capsule. Researchers compared Zohar’s readings with those from Helga, a second manikin that wore no vest. Both contained thousands of radiation sensors.

The flight did not encounter a major solar storm. Instead, researchers used the data collected as Orion passed through the inner Van Allen radiation belt to model the vest’s performance during storms similar to those seen in 1972 and 1989.

The study found that the vest could lower an astronaut’s radiation dose by about 60% during a solar event like the 1972 storm, while it could reduce exposure by nearly 40% during a storm comparable to the 1989 event. Researchers said those reductions are equivalent to 193 days and 131 days, respectively, of deep-space radiation exposure.

The findings are relevant as NASA prepares for longer lunar missions. Artemis I was an uncrewed test flight, but NASA plans to send four astronauts on a lunar flyby mission in April. Future missions are expected to support longer stays on and around the moon as the agency works toward a lunar base.

Radiation is a major safety concern beyond low-Earth orbit. Astronauts closer to Earth benefit from the planet’s atmosphere and magnetic field, which limit exposure. That protection falls away during missions to the moon and, eventually, Mars.

Zohar wore the 57-pound vest, developed by Israeli startup StemRad and US aerospace company Lockheed Martin. The garment is not intended for daily use. It is designed for periods of elevated solar radiation, when crews may need additional protection inside a spacecraft or shelter.

Rather than covering the entire body, the vest focuses on areas considered especially sensitive to radiation. It protects the lungs, stomach, bone marrow, breasts, and ovaries. Women are believed to face a higher risk of radiation-related cancer than men.

StemRad and Lockheed Martin previously tested a version of the vest on the International Space Station. The design was modeled after gear used by nuclear-disaster response teams. It was not included on Artemis II because that mission is expected to last 10 days and Orion has limited space, but it could be considered for later flights.

The researchers based their simulations on two past solar events. The 1972 storm occurred between the Apollo 16 and Apollo 17 moon landings. If astronauts had been traveling to or working on the moon at the time, they could have faced potentially lethal exposure. The 1989 event was linked to a power-grid collapse and the detonation of magnetic sea mines.

The team is now trying to make the vest lighter without weakening its protective properties. It is also studying whether spacecraft supplies and materials could be repurposed as radiation shielding during emergencies.

Jordan Houri, the study’s lead author and a researcher at StemRad, said additional deep-space testing is not expected soon because Zohar and Helga “provided us with an exceptionally large, high-fidelity dataset.”

Zohar and the vest are now on display at Kennedy Space Center in Florida.

Image credit: The AP

Anthropic has told prospective investors that second-quarter revenue exceeded $11.5bn, more than 14 times the $787mn it made a year earlier, and that it recorded positive adjusted operating income. The figures are preliminary and come as the company prepares a listing.

Anthropic has told prospective investors that its second-quarter revenue passed $11.5bn. That is more than 14 times the $787mn it recorded in the same quarter of 2025, according to documents seen by Bloomberg.

The sequential jump is almost as steep. First-quarter revenue was $4.73bn, so the company has more than doubled inside three months and made roughly $16.2bn in the first half of the year.

The quieter line matters more. Anthropic recorded positive adjusted operating income in the quarter, which is not something the leading AI labs have been able to say.

All of it is preliminary and could be revised, and Anthropic declined to comment.

What is genuinely new here is the shape of the disclosure. The industry has asked to be valued on annualised run rates, a figure that takes current revenue and multiplies it out, and Anthropic has now put an actual completed quarter and an operating income line in front of investors instead.

The numbers also hold together. Four quarters at $11.5bn works out at $46bn, close to the $47bn run rate the company disclosed in May, which is more consistency than these figures usually show.

Comparison with OpenAI needs care. Its reported figure of more than $40bn is a run rate rather than a quarter’s revenue, and as Bloomberg notes, the two may not be calculated the same way.

The purpose of the disclosure is a listing. Anthropic has filed confidentially and is working with Morgan Stanley, Goldman Sachs and JPMorgan, with backers telling the Financial Times they expect a $2trn valuation in October.

Timing is part of the strategy. An autumn debut would put Anthropic on the public market before OpenAI, whose own valuation is under scrutiny from its investors, and before DeepSeek, which is preparing to file.

There is capacity waiting. Listings have raised $256.4bn this year excluding blank-cheque vehicles, the most since 2021.

Aurora Innovation and Kodiak AI, two companies developing self-driving trucks, have received permits from the California Department of Motor Vehicles to test their autonomous vehicle technology on public roads. And Kodiak has already started.

Kodiak said it is starting with a handful of test trucks in California, primarily around its Mountain View office.

Both companies applied for testing permits this spring under updated rules approved by the California DMV, the agency that regulates autonomous vehicles in the state. Those rules, approved April 28, lifted a ban on driverless vehicles weighing over 10,000 pounds from testing on public roads and provided a regulatory path for companies to test, and eventually deploy, autonomous heavy-duty vehicles.

Kodiak and Aurora met the DMV’s permitting criteria, including safety, insurance, vehicle registration, safety driver qualification, and other requirements, according to the agency. The testing permits require a human safety operator to be behind the wheel and prohibit companies from operating on roads where the posted speed limit is 25 miles per hour or less, unless they are on a “direct route” between destinations.

Despite those restrictions, the prospect of self-driving trucks on California’s highways continues to fuel opposition.

Last week, Teamsters California sued the state’s DMV, alleging that the agency circumvented laws requiring it to study and publicly disclose the possible economic impacts of allowing self-driving trucks on public roads. The lawsuit, filed in the state’s Superior Court in Alameda County, also accused the DMV of failing to consider the safety risks to motorists.

For autonomous vehicle developers, the new rules allow them to test and eventually deploy in their home territories.

Aurora and Kodiak are both based in California, but until recently focused their testing and deployment activities in Texas because of their home-state’s ban on autonomous heavy-duty trucks. Aurora launched a self-driving truck service in Texas in May 2025, starting with a Dallas-to-Houston route. The company has added more routes since, including between Fort Worth and El Paso, El Paso and Phoenix, Fort Worth and Phoenix, and Laredo and Dallas.

Kodiak began commercial driverless truck operations in January 2025 in an off-road environment in West Texas’ remote Permian Basin. Kodiak has since expanded to on-highway operations, including a route between Dallas and Houston.

We think it’s kind of neat.

How the sensor reads your skin

How accurate is the Galaxy Watch 8’s Antioxidant Index?

Star Wars is at its best when it doesn’t take itself too seriously, and there’s no greater example of this than Lego Star Wars. The franchise, complete with its own video game on multiple platforms, prides itself on delivering a cute story with loads of laughs and inside jokes just for the biggest fans.

Fast on the heels of The Mandalorian and Grogu leaving theaters and hitting Disney Plus, the duo and a lot of their friends are getting a Lego Star Wars film just about them. And if you happened to be at D23, there was a chance to watch the whole thing early.

Dropping on Disney Plus September 2nd, Lego Star Wars: The Mandalorian is a hilarious trip down memory lane done Lego style. Recaps of the entire Mandalorian story so far, plus or minus some embellishments from the storytellers take up the one hour special in a way that kept me laughing the entire time. Keep an ear out for some of the original actors from the series showing up to voice their Lego counterparts as well as the standard array of Lego hijinks.

This is a cute, family-friendly way to catch up or refresh your memory. It includes sections of The Mandalorian and The Book of Boba Fett, with the story bringing you to just before the start of the film. And since The Mandalorian and Grogu movie will hit Disney Plus the same day as its Lego preview, you can treat yourself to some giggles before locking in and enjoying the action-packed film right after. Just be careful with your wallet after, or you’ll end up with a shopping cart full of your own Star Wars Lego kits.

Russell Holly is a Managing Editor on the Commerce team at CNET. He works with all of CNET to assemble top recommendations as well as helping everyone find the best way to buy anything at the best price. When not writing for CNET you can find him riding a bike, running around in Jedi robes, or contributing to WOSU public radio’s Tech Tuesday segment.

See full bio

By Rajan Kapoor, VP Security, Material Security

Over the past two months, I’ve written about the Vercel breach and the Composio breach separately. Both offer lessons to learn on their own. But reading them together, I keep coming back to the same observation: these aren’t isolated incidents

They’re the same attack, run twice, against different targets, where email was not the entry point into the workspace. And once you see the pattern clearly, it changes what you think you need to defend.

It also raises an uncomfortable question that I’ve been sitting with. The pattern I’m describing, where an OAuth grant is used to access an account, read sensitive data from email and Drive, and use that access to move past the workspace, doesn’t only describe what attackers do. It increasingly describes what AI agents do, by design, every day.

Before jumping into that discussion, let’s take a moment to map out the workspace attack chain.

The old mental model: email is where the danger lies

For most of the last decade, the dominant mental model for workspace security looked something like this: email is the dangerous channel, and everything else in Google Workspace is relatively safe.

That model made sense when attackers were primarily trying to steal credentials through phishing. It doesn’t hold anymore because attackers have learned to chain their way through the workspace, not just get in via an inbox.

The model most security teams are familiar with looks something like this:

• Email is the entry point: A malicious email (a phishing link, a weaponized attachment, a convincing pretext, a prompt intended to misdirect an AI agent) is how most attacks begin.
• A credential is stolen: The attack results in a valid credential being stolen and an account takeover happening.
• Sensitive data is accessed in Gmail and Drive: Once the takeover happens, the attacker easily jumps to connected apps within Google Workspace.
• Lateral pivots: An attacker inside an inbox can reset passwords and get into additional apps via magic links.
• Establish persistence: Attackers can sit undetected within accounts for days, weeks, or months, quietly exfiltrating data across systems.

Taken together, this is the nightmare scenario that is widely recognized as the account takeover (ATO). The workspace attack chain begins with an identity compromise via email and expands from there.

The evolving attack chain: OAuth is the entry point

The elements of the workspace attack chain attack haven’t changed, but the order in which the attacks unfold has evolved. The sequence I’ve now watched play out across Vercel, Composio, and a growing number of incidents we’re tracking doesn’t start with email at all.

Instead, the script gets flipped and an OAuth token becomes the entryway into email, not the other way around.

Here’s what these attacks looked like:

• OAuth is the entry point: These attacks started by establishing persistence through a stolen OAuth token. These tokens survive password resets, don’t expire, and are hard to observe. They are invisible to users and largely invisible to security teams who aren’t monitoring app behavior. Even scarier, the stolen token is a supply chain attack. A supplier is compromised and the result is access into your environment.
• Sensitive data is accessed: Using the stolen token, the attacker was able to get into data stored in Gmail and Drive.
• Email accounts are taken over: The ATO is initially executed using OAuth, not email. The access to email converts a compromised inbox into a much broader incident.
• Lateral pivots: Using a combination of credentials stored in Drive and password resets or magic links via email,the attacker can then move laterally across connected systems.

We can anticipate that the building blocks of the workspace attack chain will remain consistent, but that attackers – equipped with AI tools to sniff out vulnerabilities and scale their efforts – will continue to find ways to recombine them.

These OAuth-centric attacks are only one example of this evolution.

The same chains, a different actor

Now let’s shift our thinking, while keeping those four-step sequences I just described in our minds.

Your employees are connecting AI agents to Google Workspace right now. Those agents are authorized. They’re using legitimate OAuth grants. They’re reading email, searching Drive, operating on behalf of real users to do real work. In most organizations, this is happening faster than security teams can track it.

When an AI agent behaves unexpectedly — because its instructions were ambiguous, because it followed a chain of reasoning its developers didn’t anticipate, because it was fed a prompt through content it encountered in the environment — it can walk the same path as an attacker:

• It accesses an inbox or Drive folder it wasn’t explicitly intended to reach, because its scope was broader than its task required.
• It reads sensitive content like credentials in email threads or confidential documents in shared drives and makes use of that information.
• It takes an action downstream from that access: sending a message, following a link, making a request to another service.
• It moves laterally across applications and sensitive information winds up getting exfiltrated to a third party.

No malicious actor. No compromised credential. Just an agent doing something its operator didn’t intend, in an environment that didn’t have the controls to stop it.

Why this matters for how you think about defense

Most conversations about AI agent security are framed around preventing prompt injection, red-teaming agent behavior, or reviewing what apps your employees are connecting. Those are real problems and worth solving.

But the threat I’m describing isn’t about an agent being weaponized. It’s about an agent operating exactly as it was built to operate, in an environment where the guardrails weren’t designed with that kind of actor in mind.

A human operator acting in an environment where they’ve been overpermissioned will generally know how to navigate this situation using a combination of common sense and understanding of company norms and policies.

OAuth tokens granted to an AI agent carry the same access as tokens granted to a person, but the agent won’t understand that it’s been overpermissioned before it acts. It will simply do what it needs to do in order to execute the task.

The controls that matter here aren’t controls on the agent. They’re controls on the environment the agent operates in.

If you know where sensitive data lives across email and Drive, you can enforce policies that restrict access to it before an agent (or an attacker) gets there. If you’re investigating OAuth grants, you can understand and limit exposure to the prying eyes of an attacker or an errant agent.

If you can redact password reset links and require step-up verification before sensitive inbox content is readable, it doesn’t matter whether the entity trying to access that content is an attacker or an agent acting outside its intended scope.

The same coverage that defends against the modern attack chain also defends against the modern agent risk. They’re the same problem, wearing different hats.

What defense looks like across the full chain

I don’t think the answer is to add more point solutions to each stage of this chain. I think the answer is coverage that understands the chain as a chain, that can see what’s happening across email, OAuth, Drive, and account behavior, and connect the dots before things go wrong at step three or four.

That’s what we’ve built at Material. Here’s how our coverage maps to each step:

Blocking the initial email payload. Our email security is designed to catch what native controls miss: sophisticated phishing, payloads that bypass reputation-based filters, attacker-in-the-middle techniques. Stopping the most common attack method before it starts remains the highest-leverage intervention for the malicious threat.

Detecting suspicious OAuth behavior. Material goes beyond cataloging what apps exist and what scopes they hold.

The platform watches what apps actually do: what they read, when they read it, how that behavior changes over time. Whether an OAuth token is being used by an attacker or an AI agent operating outside its intended parameters, anomalous behavior at the activity layer surfaces the danger.

Detecting and protecting sensitive data at rest. You can’t protect what you can’t see, and you can’t design a policy around access you don’t know exists.

Material’s file security gives teams visibility into where sensitive data lives across email and Drive: which shared drives carry broad access, which email threads contain credentials or PII, which Drive folders are exposed beyond their intended audience. This is the foundation for enforcing least-privilege access against any actor, human or automated.

Blocking lateral movement via password resets. Material can redact sensitive message content, including password reset links, and require step-up verification before that content becomes accessible.

An attacker with inbox access can’t use it as a pivot point if the reset links aren’t available in plaintext. An AI agent reaching the inbox looking for something to act on encounters the same restriction.

The pattern is going to repeat

Vercel. Composio. I expect this list will keep growing, and I expect the next entries on it won’t always fit neatly into the category of “external attacker.”

Some of them will involve AI agents doing something unexpected. Some will involve overpermissioned integrations that reach data they were never supposed to see. The mechanism will look familiar even when the story around it doesn’t.

The right response isn’t to be alarmed about AI agents or to slow down adoption. Agents are genuinely useful and the productivity case for them is real.

The right response is to recognize that the workspace those agents operate in needs controls that are appropriate for a world where OAuth-authenticated software (authorized or not) is a first-class actor in your environment.

If your Google Workspace security strategy ends at the inbox, it has a gap. That gap is exactly where the modern attack chain runs, and it’s exactly where an AI agent operating outside its intended scope will run too.

If you want to talk through what full-chain workspace coverage looks like for your environment, reach out to us at Material Security.

Sponsored and written by Material Security.

It’s nearly two hundred years old.

BT’s 1846 roots took a few detours

Ilbusca/Getty Images

What BT looks like today

John Keeble/Getty Images

Just like any other online service, hackers can target and break into your accounts on popular AI platforms such as ChatGPT, Claude, and Perplexity. 

TechCrunch has created a comprehensive guide to help you protect yourself if you suspect someone has broken into your account on one of the internet’s most popular platforms, social networks, or messaging apps. Now, we’re here to show you how to check whether your accounts on AI platforms have been hacked.

As usual, we recommend using unique passwords stored in a password manager, and turning on multi-factor authentication (MFA), so that even if someone steals your password, they won’t be able to log in without that second piece of information. 

ChatGPT and Perplexity offer MFA. Claude doesn’t, because instead of asking for a password, Anthropic’s AI chatbot sends a login link to your email address.

All three of these AI platforms offer similar ways to check if there’s a suspicious device logged into your account. Here’s exactly how each platform works.  

ChatGPT

To find out if someone has broken into your ChatGPT account, open it on your computer’s browser, click on your username in the bottom left corner, go to “Settings,” then “Security and Login,” and finally click on “Active Sessions.” 

You will see where you are logged into your ChatGPT account. If you see any device you don’t recognize, you can log out of that single device. You can also click on “Log out all.”

At this point, if you want to change your password, you need to log out of your account. 

Then, on ChatGPT’s website, click “Log in” located in the bottom-left corner, enter your email address, click on “Forgot password,” and then “Continue.”  

ChatGPT will then send you an email containing a six-digit code. Enter the code on the ChatGPT login page, click “Continue,” and then enter a new password. 

You can also click “reset your password” in the email you received to see the official instructions on how to do that. 

Claude

For Claude, open it in your computer’s browser, click on your username in the bottom-left corner, then “Settings,” and click on “Account.” That’s where you will see your “Active sessions.” 

If you don’t recognize one of them, hover over it, click on the three vertical dots that appear on the right, and click “Log out” or “Terminate.”

If you want, you can click on “Log out of all devices.” 

At that point, you’ll be able to log back into your account using your email address. You will receive an email with a link to log in. Claude does not allow you to use passwords at all, so there’s no password to change.

Perplexity

In the case of Perplexity, the AI-powered search engine does not show you where you are logged in. 

So if you’re worried someone may have broken into your account, go to Perplexity in your browser and click your username in the bottom-left corner, then “All settings.” Finally, click on “Sign out of all sessions,” and then “Confirm.”

At that point, you can log back in by entering your email address. You will then receive an email with a unique six-digit code. Enter the code on the website to log in, or click on the “Sign in” button in the email to log in directly.