AI and disinformation...and the risk to personal data. Dangerous VPNs.
The scale of cybercrime. The threat a cheap USB cable (with more memory than the moon landing computer) .
False
The leading AI chatbots are now regurgitating false information on topics in the news more than a third of the time, according to an audit by NewsGuard. That’s nearly double the rate seen in August 2024 and includes false claims on topics such as health, politics, international affairs, companies, and business brands. A key cause appears to be that AI chatbots now try to answer questions on all topics, even controversial ones or real-time news events. NewsGuard says last year they refused to answer 31% of such prompts. This year that fell to zero.
Worse, the data the chatbots are using is increasingly polluted by malign actors, including Russian disinformation operations. There have been repeated warnings that Russia is seeding falsehoods in support of its international and political ambitions. One study by the American Sunlight Project estimates that Russia is producing at least 3 million propaganda articles per year. NewsGuard says in the race to make AI chatbots work in real-time concerns for safety and accuracy have been ignored.
Threats
VPNs: Free tools are to be avoided whenever possible - and this applies particularly to VPNs. New analysis reveals that while many popular free VPNs market themselves as independent, in fact they share the same source code as well as serious vulnerabilities. The research, by US academics, examined VPNs which have been downloaded from the Google Play Store over 700 million times.
Facebook: Watch out for fake security alerts saying that a Facebook account is about to be suspended. They look pretty realistic but in fact are the first step in a chain designed to steal information and install malicious software. The key part of the process is persuading the victim to copy and paste a command into Windows File Explorer which, of course, we should never do. Acronis
SMS: The Swiss cybersecurity agency is warning about “SMS blasters” which are used to send malicious text messages to cellphones within their vicinity. The agency says the messages purport to be notices of parking fines and recipients had all been in the specified locations. We see more and more reports of these systems being used around the world, often with apparent connections to Chinese organised crime groups.
Physical: Although we focus on cybersecurity, our awareness courses for high risk deployments also cover physical threats. Given the febrile nature of public discourse this is more important than ever, as evidenced by an incident in Utah in which an incendiary device was placed beneath a news media vehicle. The device had been lit but “failed to ignite.” Two men have been arrested in connection with the incident. The vehicle belonged to a Fox affiliate station. Fox 13
AI
AI: A reminder to be very careful about giving AI tools access to personal data. Examples of this include asking ChatGPT to look at a calendar to summarise the day’s activities. The problem is that the calendar could include a malicious invite designed to steal information from linked apps such as email. Google advises using the “known senders” setting to mitigate this threat but unfortunately the problem extends further than this. As AI guru Simon Willison puts it, “Any time you ask an [AI] system to summarise a web page, read an email, process a document or even look at an image there’s a chance that the content you are exposing it to might contain additional instructions which cause it to do something you didn’t intend.” This is called “prompt injection,” and a security firm has just demonstrated how it used the technique to persuade a ChatGPT-integrated agent called Deep Research to obtain confidential information from Gmail. There’s no question about how useful AI tools can be but, as they’re given more autonomy, there’s an increasing risk of them being manipulated to do something harmful. Ars Technica
Tricked: Together with a Harvard researcher, Reuters set out to create the perfect phishing scam. They found that major AI chatbots were happy to help. The result was an email inviting senior citizens to learn about a new charity dedicated to providing the elderly with care and companionship. It was tested on 108 elderly volunteers. 11% of them clicked on links in the emails they received. Reuters’ long-read is an excellent summation of how easy it is to circumvent the controls that are supposed to stop the chatbots doing this sort of thing.
LinkedIn: Microsoft’s employment platform is changing its User Agreement and Privacy Policy to expand the number of countries where it can use member data to train its generative AI models. This setting will be on by default when the change comes into force on November 3. We suggest users opt out before then by going to Settings | Data Privacy.
In brief
Android: A vast advertising fraud operation has been taken down, but only after it managed to generate up to 2.3 billion clicks per day. The operation involved domains that impersonated games and news sites and operated through more than 220 apps in the Google Play store. Satori
Kit: Meanwhile, Microsoft has seized 338 websites associated with what it calls the “fastest-growing tool used by cybercriminals to steal Microsoft 365 usernames and passwords.” The RaccoonO365 Suite of phishing kits was available at prices that began at $335 for a 30-day subscription.
Arrests: UK police have arrested two teenage members of the Scattered Spider hacking group. Thalha Jubair and Owen Flowers are 19 and 18 respectively and are accused of being responsible for last year’s attack on Transport for London. Jubair has also been charged in the US for hacking 47 companies and successfully extorting ransoms of at least $115 million.
Compunds: Operators of scam compounds appear to be setting up in new countries in response to sanctions against their activities in Myanmar and Cambodia. The UN says it found “indications of scam centre activity, including SIM cards and satellite internet devices” in Timor-Leste, a nation whose gross domestic product was $1.88 billion last year.
Romania: The country’s communications regulator says XTwitter has refused to cooperate in action against Russian propaganda on social media platforms. The regulator said it sent more than 400 takedown orders this year and most of them were carried out with the exception of XTwitter. Stirile Pro TV
Memories: A USB-C to headphone converter on sale for €3 through Ali Express has more memory that than the Apollo Guidance Computer used in the first moon landing. Predictably, the headphone converter is also ideal for being repurposed as a means to attack a device to which it’s connected. The jargon for this is a “USB Rubber Ducky.” It sounds silly but it’s a real threat, particularly in high-risk environments where it’s important to be very careful about what is plugged into any device.
Fridges: Samsung is to start using its Family Hub fridges in the US to display adverts, despite insisting in April that it had “no plans” to do so. The Verge
Updates
Apple: This week brought a series of major updates to the iOS and macOS platforms. iOS/iPadOS 26 and macOS Tahoe 26 include important security updates, but they also represent significant changes to the look and operation of the platforms. For that reason, you might wish to wait a little while before installing them. Major changes like these are invariably accompanied by initial problems and it’s probably better for someone else to find out what they are. If you don’t install these versions, security fixes are also available for their predecessors.
Chrome: Version 140 fixes a serious vulnerability that has already been used to attack Google’s web browser.
Firefox: Version 143 of Mozilla’s web browser has important security updates along with new features that will mean we spend even more of our online time inside it.
Microsoft: The Copilot app will be force-installed on all Windows computers with effect from the start of October. Unless the computer is in the EU where it remains banned because of privacy regulations.
SonicWall: Customers have been told to reset their credentials after backup configuration files were exposed in a security breach that affected MySonicWall accounts.
Tails: Version 7.0 has multiple app upgrades and adds support for newer hardware including graphics and WiFi.