Phishing: voice trumps email. The abiding risk from browser extensions.
AI getting smarter by the day, but still can't tell the time. And the return of the landline.
Audio
Voice-based phishing has replaced email as the most common social-engineering mechanism for compromising victims, according to research by Google’s Mandiant subsidiary. The reason is simple. Email scams rely on volume and luck, while so-called “vishing” involves trained operators building trust in real time, making it far harder to detect and block. In practice, attackers are increasingly calling staff directly, impersonating colleagues or IT support, and persuading them to hand over access or approve requests that bypass normal safeguards.
More broadly, Mandiant says most breaches are still driven by human and organisational weaknesses rather than advanced technology. Attackers are combining these social techniques with other methods, including exploiting software flaws and reusing stolen access, often handing off compromised accounts between different groups within seconds to carry out more serious attacks such as ransomware. At the same time, incidents are taking longer to detect, with attackers remaining inside systems for an average of two weeks, increasing the damage they can cause.
The practical response is to adapt defences to this more personal style of attack. Staff should treat unexpected calls with scepticism, especially those involving urgency or requests for access, and verify requests through a separate, trusted channel. Organisations should also reduce reliance on human approval alone by tightening controls around account changes and access requests, and assume that a convincing voice on the phone is no evidence of legitimacy.
Threats
Extensions: Whether they’re being used to circumvent controls on AI deployment or to steal information, browser extensions (or add-ons) are a 24-carat menace. By their nature, they have broad access to browser data, and many security professionals either limit their use or avoid them altogether. Their presence in official stores is no guarantee that they’re safe. If you do decide to install one, check, check and check again that it’s safe.
Booking: The travel platform has confirmed a data breach which may have resulted in attackers accessing customers’ personal data, including names, email addresses, phone numbers, and booking details. Booking[.]com hasn’t said how many people are affected, nor how the breach happened. It’s not the first such incident. Users should be sceptical about any communications related to previous or current reservations…or anything to do with the platform. TechCrunch
Poison: A previously unknown phishing service is being used to target senior executives across multiple industries to try to steal their credentials. The initial messages masquerade as Microsoft SharePoint document-sharing notifications, and they’re highly personalised. Abnormal says the attack is designed to bypass traditional multi-factor authentication methods - and it underlines the need for phishing resistant solutions.
Apple Pay: A US consumer news group is warning about an increase in scams targeting Apple’s payment platform. Scammers are sending fake “Apple Pay fraud alerts” to trick victims into calling them. Once someone calls, they’re pressured into moving money because of the risk of fraud. Longstanding advice applies: if you’re concerned about something, call the institution using a number that you know to be genuine (never one you’re sent in an alert or a message). Consumer Affairs
Claude: Beware of fake websites impersonating an installer for Anthropic’s chatbot. As Malwarebytes recommends, only download Claude from the official site, and avoid links in emails, ads, or “Pro” versions offered outside official channels.
AI
Capability: Some headlines from Stanford University’s 2026 AI Report:
-
AI capability continues to accelerate and the number of users keeps growing.
-
Capability may be increasing, but AI models still can’t tell the time reliably.
-
There’s no longer a gap between the performance of US and Chinese models.
-
The US hosts most AI data centres. Most of the chips they use are made in a single Taiwanese factory.
-
Safety isn’t keeping pace with capability.
Mythos: Officials in the US and UK are reported to be talking to financial institutions about the security risks that could result from Anthropic’s Claude Mythos Preview model. Anthropic made much of restricting its use because of concerns over its ability to identify security vulnerabilities. Commentators have suggested Anthropic exaggerated its capabilities for marketing purposes. That’s entirely possible, but equally there’s no doubt that AI models do facilitate the identification of security issues (both for good and ill). One comment that caught our eye came from AI evangelist Daniel Miessler who said Mythos represents a leap forward, but that it’s just one step on AI’s evolutionary path; “The AI we have today will be studied in museums like Commodore 64’s and punch cards,” he added.
Identity: Anthropic is to start asking some users to provide their identity with a government-issued document. It says verification will be required when accessing “certain capabilities,” during “routine platform integrity checks,” or as part of safety and compliance measures.
Heart: Researchers at the University of Oxford have developed an AI system that detects subtle changes in heart fat from routine CT scans, predicting heart failure risk up to five years ahead with 86% accuracy. MPost
Personal: Chatbots build a detailed profile of the humans that use them. As a Wall Street Journal columnist suggests, asking “What do you know about me?” can produce a surprisingly detailed result. In the case of the columnist, too detailed to print.
In brief
Supply chain: Recent incidents reveal a shift in attackers’ strategy; they’re focusing on open source tools used in software development as a way of compromising the environments where the tools are used. In one case, a popular vulnerability scanner was hit. In another, a North Korean group hijacked a JavaScript library that has about 100 million weekly downloads. One positive outcome from the attacks is that they have drawn attention to the issue and to the need for developers to have granular control over the tools they’re using.
Turkey: A Turkish government proposal would require people to provide a national ID number to use social media. The Justice Minister says social media companies have agreed to implement such a system if (read when) parliament approves it. Democracy Without Borders
Sovereign: France is taking its first steps towards reducing its dependency on US technology. The French Inter-Ministerial Directorate of Digital Affairs (DINUM) is the lucky guinea pig which will replace Windows with Linux. Meanwhile, in the UK a digital rights organisation says the country’s over-reliance on a small number of tech giants has become an issue of national security.
Preppers: Finland, Sweden, Norway, Denmark and Estonia are creating a system to support offline card payment systems in case internet connectivity is cut. “The likelihood of major disruptions has increased because the geopolitical situation has changed worldwide. There is a war in Europe, and around that war, there is all sorts of hybrid influence and harassment, which may involve disrupting or cutting connections,” Bank of Finland board member Tuomas Valimaki told Reuters.
Connected: South Korean mobile phone users will have access to basic internet connectivity even after they’ve used up their monthly allowance. Speeds will be limited at 400kbps but the government says the aim is to support basic activities like messaging…or accessing government websites. Yonhap News
Jail: Under new UK laws, senior executives could face prison terms if their platforms fail to remove people’s nonconsensual intimate images. The measures are part of government efforts to counter the spread of AI nudify apps and non-consensual pornography.
Archive: Twenty-three major news outlets are preventing the Wayback Machine from adding their content to the Internet Archive. Most say the decision is due to concerns about the data being used to train AI models. WIRED
Landlines: In an effort to keep their kids off smartphones, the Wall Street Journal reports that US families are rediscovering the advantages of the humble landline. Mind you, there does appear to be a learning curve both in how to hold the phone (not in front of your face) and in terms of etiquette (17 calls a day are less than ideal).
Updates
Microsoft: 167 updates in this month’s security release, which we believe is the second highest number in the company’s history. Meanwhile, the US cybersecurity agency has warned that 4 vulnerabilities are being exploited, including one that was addressed 14 years ago. On the plus side, Microsoft is introducing new protections in Windows to defend against phishing attacks that abuse Remote Desktop connection files. These will add warnings and by default disable some shared resources. It’s worth noting that the protections only apply to connections initiated by opening RDP files, not to those made through the Windows Remote Desktop client.
Adobe: Emergency updates for a critical Acrobat and Reader vulnerability that has been exploited for several months.
Patch Tuesday: Monthly security updates from IBM, SAP, Fortinet, Ivanti, AMD, and Joomla.
Google: New rules will penalise websites that hijack what happens when you click the Back button. This has long been abused to redirect users to adverts or unrelated websites instead of the previous page.
SecureDrop: Version 2.15.0 has been released with support for a new Inbox application that is to be rolled out soon. There’s also an update for SecureDrop Client to address a low-priority security issue.
Tails: Another emergency fix to address a vulnerability in Tor Browser.