Sucking up our data. Abusing trust. AI gets personal. Spyware alerts.
And turning internet gateways into security sensors.
Greedy
A study of mobile apps reveals the voracious appetite of big tech companies for data about their users. The research by Surfshark analysed the privacy information submitted for 171 apps to Apple’s App Store, covering data including precise location, browsing history, and purchase details. It found Meta’s apps collected on average three times as much data as those from Apple and Microsoft, with Meta AI recording 33 data types. “Our research shows that the Big Tech business model is often built on an insatiable appetite for user information,” Surfshark said.
In an effort to shed light on how the advertising ecosystem tracks us, a new tool has been launched which traces and identifies the adverts running on a particular website - and where else they appear. DecryptAds pulls together data which has previously been difficult for consumers to access but which is bought and sold by data brokers and can be used to track people both online and in the physical world. As Krebs on Security reports, using the tool to research espn.com reveals 143 advertising partners and 19 registered data broker domains. Almost half of those brokers are collecting geolocation data from visitors who don’t block ads, while another three disclose that they collect device fingerprints and sensitive personal information.
The two reports tell the same story from opposite ends: Surfshark shows how much is scooped up, DecryptAds shows where it goes. Put together, they define an ecosystem so sprawling that “consent” to the use of our data is effectively meaningless. There’s one way to regain some control. A reputable content-blocker won’t stop every form of tracking, but it will absolutely will limit the fuel on which brokers rely.
Threats
Trust: Attackers are increasingly focusing on hijacking things we already trust, rather than trying to infect victims with malicious software. In its half-year threat report, Gen Digital says it saw sharp increases in scams exploiting hotel booking systems, messaging accounts, browsers, online shops and AI tools. Collaboration platforms like Slack are also a target. Palo Alto Networks says alerts of malicious activity have more than quadrupled to nearly 6,800 a month. The underlying message is clear, even if you recognise the sender, it doesn’t mean the message should be trusted.
Bypass: It’s depressing, but basic multi-factor authentication can no longer be viewed as a reliable defence. In a recent incident, an employee in a company’s finance department clicked on a convincing HR-themed email and approved a Microsoft MFA request. Attackers were able to steal the resulting authorisation “session token,” giving them access to email without further authentication. They then spent weeks monitoring messages, creating rules to hide their activity, and ultimately exploiting genuine information to divert vendor payments. As well as having cast-iron checks for payment changes, phishing-resistant solutions such as passkeys are now essential.
Retirement: Scumbags love old domain names and they’re snapping up tens of thousands of them as soon as they become available. As Infoblox reports, there’s even an auction service called dropcatch.com to facilitate (and monetise) the process. The rationale is simple: unless it’s retired gracefully, an old domain can retain its reputation, incoming email, links from other websites and existing traffic, making it much more useful than a newly created one.
AI
Personal: AI assistants want to move out of a chat window and make deeper connections into our work. Take the ChatGPT Mac App for example. If you allow it, a feature called Computer History will record activity across selected apps and websites so ChatGPT can remember what a user has been doing and pick up where they left off. Meanwhile, Meta AI’s beta Mac app can see shared windows, take dictation, and connect to services including Facebook, Instagram and Google Workspace. Enabling these features can turn chatbots into genuinely useful personal assistants, but doing so is not risk-free because of the possible exposure of sensitive information, especially in the event of an account being compromised.
Gemini: On that subject, by default, Google exposes an organisation’s Workspace data to Gemini AI. Again, that could be useful - but could also expose sensitive data that should remain confidential. ZDNet has a useful guide on managing Gemini access and reducing that risk.
Infighting: Some new research from Anthropic provides valuable insights into the challenges facing the AI developers. In one experiment, Anthropic deliberately put Claude agents with conflicting objectives into the same environment. The consistent result was a “turf war” in which the models assumed others were deliberately impeding their work. They reacted by deploying “increasingly aggressive, self-replicating malware” in an attempt to sabotage their peers. This was just one of the findings of the research which, as Anthropic says, underlines the difference between human social systems built up over millennia and large language models which might understand that framework but lack the “disposition” to apply it.
Court: A person representing themselves in a Connecticut court hid text in a filing in an abortive effort to manipulate AI and influence his case. The attempt failed on two counts: officials spotted the document’s unusual formatting and, more importantly, the court doesn’t even use AI. 404 Media
In brief
Spyware: An unprecedented number of Apple customers have reported receiving a recent threat notification alerting them to suspected spyware attacks targeting their devices. Access Now, which provides digital security support, told TechCrunch the number of people requesting help following the notifications was 30% to 40% higher than usual. Among those alerted was a Ukraine Armed Forces soldier who said he wouldn’t have thought he was important enough to be targeted. A senior researcher at The Citizen Lab said the reports show that spyware attacks may be more prevalent than people realise. The good news is that, for the loss of some convenience, enabling Lockdown Mode appears to provide considerable protection. Anyone at high risk should turn it on.
Spies: You might remember the Encrochat affair, in which French intelligence hacked the encrypted phone network favoured by organised criminals (leading to more than 6,500 arrests). France has repeatedly refused to disclose its method, citing national security. Now, researchers have reverse-engineered the secret implant and concluded that parts of it were copied from code freely available on GitHub. The implant exploited an Android security vulnerability that went unpatched for two years and potentially affected 2.5 billion phones. It allowed the government hackers to take complete control of infected devices and copy or change users’ data, programs and files at will. But that’s just one headline from an ongoing saga which is covered in detail by Computer Weekly. It reads like the plot for a movie.
Journalists: Israel’s security services have warned reporters in the country that Iranian state hackers have intensified targeted phishing attacks against them. The campaign appears to be designed to obtain private information from journalists covering political and national security.
Reversed: Surveillance outfit Flock Safety has long insisted that its cameras “cannot recognise, identify, or track individuals.” Well, they can now, according to an investigation by WIRED. It analysed software found on Flock’s website that can identify drivers and track their vehicles through their movement patterns. Flock says it’s testing the product and describes it as still in development.
Monitor: US internet service provider Comcast is turning its newer Xfinity broadband gateways into home security sensors, using disturbances in WiFi signals to detect people moving around the house. The technology itself isn’t new, but this is the first time an ISP has deployed it via its routers. In the UK, Virgin Media’s Hub 5 uses closely related hardware and software, though it hasn’t published any plans to follow Comcast’s lead.
Collateral: As well as attacking Wildberries’ physical infrastructure, Ukraine says it has also disrupted the IT systems of Russia’s main online retailer. The Ukrainian attacks demonstrate that in the event of a conflict, civilian organisations risk being dragged onto the digital battlefield.
Updates
Apple: Security updates for iPhones, iPads, and Macs to address 28 vulnerabilities. New versions include iOS 26.6.1, iPadOS 26.6.1, macOS Tahoe 26.6.2, and iOS 18.7.10/iPadOS 18.7.10 for older devices. The updates include a fix for a vulnerability in the way images are processed, a type of issue previously exploited by “zero-click” spyware that requires no interaction to infect a device. The need to apply security updates as soon as possible was underlined this week by news that a security company used an AI agent to build working exploits for two previous macOS vulnerabilities. It says the work took just four hours.
Browsers: We’ve always said it’s worth closing down web browsers at the end of every day to make sure they stay up to date. Given the pace of updates, that’s now a fundamental security requirement. This week, both Chrome and Firefox addressed dozens of vulnerabilities, many of them high-severity.
Microsoft: Fix for an issue that caused Windows Defender to crash after a recent security update. Affected systems displayed error messages saying “Threat service has stopped.”
Oracle: August 2026 Critical Security Patch Update includes 943 security fixes addressing newly disclosed vulnerabilities across its enterprise software portfolio.
Tails: Version 7.11 updates Tor Browser and fixes problems with persistent storage.