The IT Nerd

Cybernews has posted a new survey where they polled over 100 AI researchers on a wide range of AI topics, including their views on the pace of AI development, AI’s long-term impact, AGI and ASI timelines, and the greatest source of AI harm.

Their survey comes amid heightened concerns over AI’s existential risk, reignited by a public resignation of Anthropic researcher Jacob Coxon, who said that he and other employees in AI labs earnestly believe the technology can wipe out humanity by the end of the decade.

Their survey also indicates a widespread anxiety among the AI research community, as 30.4% of respondents say it is very plausible that a self-improving AI could pose a major risk to humanity within the next 5 years. Another 33.3% say that it’s somewhat plausible.

Other key findings:

• Nearly 40% call for frontier AI to be slowed down, while 15.7% want it to be stopped completely.
• More than four in ten (43.1%) believe there’s at least a 10% chance that AI will cause a catastrophic event this century.
• 52.4% say a fundamental breakthrough is probably or definitely needed for AGI; 69.6% say the same for ASI.
• When asked about the biggest source of AI harm over the next 10 years, 37.9% of researchers pointed to human misuse, such as cyberattacks, bioweapons, and disinformation.

Find many more insights by exploring the full report here: https://cybernews.com/ai-news/ai-researchers-survey/

With the US midterm elections approaching, a new Cybernews study reveals that Democrats are portrayed negatively in political deepfakes far more frequently than Republicans, while President Donald Trump and Republican-linked sources are responsible for over 40% of the analyzed deepfakes.

Cybernews researchers analyzed 104 unique political deepfakes reported in the Resemble AI database from early 2025 to October 13th, 2026, examining how Democrats and Republicans were portrayed and who shared the AI-generated content.

Key findings:

• Democrats were depicted negatively in 49 deepfakes, compared to 28 targeting Republicans. Only 3 Democratic-related deepfakes were positive, compared to 13 Republican-related ones.
• Trump posted 20 of the 104 analyzed deepfakes. Republican candidates, officials, and party groups shared another 22, compared to 7 linked to Democrats.
• 60% of deepfakes posted by Republicans portrayed Democrats negatively.
• Seven deepfakes were traced to foreign state actors, including Russia, Iran, and China.

For more information and specific examples of the deepfakes, read the full study here: https://cybernews.com/ai-news/political-deepfakes-2026-midterms/

Two critical vulnerabilities have been discovered in PHP, the programming language powering around 70% of websites, potentially putting businesses and internet users at risk.

Voldemaras Kadys, Head of Security & Platform Engineering at Mediatech, explains what these flaws mean for websites, how ordinary internet users could be affected, and why the release of security updates doesn’t necessarily mean websites are safe. Please find Voldemaras’s comments below:

What is PHP?

“PHP is a programming language that powers around 70% of all websites, including online services, stores, booking platforms, and many others. In other words, most websites you browse online are based on PHP and are therefore affected by the recent vulnerabilities.”

What were the flaws in it and why was an update needed?

“Two critical security flaws were discovered in PHP that could allow hackers to take control of vulnerable website servers and potentially steal customer information, change website content, install malware or take the website completely offline. The potential impact of these vulnerabilities may be huge, as most websites run on PHP.

PHP developers have already issued updates for these vulnerabilities, but that doesn’t mean all websites are automatically safe – website operators and hosting providers need to install the updates themselves, and hopefully they do that as soon as possible. The seriousness of the vulnerabilities is also evident from the fact that PHP released updates even for End-Of-Life (EOL) PHP versions, which are no longer actively supported and typically do not receive any more updates.”

Do regular web users need to worry?

“For regular internet users, simply visiting a PHP website doesn’t necessarily mean their device will be compromised. The risk is that attackers may exploit a website that holds their personal information. For example, if a user has an account with an online shop, attackers may be able to hack that website through this PHP vulnerability and extract the information the user has stored there, such as their email address, physical address, name and date of birth.

This kind of information may then be used for phishing, impersonation or other scams. However, not every PHP-powered website is vulnerable, and the vulnerability alone does not instantly mean a breach has happened.

There’s no reason to stop browsing or shopping online, but users should still be extra cautious about suspicious messages because we don’t know what kind of data hackers may have stolen through this vulnerability, and from which websites.”

Cybernews researchers discovered that attackers targeted numerous universities worldwide, attempting to compromise their login systems and access students’ and teachers’ personal data. The attack was carried out by exploiting QuickLaunch, an AI-powered tool that helps colleges and universities manage student and staff sign-ins to school systems and access to online services.

Key information:

• At least 42 universities were identified as possible QuickLaunch-related targets, and 40 more were listed as using a different sign-on setup.
• Attack summaries in the Mandarin language point to Chinese hackers.
• The discovered server stored a large dataset linked to QuickLaunch environments, including roughly 32,400 unique user records across 1,079 domains, with names, email addresses, and user IDs.
• The CVE-2024-7097 vulnerability allowed attackers to create unauthorized accounts via the QuickLaunch API, gaining a foothold in the environment.
• In this specific case, the affected educational institutions face risks beyond just account compromise.

For more information, here’s the full report:

https://cybernews.com/security/hackers-universities-login-data-leak

A recent investigation by 404 Media revealed that advertisers can target ChatGPT users using email addresses, phone numbers, and advertising IDs collected outside the platform. While OpenAI says advertisers cannot access private conversations, the findings raise questions about privacy and advertising within AI assistants.

Jurgita Lapienytė, Editor-in-Chief at Cybernews, argues that the introduction of targeted ads could have an unexpected upside: reminding people that ChatGPT is a commercial product, not a trusted friend, and making them think twice before sharing personal information.

Here’s her full commentary:

“People turn to ChatGPT for personal advice because it feels like talking to a supportive friend. Unlike traditional search, chatting with AI seems less transactional, giving users the sense of having an unbiased, deeply helpful personal assistant.

But how unbiased can these conversations truly be when advertisers can pay to target users with ads based on personal identifiers they collected themselves outside of the platform?

People confide in chatbots, turn to them for answers to delicate life situations, and tap into previous sessions for more personalized responses. They basically lay out their whole lives in front of a bot, so it would help them.

What happens when a targeted ad appears at the end of what felt like a very private conversation, even if only with a machine?

When a doctor offers you a specific medicine, do you wonder if they’ve been paid to do that? In ChatGPT, even with labels and privacy toggles, the line between an objective AI response and targeted advertising feels blurry. Maybe that’s a good thing? Seeing an ad in a chat you thought was private might make you stop oversharing and realize this is just another piece of technology made by a for-profit company.”

Having OpenAI do this may push other people into not trusting them either. Perhaps then there will be questions asked about OpenAI. One can hope.

Cybernews researchers discovered an exposed server linked to a documented large-scale fraud operation centered on online gift card trading platforms. The attacker appears to have focused on intercepting gift cards uploaded to peer-to-peer (P2P) trading platforms, allowing them to capture valuable card data before the normal transaction process was completed.

What did Cybernews researchers find:

• The server’s data indicated that the attacker chose from 58 gift card trading platforms and 3 peer-to-peer crypto exchange platforms that accepted gift cards.
• Exposure of data suggests that some platforms were not vulnerable to abuse, while others exposed internal data, gift card references, or infrastructure details.
• Exposed platforms included Cardlumi, Cardtonic, Trustonic, and Octoplus Box.
• According to the researchers, Anthropic’s Claude was used as the primary AI agent in the attack workflow.

According to Aras Nazarovas, Senior Information Security Researcher at Cybernews, platforms and their customers and sellers are potentially at risk, as customers would purchase redeemed gift cards, and it depends on how well the platform’s customer support handles these disputes. 

For more information, read the article here:

https://cybernews.com/security/exposed-server-gift-card-fraud-operation-analysis

Amazon’s Fire TV Cube 3rd Gen has been hacked using a four-year-old GPU vulnerability, allowing users to gain complete control over the device, Cybernews reports.

The discovery has sparked excitement among home theater enthusiasts, who are exploiting the flaw to turn the Cube into a powerful media player. However, the same vulnerability could also pose security risks.

Here are the key findings:

• The vulnerability (CVE-2022-38181), rated 8.8/10 in severity, allows arbitrary kernel code execution and root access.
• ARM patched the flaw in August 2022, two months before the Fire TV Cube 3rd Gen was released.
• Enthusiasts are using the exploit to install custom operating systems, remove Amazon’s software, and unlock advanced video and audio playback capabilities.
• Attackers could potentially exploit the vulnerability to spy on users, steal data, or incorporate devices into botnets, although there’s no evidence of active exploitation in the wild.

Amazon told Cybernews that the demonstrated method poses little risk to customers, as it requires physical access and deliberate, multi-step modifications. The company says it’s reviewing the details and will take appropriate action if needed.

You can read the full story here: https://cybernews.com/security/amazon-fire-tv-cube-root-exploit-enables-full-unlock/ 

With Samsung reportedly preparing to launch its first Android XR smart glasses in November, a new Cybernews analysis of 7 existing AI smart glasses brands reveals the privacy risks users may face as the technology becomes more widespread.

Researchers examined smart glasses from Meta, Rokid, RayNeo, INMO, Solos, Even Realities, and Halliday, looking at app permissions and trackers, as well as 9 privacy criteria covering areas such as camera indicators, data retention, on-device processing, and dedicated privacy documentation.

Key findings:

• Meta requests the most permissions overall (70) and the most dangerous permissions (18), including access to read external storage and SMS and to record audio.
• INMO Global and Solos AirGO contain the most trackers, with 6 detected in each app, while Hi Rokid is the only app with no trackers detected.
• All 7 analyzed apps rely on cloud processing for their AI features – none runs core features such as voice, translation, or image analysis locally on the glasses.
• 5 of 7 brands don’t specify any data retention limit.
• Only Meta and Rokid have a confirmed response when the glasses’ recording indicator light is covered.
• Only 3 of the 7 brands – Meta, Rokid, and Even Realities – have privacy documentation specifically written for their glasses.

The findings show that the privacy risks of smart glasses extend beyond people who may be unknowingly recorded to the people wearing the devices themselves.

Full research:

https://cybernews.com/ai-news/smart-glasses-privacy

A hacker is selling 54 million records allegedly linked to Airbnb, Uber, PayPal, Booking.com, and Google.

The threat actor, known as Marx, claims to have live access to the source of the data, potentially allowing them to intercept sensitive communications, including authentication codes, as they reach users.

Cybernews researchers looked at the claims, here’s what they found:

• The samples of data appear to originate from a single third-party SMS service.
• The samples contain phone numbers and mobile carrier information, while the alleged Uber data also includes names of people who booked rides. The message contents seem to be heavily stripped.
• The samples examined appear geographically limited to India and Oman, and researchers could not determine the full scope of the alleged breach.

If proven to be legitimate, such exposed data could result in increased risks of phishing attacks and, in some cases, account takeovers.

Here’s the full investigation:

https://cybernews.com/security/airbnb-uber-google-data-breach-claims

Created by Markus “Notch” Persson and developed by Mojang Studios, Minecraft has over 212 million monthly active players, and now the players’ data surfaced in an alleged 18 million breach

Two posts on underground cybercrime forums claim that Minecraft players’ data has been breached. One threat actor says 18 million user records are for sale, while another post cites a more modest 9 million records.

Cybernews researchers investigated the claims, and here is what was found:

• Supposedly, different datasets appear to be the same 1,000 records.
• The sample contains usernames, email addresses, and, in some cases, password hashes. 
• The records also appear to have been collected from specific Minecraft servers, with only 3 unique servers appearing across the sample.
• Information could be used for credential stuffing and social engineering scams, particularly if email addresses are paired with passwords victims have reused elsewhere.

The team found that email addresses in the sample appear in Have I Been Pwned and are referenced in multiple infostealer combo lists. This suggests that the data source may be infostealer malware.

For more information, read the article here:

https://cybernews.com/security/minecraft-players-data-breach