The IT Nerd

Cybernews researchers discovered that Meta’s personal assistant, which is integrated into Messenger, WhatsApp, Instagram, and other apps, is easy to manipulate into revealing harmful information. The Llama 4-based chatbot was easily tricked into providing instructions on making a Molotov cocktail.

The assistant was easily tricked by utilizing the so-called “narrative jailbreaking” practice. The technique masks the harmful request by asking the bot to tell a “story” to bypass safety filters. To execute the jailbreak, the team simply asked the chatbot to tell a story about the Winter War between Finland and the Soviet Union, requesting details about how the incendiary devices were made back then. 

While it’s unlikely that people will flock to Meta for advice on Molotov cocktail-making, the issue highlights the possibility of abusing the chatbot for purposes that appear to be beyond the scope of what an AI assistant ought to be capable of.

The team disclosed the issue to Meta immediately after discovering it. After the publication went live, the company told Cybernews it had resolved the problem.

Also, Cybernews researchers recently discovered that Lenovo’s customer service assistant, Lena, had an XSS vulnerability that allowed the running of remote scripts on corporate machines if you asked nicely.

Meanwhile, another chatbot, used by the travel agency Expedia, allowed users to ask for a recipe for making a Molotov cocktail. The company eventually fixed the issue, and the chatbot stopped advising on making incendiary devices.

To read the full research report, please click here.

Cybernews has just released a report about enterprise chatbot maker Yellow.ai. It reveals that Yellow.ai’s customer service chatbot allowed users to steal cookies and execute malicious code upon request, with zero pushback.

The flaw highlights multiple security issues, such as improper user input sanitization, improper chatbot output sanitization, the web server not verifying content produced by the chatbot, running unverified code, and loading content from arbitrary web resources. This leaves many options for Cross-Site Scripting (XSS) attacks.

Cybernews reached out to Yellow.ai before publishing the research, and the company fixed the issue by sanitizing the generated code, ensuring that it would not get executed. However, the bot still generates malicious code if asked.

Yellow.ai customers include major companies like Sony, Logitech, Hyundai, Domino’s, and hundreds of other brands. It is unclear whether the chatbots used by Yellow.ai’s clients have the same security flaw as the customer service bot had, so Cybernews urges all Yellow.ai clients to be wary.

For more information, here’s the full report: https://cybernews.com/security/yellowai-customer-chatbot-cookies-flaw

The EU’s top official, European Commission’s president Ursula von der Leyen, was on her way to Bulgaria when a suspected Russian attack forced her plane to land without essential navigation tools. 

This harrowing episode was no accident but what officials suspect to be a deliberate act of Russian interference – an electronic attack targeting critical infrastructure in the heart of the European Union.

This incident exposes not only the elevated state of geopolitical hostility but also the cybersecurity weaknesses within EU institutions themselves.

According to the research by the Business Digital Index, or BDI, the EU’s cybersecurity defenses resemble an office where nearly half the doors are unlocked, passwords are scrawled on sticky notes, and the alarm system is known to be broken but left unfixed.  The BDI findings reveal the reality that EU institutions may not be robustly prepared to withstand or respond effectively to high-impact cyber-physical attacks like GPS jamming.

The researchers looked at 75 EU institutions and found that none got an A or B for cybersecurity efforts. 35% got the lowest grade, an F. The problems are especially clear with basic security: in the F-rated institutions, 85% of employees reused passwords that had already been breached. In C-rated ones, only 8% did this. SSL/TLS configuration issues were identified in 100% of F-rated institutions. 

These findings point to very real – and these days accelerated by AI – risks for phishing, malware, and stolen data. Attackers can now do such things as mimicking colleagues using deepfake technology, and deploying malware that adapts in real time to avoid detection. Needless to say that these potential threats can result in financial loss, reputational damage, and regulatory penalties for EU organizations. 

The EU’s main response to growing cyber threats has been to add more rules in order to improve cybersecurity. But the data shows that just having rules isn’t enough. Despite these new rules, nearly half (46%) of the EU’s lowest-rated organizations have already suffered data breaches.

I believe that the real problem is that leaders aren’t acting urgently or taking responsibility. For example, almost all D-rated and F-rated institutions had insecure hosting environments. Domains vulnerable to email spoofing were found in every C-rated organization and in 96% of D-rated and F-rated ones.

The EU needs to do more than merely add more rules and formally follow them. It needs to make sure leaders are held responsible for breaches. That means executives should have part of their pay tied to cybersecurity results. It also means having real, independent security checks with actual consequences for failure. The Transport sector is doing a little better than others, and the EU should learn from that.

Some might argue that more rules will solve the problem, or that it’s just too big to fix in a short amount of time. But the numbers tell a different story: the institutions with the worst track records are the same ones that don’t pay attention to basic security practices such as using strong and uncompromised passwords. At the end of the day, this comes down to leadership.

Given that cyber threats keep on evolving and the geopolitical situation isn’t exactly what we want it to be, the risks are really high. Every day the EU waits, it puts sensitive data, economic stability, and public trust at risk. If the EU wants to be a leader in digital governance, it needs to make cybersecurity a top priority for executives, invest in training, and hold leaders to account.

If nothing changes, the next headline won’t be about bad grades or landing with paper maps. It might be about a real crisis that rules can’t fix. The question now is whether the EU will act in time.

ABOUT THE AUTHOR

Jurgita Lapienytė is the Editor-in-Chief at Cybernews, where she leads a team of journalists and security experts that uncover cyber threats through research, testing, and data-driven reporting. With a career spanning over 15 years, she has reported on major global events, including the 2008 financial crisis and the 2015 Paris terror attacks, and has driven transparency through investigative journalism. A passionate advocate for cybersecurity awareness and women in tech, Jurgita has interviewed leading cybersecurity figures and amplifies underrepresented voices in the industry. She’s recognized as the Cybersecurity Journalist of the Year and featured in Top Cyber News Magazine’s 40 Under 40 in Cybersecurity. Jurgita has been quoted internationally – by the BBC, Metro UK,  The Epoch Times, Extra Bladet, Computer Bild, and more. 

Cybernews researchers have uncovered a massive data exposure affecting more than 250 million identity records across seven countries, including Turkey, Egypt, Saudi Arabia, the United Arab Emirates (UAE), Mexico, South Africa, and Canada.

Three misconfigured servers — hosted on IP addresses in Brazil and the UAE — contained detailed personal information, resembling government-level identity profiles, now confirmed to have been publicly accessible.

What data was exposed? 

The leaked information included ID numbers, full names, dates of birth and gender, contact details, and home addresses.

Record numbers exposed:

• Turkey (88,396,572 records)
• Egypt (77,744,912 records)
• Saudi Arabia (26,827,301 records)
• UAE (4,856,942 records)
• Mexico (8,740,000 records)
• South Africa (44,472,288 records)
• Canada (9,322,549 records)

Cybernews contacted the hosting providers, and as of now, the data is no longer publicly accessible.

To read the full research, please click here.

Cybernews’ latest overview of the first half of 2025 reveals that the total number of breached accounts fell by 20 times compared to the same period in 2024, according to Cybernews’ Personal Data Leak Checker tool.

However, while breaches have dropped significantly compared to last year, several countries remain highly vulnerable as millions of accounts remain exposed. The top three countries with the highest number of breaches in 2025 are the U.S., with 2.5 million breached accounts, France, with 1.8 million breached accounts, and India, with 1.2 million breached accounts.

This report offers key insights into global data breach trends in the first half of 2025, highlighting that unsafe online practices such as subscriptions to suspicious websites and weak password reuse remain the leading causes of breaches.

Key findings of this research:

• The first six months of 2025 recorded 15.8 million breached accounts globally.
• The top five countries most breached worldwide in the first six months of 2025 are the U.S., France, India, Russia, and Venezuela.
• January has the highest number of breached accounts in 2025 so far.
• Countries experienced a spike in breach numbers in March, making it the second most breached month in the first six months of 2025. 
• The second quarter of 2025 saw a 77% dip in breached accounts.
• Despite a 20-fold drop compared to 2024, breach density per capita shows the U.S. remains the most affected nation, with 8 in every 1,000 internet users impacted.

To read the full research, please click here.

Cybernews researchers have uncovered critical misconfigurations on two Tencent Cloud subdomains that exposed sensitive credentials and internal source code for months, posing risks to millions of Tencent Cloud users.

Tencent Cloud is one of the largest global cloud providers, serving over 10 million users worldwide across gaming, finance, communications, and enterprise sectors. 

Key takeaways: 

• One of the affected services was related to Tencent’s internal load balancer, and another subdomain was a deployment of JEECG, an open-source development platform promoted by Tencent Cloud.
• The hardcoded credentials appeared to grant direct access to Tencent Cloud’s administrative console. 
• The environment files with hardcoded administrative console credentials and the .git directory have been publicly accessible for months.
• The exposed passwords were also weak and vulnerable to dictionary attacks. They were composed using the company name, the year, and some symbols.
• Tencent acknowledged it as a “known issue” and closed access. 

Leak significance 

A malicious actor with access to the misconfigured files and directories could potentially:

• Gain full administrative access to the production systems
• Tamper with internal API services 
• Attach malicious payloads to the trusted front-end code 
• Pivot further into Tencent’s internal cloud infrastructure 
• Or simply abuse the trusted Tencent domain for malicious phishing campaigns

To read the full research, please click here.

The Cybernews research team uncovered that MagentaTV, Deutsche Telekom’s TV and streaming platform, exposed over 324 million log entries — totaling 729GB of data — via an ad delivery platform. MagentaTV is estimated to have around 4.4 million users.

Researchers also claim that the exposed instance received new log entries every day, with anywhere from 4 to 18 million logs being added daily.

Moreover, the team believes that the instance was publicly accessible at least since early February 2025, with the company removing it from public view after our team contacted them in June.

What details did the MagentaTV leak?

While the majority of the information accessible via the exposed instance could be considered non-sensitive, some of the leaked logs contained HTTP headers from requests sent by MagentaTV customers.

According to the researchers, some user data was also exposed in the leak, including:

• IP addresses
• MAC addresses
• Session IDs
• Customer IDs
• User agents

Potential risks for users

Attackers could potentially utilize leaked data to track user locations, identify them, and direct targeted attacks against specific devices.

Additionally, customers may face additional risks if attackers cross-reference the leaked data with older breaches, especially using IP addresses to identify users.

To read the full research report, please click here.

Thousands of ChatGPT conversations are appearing in Google search results, sparking privacy concerns as users discover that their shared chats, including personal details, are publicly accessible with the right search terms.

While OpenAI has responded by removing the “Share” feature from the ChatGPT app, the Cybernews research team warns that the issue poses real-world risks:

“Many cases discussed online involve exposure of personally identifiable data such as names and addresses. If combined with controversial content, this data could be weaponized for harassment or doxxing.”

“Removing the share feature in the ChatGPT app may be an extreme way to deal with the problem. However, it’s great that OpenAI is working on removing shared content from search engine indexes, and hopefully, this will allow for faster removal of unintentionally shared content,” commented the Cybernews research team.

On the technical side, the team has added:

 “This problem could have been mitigated by blocking crawlers via robots.txt — but with AI operators increasingly ignoring these rules, OpenAI may have had no choice but to retire the feature entirely.”

Cybernews also advises concerned users to manually delete any shared links via: Settings → Data controls → Shared Links.

You can read the full article here.

The Business Digital Index (BDI) platform evaluated and ranked 20 of the most visited travel and tourism websites by their cybersecurity posture.

Each company’s score reflects performance across seven security categories: software patching, web app security, email protection, system reputation, hosting infrastructure, SSL/TLS configuration, and data breach history. 

Here are the 20 most popular travel & tourism websites ranked by cybersecurity (best to worst):

Out of the 20 travel and tourism websites analyzed, just two—Trip.com and Flightradar24—achieved an A grade.

These four—Wetter.com, Hilton, Marriott, and Skyscanner—received the lowest cybersecurity scores.

To find out what critical vulnerabilities these websites have and why some of them score so low, please see the full report here: 

https://businessdigitalindex.com/research/cybersecurity-analysis-reveals-critical-vulnerabilities-across-20-major-travel-tourism-websites

As artificial intelligence becomes increasingly central to the operations of America’s largest corporations, recent research reveals potential security vulnerabilities that could affect both organizations and their customers. 

An analysis by cybersecurity experts at Cybernews examined AI deployments across the S&P 500 and uncovered close to 1,000 potential weak points that may lead to data exposure, theft of proprietary information, and erroneous AI actions.

The study found that 327 S&P 500 companies publicly report using AI tools in their operations in sectors including finance, healthcare, manufacturing, and energy. 

While these tools have accelerated innovation and efficiency, safety measures have yet to fully catch up, leaving systems open to misuse or failure. This includes AI outputs that may be inaccurate or misleading, unintended disclosure of confidential data, and risks of corporate secrets being compromised.

The potential vulnerabilities extend across multiple industries. Technology and semiconductor companies are especially vulnerable to data leaks and intellectual property risks. Financial institutions might face challenges protecting client data while ensuring AI does not reinforce unfair bias in lending. 

Healthcare providers carry the added responsibility of protecting patients from flawed AI-driven recommendations. Meanwhile, industrial and infrastructure sectors must guard against disruptions that could affect critical services, such as power supply or supply chain operations.

For consumers, the consequences are tangible. Unsecured AI systems risk leaking private details – ranging from medical histories to financial records – while flawed AI judgments could influence decisions that directly affect people’s health and finances. 

As AI tools play a larger role in retail, banking, transportation, and other areas, protecting these technologies becomes essential for public protection.

The report highlights past incidents that illustrate these dangers. IBM’s Watson once offered unsafe cancer treatment suggestions. Apple’s credit system faced scrutiny after allegations of gender bias. Zillow’s AI-driven pricing led to substantial financial losses. Additionally, Samsung experienced unintended source code disclosures due to inappropriate use of AI chatbots by employees.

As AI further transforms businesses, past incidents and potential threats show how crucial it is to improve security strategies in parallel.