Cockroach spyware. AI-enabled social engineering. A new protected mode for ChatGPT.
Total cyber attacks down, but ransomware up. And US police abuse of surveillance tools.
Spyware
Despite a court injunction, Meta says the maker of Pegasus spyware is continuing to use WhatsApp as a means to infect targets. The latest campaign, which Meta says it disrupted, was less sophisticated than previous “zero-click” attacks which required no user interaction. Instead, it was a highly-targeted (“spear-phishing”) operation that tried to lure people into clicking a link sent to their WhatsApp accounts. Meta has filed a complaint against the NSO group, asking it to hold the company in contempt. We doubt it will have much effect.
The reality is that spyware is almost impossible to control, and its widespread use by governments means that in many cases there’s no incentive to do so. In the case of NSO, despite losing many of its staff and most of its contracts, it is clearly still operational and has been busy lobbying lawmakers in Washington DC to have sanctions against it lifted. Citizen Lab has spent years investigating “mercenary spyware” and its use against journalists, opposition figures, activists and human rights defenders. Citizen Lab’s director has called the expansion of the commercial surveillance industry an escalating threat to democracy. Spyware remains a significant threat to individuals whose work governments dislike.
Threats
AI: There’s a steady increase in the use of AI-enabled social engineering to circumvent technical security measures, according to research by Visa. It says it identified nearly $1 billion in scam activity from July to December last year. As an example, two current campaigns are using AI-generated TikTok videos and Instagram Reels to direct users to sites designed to deploy malicious software. Reversing Labs says they offer fake technical advice or tutorials on using premium software for free. Both use professional-sounding voice-overs and clean graphics. AI makes it worryingly simple to create campaigns like these and it’s vital to be on guard against them.
Fakes: Another use of AI involves inserting fake retail sites into ChatGPT shopping results. Scam-checking service Ask Silver says it found ChatGPT linking to fake versions of a UK shoe retailer and a home furnishing chain. The counterfeit sites use domain names that look plausible at first glance. They take orders, deliver nothing, and steal payment information. The giveaway is that they also offer improbably high discounts. Ghacks
Stealer: A phishing kit previously linked to Microsoft 365 account theft, has expanded its reach to target AWS, Okta, Xerox and messaging platforms.
Users should be wary of unexpected login prompts, emails claiming account issues or document access requests, and pages asking for MFA codes immediately after sign-in. If something feels slightly off, go directly to the service rather than following the link. Arctic Wolf
Meta: UK banking giant Lloyds says more than two-thirds of fraud cases reported by its customers started on Facebook, Instagram and WhatsApp. People in their late twenties and early thirties are the most at risk, with the most prevalent scams involving fake tickets for concerts, festivals and sporting events. Lloyds said consumers were sending about £66 million a year to fraudsters after falling victim to a scam advert via Meta, up from £27 million in 2023. Sunday Times
AI
ChatGPT: OpenAI has begun rolling out a new security setting designed to protect against “prompt injection.” (This is a tactic that involves hiding malicious instructions on web pages and other sites to try to trick chatbots into doing something they’re not supposed to.) Lockdown Mode is designed to stop sensitive data being exfiltrated in the event of a malicious prompt being activated. OpenAI says it limits some functionality and is intended for users with an “elevated risk profile due to who they are, what they work on, or the types of data they work with.”
Cutdown: Anthropic has released a new version of its Mythos large language model which is designed for general use and has guardrails designed to block responses in specific high-risk areas of cybersecurity and biology. Early reactions suggest the restrictions are indeed extensive, with users reporting that it refuses to answer even basic biology questions. Despite the precautions, new research suggests the real problem may be that any set of guardrails can be broken - it’s just a question of finding the prompt that does so.
History: Google is rolling out changes to its search functionality that will mean content uploaded may be saved and used to train its AI models. Examples include images, files, and audio/video recordings, such as Google Lens images and voice searches. There is an option to turn this off.
TV: More evidence that smart TVs - and other internet-connected devices - are being exploited to gather content from the public web to provide content for AI training. A detailed report from Include Security explains how the devices are put to work without the explicit consent of their owners. The company involved even advertises the AI web scraping features on its website. This doesn’t necessarily harm the owner, but it does mean many of us are part of the AI training ecosystem without knowing it.
Trump
Whistleblower: WIRED has a wild story about a federal employee who filed a complaint about the so-called Department of Government Efficiency (DOGE) and days later found the brakes of his car sabotaged. The employee had claimed DOGE exfiltrated data from the National Labor Relations Board, which was then accessed from Russian IP addresses using valid credentials.
In brief
Overview: The number of cyber attacks recorded globally fell slightly in May, but that’s not as reassuring as it sounds. Check Point found ransomware attacks rose by almost 50%, suggesting criminal groups are concentrating on fewer but more profitable targets. Rather than casting a wide net, attackers appear to be focusing effort where disruption creates pressure to pay. For organisations, the message remains familiar: secure backups, install updates, and enforce strict access controls.
Outrage: Social media was supposed to connect people. Increasingly it rewards outrage, performance and whoever can hold attention for longest. Data guru Nate Silver argues that platforms have evolved into a kind of digital freak show where the loudest, strangest and most extreme voices are promoted because they keep people scrolling. As he points out, on XTwitter, the “Catturd” account gets far more engagement than the New York Times.
Cameras: The Financial Times reports that Russia’s security services shut off parts of a camera surveillance system protecting President Putin. After inspecting the system, only devices without an internet connection were reconnected. The move came after Ayatollah Ali Khamenei was assassinated in Tehran at the start of the current war with Iran. He is reported to have been killed in his office when the Israel air force attacked his compound.
GPS: Evidence suggests that for the past twenty years the US has been using the public GPS network to broadcast codes for its global encryption network. A University College London professor analysed a forgotten part of the GPS signal and concluded the data contained in it was most likely an elegant way of distributing codes to GPS receivers around the world - with only military devices able to decipher them.
Stalkers: 404 Media examines the endemic misuse of surveillance systems by police in the US. The report includes multiple instances of officers using vehicle-tracking systems to monitor people for personal reasons rather than law enforcement. The system providers insist they have safeguards to prevent this kind of misuse. The evidence suggests they’re not working very well.
Updates
Deluge: Microsoft and Adobe released a record number of updates to address vulnerabilities across their products. With over 200 vulnerabilities from Microsoft and 123 from Adobe, the update process has become an exercise in triage. More worryingly, it marks a steady upward trend in the number of issues; the number of vulnerabilities (“CVEs”) disclosed by Microsoft this year exceed the total for the whole of 2018. Given the sheer number of updates, it’s vital to prioritise internet-facing systems, externally accessible services and machines used to open files from outside the organisation. Indeed the US cybersecurity agency has just updated the rules for federal civilian agencies on applying updates. The new order cites the rise of AI-automated attacks as the key reason to prioritise issues based on the risk they pose. In some cases, updates will have to be applied within three days.
WhatsApp: Support for multiple accounts is finally rolling out to iPhones. Users can check whether the feature is available to them by going to Settings > Account and looking for an “Add account” option. Just note that if you do add an account, you will have to relink any other devices connected to WhatsApp.
Patch Tuesday: There were also monthly security updates from Adobe, Chrome, Opera, AMD, NVIDIA, Cisco,SAP,IBM,HPE,Dell,Fortinet,TP-Link, and Zoom.