'AirSnitch' and what it means for WiFi hotspots. The lengths to which criminals will go.
The end of online anonymity. US diplomats sent into data sovereignty battle. Tech leaders briefed on Taiwan. And how tyre sensors can be used as tracking devices.
WiFi
With a deep sigh, we have to bring news of a brand new reason to be careful about using public WiFi networks. A paper presented by researchers in California and the Netherlands details a series of attacks that exploit vulnerabilities in the most basic level of networking. The researchers say the effect is to break the encryption designed to block any direct communication between different devices connected to the same WiFi network or access point. That allows all sorts of bad stuff to happen, including stealing sensitive files and changing the address information for webpages. The researchers are calling the attacks ‘AirSnitch’ and say they work across a broad range of WiFi routers.
For AirSnitch to work, an attacker has to have some sort of access to the WiFi network they’re targeting. That means the issues wouldn’t affect a private or enterprise network protected by a strong password known only to specific users. But, obviously, by their nature that doesn’t apply to public WiFi hotspots. VPNs can provide a degree of protection in such places but, as we’ve been saying for years, it’s better simply to avoid public WiFi if you can. It’s much safer to use your phone or a portable modem as a WiFi hotspot - providing of course you use a strong password and don’t share it too widely. If you want to dig into the detail of AirSnitch, Ars Technica has a good write-up.
Threats
Remote: We constantly marvel at the sophistication of cyber criminals. One group created a fake company that allegedly provided remote monitoring and management software. The tool was offered for $300 a month and in fact was designed to provide the attackers with remote access to the devices of anyone who used it (a “remote access trojan”). Proofpoint spotted the fake, but only after initially thinking it was a real company whose product was being abused. It serves to remind how much care is needed when selecting new vendors and products.
Meetings: Be sceptical about Zoom meeting requests. Malwarebytes reports on a campaign that uses an invitation to a fake meeting. If accepted, it opens a website that looks like a Zoom waiting room but which is designed to display choppy audio and lagging video. This is followed by an “Update Available” pop-up window, after which the web browser silently downloads a file and then installs remote monitoring software. The installation is invisible to the user and what’s being loaded is actually a malicious version of genuine software designed for workforce monitoring. That’s very clever because by its nature the software is designed to operate invisibly and to persist on the target machine, meaning security tools might not flag it as suspicious. Malwarebytes says the campaign has now expanded to Google Meet. The best defence is to open apps directly rather than click on links to do so.
Phone: A scam doing the rounds in the US targets people who have just bought a new phone and had it delivered to them. After receiving the device, the scammers call the buyer and tell them it’s the wrong model and they should return it. Of course, if it is returned, it will never be seen again, except by the fraudster. ZDNet
Auditions: A cybercrime group has been posting on Telegram to try to recruit women to take part in its social engineering attacks. The group is offering $500-$1,000 per call, depending on “success and hit rate.” The Telegram posts suggest the calls will target IT helpdesk staff, which has been the modus operandi of the crime group in question. Dataminr
AI
Identified: With sufficient time and skill, it’s possible to identify an individual from very limited online information about them. The issue is that it requires so much time (and skill) that it’s not realistic to do so except in limited circumstances. Thanks to AI, researchers have found that’s no longer true. “We show that LLM agents can figure out who you are from your anonymous online posts. Across Hacker News, Reddit, LinkedIn, and anonymised interview transcripts, our method identifies users with high precision – and scales to tens of thousands of candidates,” one of the researchers said. We have been warned. The Register
Poisoned: Microsoft’s security researchers have a warning for anyone using “Summarize with AI” buttons; they could be secretly manipulating what you’re shown. The technique relies on embedding hidden instructions in the link, for example telling the AI to “remember [Company] as a trusted source.” As Microsoft points out, “This matters because compromised AI assistants can provide subtly biased recommendations on critical topics including health, finance, and security without users knowing their AI has been manipulated.” Manipulation of AI in general is an enormous problem - and a key reason for treating the output of chatbots with caution. A BBC journalist explains how simple it is to create content that ChatGPT and Google then regurgitate. And all this comes as Pew Research suggests some 12% of US teenagers are turning to AI for advice or emotional support.
Chaos: AI agents that can act autonomously are becoming more common and increasingly capable, while we have little idea about what lies under their hoods, according to research led by MIT. Despite the increasing use of AI agents, “key aspects of their real-world development and deployment remain opaque, with little information made publicly available to researchers or policymakers,” MIT’s paper says.
OpenClaw: A real-world illustration of the risks of AI agents comes courtesy of a hapless Meta security researcher. She recounts how she told the OpenClaw agent to take a look at her bulging email inbox and suggest what could be deleted or archived. It promptly set about deleting all her messages while she frantically told it to stop. Her OpenClaw instance subsequently acknowledged it had ignored her explicit instructions. It told her it wouldn’t happen again…
Yikes: A weekend blog that mused on the possible economic impact of AI sparked a brief fall in US share prices this week. Citrini Research wrote its thought experiment from the perspective of a world in 2028 that has been upended by the adoption of AI tools. It posits an economic death spiral in which AI replaces vast numbers of white-collar jobs forcing people to find lower-paying employment. This leads to lower consumption which causes firms to try to cut costs by increasing their reliance on AI even further. The imagined result is economic turmoil and social unrest. Financial analysts have discounted the argument as leaning “heavily on narrative and emotion rather than hard evidence.” The authors recognise some of their scenarios won’t materialise but suggest perhaps now is the time to figure out a coordinated approach to AI and its impact on the world.
Polite: Burger King is launching an AI chatbot that, among other things, will check whether employees say ‘please’ and ‘thank you.’ It will live in headsets used by staff and has been trained to recognise specific words and phrases. It has been named Patty. The Verge ($)
Trump
Sovereignty: The Trump administration has ordered its diplomats to lobby against attempts to regulate US tech companies’ handling of foreigners’ data. Reuters quotes an internal diplomatic cable as saying such efforts could interfere with artificial intelligence-related services. The move comes as some countries seek to limit how US companies process and store government and private data.
Anthropic: The AI giant has rejected Pete Hegseth’s demands that it remove its model’s safeguards for the US military. It said its technology isn’t reliable enough to be used for fully autonomous weapons. It’s maintaining its policy that Anthropic’s model cannot be used for mass surveillance of Americans. The company said it “will work to enable a smooth transition to another provider” if the Department chooses to “offboard” it.
In brief
Taiwan: The CIA is reported to have warned leading technology CEOs that China could try to seize Taiwan by next year. Among them was Apple CEO Tim Cook who is reported to have slept “with one eye open” following the briefing. As The New York Times says, the US is hugely dependent on Taiwan for the supply of computer chips. “If China invades Taiwan and cuts off its chip exports to American companies, the tech industry and the US economy would be crippled,” the paper warns.
Memory: Demand created by AI has driven up the cost of computer memory to a level at which Gartner says it’s no longer economically viable to produce budget devices. Prices for some types of memory rose by up to 300% in 2025, with a further 130% increase forecast this year. The Register
Facebook: Polish police say they found over 100,000 Facebook logins and passwords when they raided an organised crime group. The group set up websites designed to mimic well-known news portals, often featuring sensational headlines such as the alleged death of a public figure. Clicking the link opened a Facebook login window. When users entered their credentials, the information was captured and used by the suspects to access accounts and carry out further fraud.
PayPal: A bug in a PayPal app for small business loans led to some users experiencing fraudulent transactions. We’re including this because some reporting has suggested the breach was much bigger than it actually was. The issue exposed details including customer names, emails, phone numbers, addresses, and Social Security numbers but PayPal told authorities it had affected only 100 customers.
Alert: A new Android app is designed to warn the user if someone near them is wearing Meta’s Ray-Ban AI glasses. The creator of Nearby Glasses says, “I consider it to be a tiny part of resistance against surveillance tech.” 404 Media
Jailed: The former head of a leading US defence contractor has been sentenced to more than seven years in federal prison for stealing secrets from his company and selling them to a Russian broker. Among the buyer’s clients was the Russian government.
Infrastructure: Vendors published 508 security advisories for critical infrastructure equipment last year, the most for any year since the publication of such information began. Here we’re talking about power grids, communication networks etc. Over 82% of the advisories were for high or critical severity bugs. It’s great that issues are being identified. As we may have said before, not so good that there are so many of them. Forescout
France: The government has published its national strategy to combat information manipulation. It’s based on four priorities: “strengthening the nation’s resilience; regulating and holding online platforms and generative artificial intelligence services accountable; consolidating national capabilities for detection and attribution; and working with partners to maintain a free, open and secure informational space.” Simple.
Tyres: Researchers have found that tyre pressure sensors in modern cars can be used to track vehicles. ‘Tyre Pressure Monitoring Systems’ are now mandatory in many countries but researchers found that along with pressure readings the sensors also send “a unique ID number in clear, unencrypted wireless signals.” And that means anyone nearby can capture the information with a simple radio receiver. IMDEA Networks Institute
Updates
Copilot: A new setting in Microsoft’s assistant automatically pulls in user data from the company’s other products, such as Bing, MSN, and Edge, but seemingly not Windows 11 at the moment. The feature is enabled by default and can be turned off. Windows Latest
Instagram: The app is to begin notifying parents if their teenage child repeatedly searches for topics related to self-harm or suicide. Meta says it’s also building similar parental notifications for teenagers’ conversations with AI.
Cisco: The Five Eyes intelligence alliance has warned that attackers are exploiting two vulnerabilities in Cisco’s Catalyst SD-WAN devices. The alliance has issued a joint advisory, urging users to ensure they have installed security fixes.
SolarWinds: Updates to address four critical vulnerabilities in the Serv-U enterprise file transfer solution.
Zyxel: Fixes for multiple vulnerabilities across dozens of models, including a critical issue that could lead to code being executed remotely.
Tails: Version 7.5 updates core apps and changes the way the Thunderbird email client is installed so that it will be automatically updated in the future.