The compromised web. Abusing chatbot content sharing. Meta's AI support goes rogue.
The threat to the US midterms. When anonymous is anything but. And what not to name your Bluetooth speaker.
Web of deceit
One in ten of the nearly 85 million new domains registered in 2025 were linked to malicious activity and were eventually added to one or more cybersecurity blocklists, according to research by Interisle. The actual figure may closer to one in five because many of the domains remain to be deployed and identified.
This matters because trust on the web is breaking down. AI enables scammers to create convincing looking websites with little effort and almost no cost. But they’re also hijacking legitimate sites to deliver fake browser updates and other attacks. SilentPush identified thousands of high-reputation sites that had been compromised. The new normal is for cybercrime to be woven into the everyday fabric of the web.
So what to do?
-
Don’t assume websites are trustworthy because they look professional. Attackers increasingly use legitimate platforms, compromised sites and realistic domains.
-
Don’t search for software downloads. Go directly to known vendors or bookmarked links.
-
Treat browser update prompts with suspicion. Modern browsers update automatically. Unexpected prompts are increasingly used to deliver malware.
-
Use a password manager and MFA. “Infostealers” are designed to harvest saved credentials and authentication cookies, password managers mitigate this threat.
-
Keep devices updated. Many campaigns rely on old vulnerabilities or security bypasses.
-
Be suspicious of requests to paste commands into Terminal, PowerShell or Command Prompt. This is the tactic known as ClickFix and it’s increasingly common.
Threats
Sharing: Attackers are increasingly abusing content-sharing features on platforms such as ChatGPT and Claude to distribute malware. The approach allows malicious content to be hosted on trusted domains such as chatgpt.com and claude.ai, making it more likely users will trust links. Traffic is then driven through search adverts promoting installation guides and software updates. Push Security
Signal: A new campaign is targeting users to try to steal backup recovery keys. It begins with a message purporting to come from Signal Support, warning about a “sync issue” and the risk of permanently losing account data. Signal will never ask for any codes, and unsolicited messages from any “Support” account should be treated with suspicion. Malwarebytes
Unrest: Geopolitical unrest provides fertile ground for scammers. Whether it’s conflict in the Middle East or disease in Africa, current events make for ideal lures. ESET has guidance about the (extensive) range of scams being seen. Fake charities and travel-related fraud are particular threats at the moment.
AI
Meta: Why bother with the palaver of phishing emails and fake websites when you can just ask AI support to do your work for you. 404 Media has the wild story of hackers who say they used Meta’s AI support chatbot to break into a host of high-profile Instagram profiles. All they had to do was ask it to change the email address associated with the target account. The hackers’ claims coincide with a series of high-profile Instagram account takeovers.
Manipulation: Also from 404 Media comes news of an internal Microsoft strategy document that says it aims to make people addicted to its new “Scout” personal assistant. Scout was launched this week and is designed to bring autonomous AI functionality to non-technical users. In general, AI chatbots are designed to keep users engaged, hence conversations that would go on forever if you let them. Research published this week identifies the features that underpin the approach - and their potential risks to privacy and well-being.
Inevitable: “Resistance is futile,” was the message from chipmaker Qualcomm’s CEO at a conference this week. The era of apps and even smartphones as the centre of digital life is ending, he said, to be replaced by AI “agents” that quietly follow users across phones, PCs, glasses and other devices, carrying out tasks in the background. The pitch is that these agents will become so seamless they effectively disappear from view and become impossible to avoid.
Deaths: Florida is suing OpenAI, accusing it of being a threat to public safety. The suit follows a mass shooting at Florida State University in which two people were killed. It claims OpenAI’s “careless introduction” of ChatGPT has “aided and abetted deadly rampages” and “encouraged suicide.” OpenAI has responded by pointing to measures implemented to protect minors. BBC
Trump
Fakes: A growing war over “reality itself” is the biggest threat to the US mid-term elections in November, according to Check Point. It warns that phishing, fake news sites, credential theft, domain impersonation and AI-generated content are increasingly being used to undermine trust in the elections. The key risk, Check Point argues, is less that attackers change the result of the elections and more that they convince enough people the result cannot be trusted.
Order: President Trump has signed a scaled-down executive order on cybersecurity and AI oversight. The order creates a voluntary system which asks AI companies to give the US government early access to new models up to 30 days before release. The administration says the aim is to balance national security with maintaining the US lead over China. Critics say AI industry lobbying appears to have prevailed over more onerous requirements.
In brief
Exposed: A supposedly anonymous video chat service called FTF Live exposed data from more than 22 million user sessions after researchers discovered publicly accessible monitoring systems left online. The leak reportedly included details affecting some 3.5 million identifiable accounts. It underlines the reality that “anonymous” services often collect information which can reveal a surprising amount about their users. Cybernews
Dashlane: The password manager outfit temporarily suspended some customer accounts after detecting repeated attempts to brute-force its two-factor authentication process. The company says no systems were compromised and fewer than 20 encrypted vaults were downloaded. Some users criticised the response, arguing that attacks against 2FA should ideally be rate-limited without locking out legitimate users.
Alert: Android is introducing a new feature designed to combat AI voice scams and caller ID spoofing by warning when someone appears to be impersonating a contact they know. It works by checking whether calls from contacts actually originate from that person’s device — and shows a warning if the digital “handshake” is missing. That means it currently only works when both parties use Google’s phone app.
Iran: An Iranian hacktivist group is recruiting individuals for physical attacks, espionage, sabotage, and assassinations. According to Recorded Future, the Handala group is behind four new online personas that recruit people for attacks in Israel and the US. The group has promised rewards of up to $500,000 for attacks on selected individuals. Previous reporting has linked the Handala group to Iran’s MOIS intelligence agency.
Russia: The government has greatly expanded the personal and technical details that mobile operators and internet service providers must collect from customers and share with state authorities. The data now includes home addresses, passport information, tax IDs, bank account details, and geo-location coordinates.
Screens: Staff at a Liverpool organisation say they were disciplined after managers allegedly accessed WhatsApp messages left open on workplace computers while they were away from their desks. The details of the case are disputed but the risk of leaving screens unlocked is not. Liverpool Echo
Names: Our standing advice is not to give laptops and smartphones your real name (because that’s what anyone can see when you join a network). We hadn’t given much thought to Bluetooth speakers, until this week when we saw news of a United Airlines flight that made an emergency diversion after someone spotted a Bluetooth device called “BOMB” on board. It turned out to be a teenager’s speaker. NPR
Updates
Microsoft: From next month older Macs and iPhones will no longer be able to use Microsoft 365, Office 2019, and Office 2021 to edit documents. Microsoft suggests affected users turn to free Microsoft 365 web apps, take out a paid Microsoft 365 subscription, or make a one-time purchase of Office 2024.
Apple: iOS 26.5.1 is designed to address a charging problem affecting iPhone 17 models.
Canon: Firmware updates for more than 200 enterprise printers to fix a major security issue that could have been used to access sensitive credentials.
Tails: Emergency release to fix a serious security vulnerability in the Linux kernel, as well as security vulnerabilities in the Tor client.