16 Billion Login Records Leak Online in One of the Largest Credential Exposures to Date

A digital breach of unprecedented scale has quietly unfolded online. Security researchers, after months of monitoring, have identified a network of exposed datasets containing a combined total of over 16 billion login records. These collections, found on unsecured servers, include usernames and passwords gathered from a wide range of platforms and services.

As reported by CyberNews, the data appears to originate from a mix of infostealer malware, credential stuffing sets, and previously unreported leaks. According to the investigators, the datasets surfaced across various storage systems left open on the internet, with some briefly accessible to the public. Although their availability was short-lived, the exposure window was long enough for researchers to capture and analyze a significant portion of the records.

The credentials come from a wide spread of online environments. Included are accounts linked to social media platforms, corporate tools, cloud services, VPN portals, and even government resources. Many of the records followed a repeating format, typically a web address, followed by login details and an associated password. This structure matches the way modern malware tends to collect sensitive information, allowing for automated use in later attacks.

Several factors set this incident apart. Unlike older leaks that sometimes recirculate in cybercrime forums, the bulk of the data here appears recent and unreported. In fact, researchers say only one previously known dataset — containing around 184 million records — was already in public discussion before this. The rest, according to early analysis, represent newly surfaced material. Some files included not only basic login credentials but also session tokens, cookies, and metadata, all of which can be exploited in targeted intrusions.

The variety in dataset naming conventions has made it difficult to trace every origin point. Some files were labeled generically, using terms like “logins” or “credentials,” while others hinted at geographic or platform-specific links. For instance, a collection exceeding 3.5 billion entries appeared tied to the Portuguese-speaking world, and another with over 455 million entries seemed connected to users in the Russian Federation. Smaller sets, like one named after Telegram, suggest the targeting of specific platforms or services.

Cybersecurity experts following the case have noted how such aggregated credential data fuels a range of malicious campaigns. Among the most likely threats are phishing schemes, identity theft operations, ransomware deployment, and business email compromise attempts. Because the records include both older and newer entries, some individuals and organizations may be at risk without even realizing it.

One of the more troubling aspects is the lack of clear ownership over the exposed data. While some believe that portions could have been compiled by security analysts for research, much of it is presumed to have passed through the hands of cybercriminal actors. The scale of the exposure makes it likely that the datasets are already being used or sold through underground channels.

Although no one can fully undo what has already leaked, security professionals are urging action. They advise individuals to review their existing accounts and update passwords, especially for any services used regularly. Enabling multi-factor authentication can reduce the risk of unauthorized access. Organizations, meanwhile, are encouraged to audit their systems, look for signs of compromise, and educate users on how to respond to potential phishing attempts or credential theft.

Massive breaches of this nature are becoming more common. Just last year, the RockYou2024 password dump revealed nearly 10 billion unique passwords. Earlier this year, another massive incident, known as the Mother of All Breaches, surfaced with over 26 billion records. This latest event, though smaller in scale than MOAB, is still notable because of the focus and freshness of its contents.

At a time when digital infrastructure underpins nearly every aspect of life and business, maintaining control over authentication data has never been more critical. While not all exposed records may be actively in use, even a small fraction of successful logins can result in major disruptions for individuals and companies alike. What matters now is not simply what leaked, but how quickly users and institutions respond to secure their systems.

Read next:

• Firms Rethink Internal AI Builds to Cut Costs, Improve Control, and Manage Risks of Autonomous Decisions

• Position Bias in AI Models Threatens Accuracy in High-Stakes Applications, MIT Warns
by Irfan Ahmad via Digital Information World

Firms Rethink Internal AI Builds to Cut Costs, Improve Control, and Manage Risks of Autonomous Decisions

Your next boss may not be just a person with a corner office. In some companies, decision-making is beginning to rely on something less visible that is automated systems built to think through problems and offer answers without human hesitation.

Recent industry analysis from Gartner suggests that artificial intelligence is becoming more than just a tool in the background. If current trends continue, by the middle of this decade, around half of the important choices made inside companies could be influenced or directly completed by AI. This shift is not simply about speed, but about how information is processed, evaluated, and turned into action.

Where AI is handled properly, executives may find they can respond faster to change and manage resources more effectively. But where it’s deployed without proper oversight or alignment with business goals, the consequences could be harder to manage. Mistakes at scale are not just expensive, they can be hard to reverse.

In practical terms, these AI agents act as a kind of middle layer between raw data and final decisions. They’re designed to pull in streams of information, assess them in real time, and guide leadership through the more complex layers of judgment. While they don’t remove the need for people, they do change how people approach strategy and planning.

Some firms have already begun to restructure how departments work together. Analysts and data teams are now expected to sit closer to management. Their role isn’t just to deliver charts, but to help shape what kind of questions get asked in the first place. AI becomes more useful when it’s matched with human judgment that knows where to focus.

Not every firm will get this balance right. In fact, the same forecasts warn that a large number of data leaders may fall short in managing the synthetic data they use to train and test models. This could introduce weak points in both accuracy and compliance, which in turn could affect broader business outcomes.

AI itself isn’t neutral. It reflects the quality of the data behind it and the rules set by those who deploy it. In the future, some company boards may even start to bring automated systems into their oversight processes. By the end of the decade, it’s expected that a portion of global boards will begin using AI to independently review and challenge high-stakes decisions made by executives. This doesn't replace accountability, but it reshapes where and how it's applied.

Meanwhile, a growing number of firms are considering whether to build their own generative AI systems instead of relying on external providers. Those who go that route often cite lower long-term costs and stronger control over how their systems evolve. But the choice also comes with increased pressure to understand the risks from the inside out.

The role of leadership is changing. It’s no longer enough to manage teams and review quarterly plans. Those in charge will need to understand what machines can do, where they fall short, and how to make the most of a future where decisions are no longer made in isolation.

Image: DIW-Aigen

Read next: Position Bias in AI Models Threatens Accuracy in High-Stakes Applications, MIT Warns
by Irfan Ahmad via Digital Information World

Position Bias in AI Models Threatens Accuracy in High-Stakes Applications, MIT Warns

A team from MIT has uncovered the inner mechanics behind a persistent flaw in large language models that is their tendency to overlook information buried in the middle of documents. The findings, grounded in a rigorous theoretical framework and supported by extensive experiments, explain why these systems often favor text that appears at the beginning or end.

The researchers traced the root of this issue, referred to as “position bias”, to architectural design choices and how these models are trained to process sequences. Central to the analysis is how the attention mechanism, a core component of models like GPT-4 or LLaMA, handles the flow of information across multiple layers.

Using graph theory, the team demonstrated that attention patterns are not evenly distributed. Instead, certain tokens become dominant simply due to their position. When the model reads from left to right, earlier tokens often accumulate more influence as the layers deepen, even when their content is less relevant. This effect intensifies as more layers are added, creating a cascade where initial tokens disproportionately shape the model's decisions.

The study shows that even without adding any formal position tracking, the structure of the model itself introduces a preference for the start of the sequence. In experiments with synthetic retrieval tasks, the performance of the models dipped when key information was placed in the middle of the input. The retrieval curve followed a U-shape, strong at the start, weaker in the center, then improving slightly at the end.

This behavior wasn’t incidental. Controlled tests confirmed that position bias emerged even when the training data had no such leanings. In setups where the data favored certain positions, the models amplified those biases. When models were trained on sequences biased toward the beginning and end, they mirrored that pattern, heavily underperforming in the center.

The paper also explored how positional encoding schemes, tools designed to help the model track where a word appears, can partially counteract this effect. Techniques like decay masks and rotary encodings introduce a fading influence based on distance, nudging the model to attend more evenly across the sequence. However, these methods alone don’t eliminate the bias, especially in deeper networks where earlier layers already tilt the attention forward.

In practical terms, this means that users relying on AI models for tasks like legal search, coding assistance, or medical records review may unknowingly encounter blind spots. If key content appears mid-document, the model might miss or misjudge it, even if everything else in the system functions as intended.

The implications go beyond diagnostics. By showing that position bias is both an architectural and data-driven phenomenon, the researchers offer pathways to mitigate it. Adjustments in attention masks, fewer layers, and smarter use of positional encodings can help rebalance the focus. The study also suggests that fine-tuning models on more uniformly distributed data could be essential in high-stakes domains where omission carries risk.

The research not only maps the bias but explains its evolution. As tokens move through the model, their contextual representations are repeatedly reshaped. Those that appear earlier begin to dominate, not because they contain better information, but because they become more deeply embedded in the model's reasoning. In this sense, the bias is baked into the system’s logic.

Rather than treating this as a bug, the team sees it as an opportunity for improvement. Their framework doesn’t just diagnose; it provides tools to reshape how models perceive position. By better understanding these internal biases, developers can build systems that reason more fairly and consistently across the full length of input, beginning, middle, and end.

Image: DIW-Aigen

Read next: Why a Wrench Might Outlast Code in the Age of AI
by Irfan Ahmad via Digital Information World

OpenAI Rolls Out ChatGPT Image Generation via WhatsApp at +18002428478

While much of the conversation around artificial intelligence tends to revolve around speed and sophistication, OpenAI’s latest move feels unexpectedly low-tech, and perhaps deliberately so. The company has introduced a way for users to access its image generation feature simply by sending a message to 1-800-ChatGPT (or +18002428478 to be exact) through WhatsApp.

This rollout opens the tool to all users globally, according to the announcement shared through the company’s official X account. But for a platform so deeply associated with cutting-edge AI, the decision to lean on a toll-free number, something more closely tied to landline-era habits, comes across as a curious throwback.

It’s difficult to gauge how many users were actively hoping to reach an AI tool through a method that predates the smartphone. The symbolic use of a “1-800” code may even be lost on those who never had to think about long-distance calling. Still, the move could signal an effort to make AI services feel more approachable to demographics that might be less comfortable navigating app stores, new interfaces, or competing platforms.

In some ways, it suggests that OpenAI is trying to widen its reach, not by adding complexity, but by lowering the technical barrier. For users already familiar with WhatsApp, sending a quick message might feel less intimidating than signing into a new app or website. And for those who recall a time when customer service meant dialing a toll-free number, this might feel oddly familiar, even if the conversation now happens with an algorithm rather than a human voice.

Read next: Marketers Brace for Soaring Content Needs as Expectations Shift Through 2027
by Irfan Ahmad via Digital Information World

Marketers Brace for Soaring Content Needs as Expectations Shift Through 2027

As customer attention continues to fragment across platforms, brands are finding themselves at a crossroads where consistency, speed, and personalization must all converge. New insights reveal just how much pressure is building: the majority of marketing professionals anticipate a fivefold surge, or more, in content requirements by 2027.

In a study, conducted by Adobe, that gathered feedback from over 1,600 marketing professionals, the overwhelming majority acknowledged a sharp rise in demand. Nearly every respondent reported at least double the content workload compared to just two years ago. More than six out of ten indicated that this growth had already surpassed five times their previous volume, and a significant share expects that pace to continue accelerating.

The main driver behind this surge? Audiences are demanding richer, more personalized experiences. Over half of marketers noted that tailored messaging now ranks as the most dominant factor pushing content output higher. At the same time, changing habits, such as the growing importance of short videos and audio-first formats, are forcing teams to rethink how they communicate. Hybrid touchpoints, blending digital with in-person channels, also require more content delivered more frequently, in more styles.

In fact, staying visible in this environment means showing up constantly. A large portion of marketers reported that their audiences now expect updated material at least weekly, often multiple times per week, placing more pressure on already stretched teams.

Among all content categories, social media leads the way in growth. The speed and virality of platforms like TikTok, Instagram, and others have shifted the goalposts. More than half of marketers believe that demand for social-first and short-form video content is growing faster than any other type. Yet understanding what resonates across diverse platforms remains elusive. Many teams find themselves producing vast amounts of content without clear feedback loops, which makes optimization difficult.

Across both organic and paid social channels, time remains a persistent bottleneck. Marketers cite challenges in producing enough brand-consistent and timely content. In paid campaigns, teams also struggle to generate enough variety to keep messages fresh or to reach distinct audience segments effectively.

Even when strategy is aligned and messaging is clear, internal roadblocks often slow execution. For many marketing departments, a single piece of content may pass through dozens, or even hundreds, of hands before it’s published. Roughly half of surveyed marketers said the typical review and approval process involved between 50 and 200 individuals. In some cases, that number was even higher.

Volume has followed complexity. About seven in ten marketers now generate at least 1,000 content assets annually, while some organizations report output in the hundreds of thousands. Still, many say that managing this level of production is unsustainable without significant changes.

Among the top challenges: limited time for ideation and creation, disconnected workflows, and prolonged manual approval stages. The result? More than half of marketers say they spend the majority of their time navigating internal reviews rather than developing impactful campaigns.

To address these pressures, many teams are integrating generative AI into their workflows, not as a novelty, but as a core tool. Marketers are applying AI in various areas: optimizing content for better performance, adapting assets for global audiences, and creating new multimedia elements. More than half already use generative AI at multiple stages of production, and a large majority intend to increase that usage within the coming year.

As expectations for speed and relevance continue to rise, marketers are focused on scaling smarter, not just faster. Through redesigned workflows and deeper integration of automation and AI, teams are working to balance high output with brand integrity. The goal isn’t simply to meet content quotas, but to deliver messages that connect, perform, and evolve with audience behavior.

In this climate, operational agility and creative adaptability aren’t just nice to have, they’re now fundamental to staying competitive.

Read next: Personalized Shopping Isn’t Always a Deal, It Might Push Prices Higher
by Irfan Ahmad via Digital Information World

Meta Faces Backlash Over Mysterious Instagram Bans and Facebook Selfie Checks

For weeks now, Instagram users have reported sudden account bans with no clear explanation or response from the platform. In many of these cases, individuals insist they have not violated any rules, yet they find themselves locked out with little recourse and no reply after submitting appeals. The volume of complaints, especially on Reddit and X, has grown enough to catch wider attention.

Many of the affected users on Reddit describe being stuck in an unresolved process. Some say they uploaded identity documents or submitted multiple appeals, only to be met with silence. Several note that they received no warning before losing access and have not been given a reason for their suspension.

The concern for some users goes beyond lost access to personal profiles. Small businesses and independent creators are also being affected, with some stating that the ban has cut them off from their main marketing platform and customer base. As a result, users are increasingly pointing to automated systems as the likely cause, though Meta has not confirmed any technical details or issued a public statement.

Moderation errors are not unusual on large platforms, especially when artificial intelligence is involved, but the number of cases and the lack of human oversight in the appeal process have left users frustrated. In some instances, people have claimed they were banned for serious policy violations they insist they did not commit, such as content related to child exploitation, a label that carries severe reputational consequences, even if wrongly applied.

Online, calls for accountability are increasing. A petition calling for Instagram to review its moderation approach has gained several thousand signatures. Some users have discussed the possibility of legal action, citing emotional and financial consequences from the account losses.

This incident follows a broader pattern among major tech platforms, where the reliance on AI moderation is being scrutinised. Earlier this year, Pinterest acknowledged an internal error that led to mass suspensions but declined to say whether artificial intelligence was responsible. That company eventually restored affected accounts after admitting fault, but gave few details about the underlying issue.

While Instagram remains under pressure, Meta is also facing a growing number of complaints from long-time Facebook users who say they have been locked out due to the platform's facial recognition checks. As per several comments shared on Digital Information World's post, the process now requires some users to submit a video selfie to confirm their identity. This requirement appears to be triggered in a variety of cases, some after years of activity with no prior issues.

Dozens of individuals have described being asked to submit facial video verification without being told why. Many of them do not own smartphones or desktop cameras and say they are now permanently locked out. Some, including elderly users who only use Facebook to stay in touch with family, expressed confusion and concern about how the change was introduced. Others worry about how the video data might be stored or used, particularly in relation to AI training or personal profiling.

One user explained they had been on Facebook for over a decade and had never posted anything questionable. They were looking at the Marketplace when they were suddenly logged out, then told to verify their identity with a facial video. Another user said they had not used Facebook for years but attempted to log in, only to find the old account inaccessible. After trying to sign up again, they were subjected to puzzles and then a facial scan prompt.

Some individuals speculated that their technical setups might have triggered the system’s suspicion. One mentioned running Linux without geolocation libraries, setting their birthday to 1905, and using privacy-focused browsers that block tracking. Others questioned whether low activity levels or old accounts might be seen by the system as potential bot behaviour.

The consistency across many of these reports suggests that the video selfie check is not just reserved for new sign-ups or flagged activity. It seems to be part of a broader verification push that is catching regular users in the net. Several pointed out that they were not warned about the change and were offered no other way to prove their identity if they lacked a camera or chose not to upload a recording.

Some worry that the system may be biased or discriminatory in how it flags accounts, especially in the absence of transparency from Meta. One user questioned whether certain political views, inactive posting habits, or refusal to participate in advertising were factors that influenced the verification demand. Others said they felt like they were being gradually pushed off the platform, not for breaking rules, but for not aligning with how Facebook wants users to behave or interact.

Despite the volume of complaints, Meta has not publicly addressed the facial recognition issue or explained the criteria triggering the video selfie requirement. For now, affected users remain locked out with no clear timeline for resolution.

Image: DIW-Aigen

Read next: Meta Adds Privacy Warning After AI App Publicly Displays Personal Chats
by Irfan Ahmad via Digital Information World

Meta Adds Privacy Warning After AI App Publicly Displays Personal Chats

Meta has added a warning label to its AI app after users were found publicly posting personal conversations without realising those interactions were being shared. The update, now visible on the app’s “post to feed” button, advises people not to include personal or sensitive information when publishing content. The change follows mounting concern over how easily private exchanges were being made publicly viewable.

Reports first emerged after users noticed that the app’s public feed, called “discover”, was showing highly personal queries submitted to Meta’s AI, ranging from health concerns to legal questions. In some cases, users even included phrases such as “please keep this private” in what were already public posts. While the app does not share conversations by default, many people appear to have been unaware that using the “share” option meant their chats would be made visible to everyone.

The volume and nature of the content being exposed drew criticism from privacy advocates, who pointed out that most AI chat apps do not display user interactions in a public-facing stream. Observers have noted that the app offers few familiar cues or design features to indicate that posts are going live. Some compared it unfavourably to Meta’s other platforms, where sharing controls are more visible and widely understood.

The new warning message now appears when users try to post a chat publicly, but based on reports, it only shows once, on the first attempt. It clearly states that prompts shared with the feed are public, may appear on other Meta platforms, and should not include private details. However, critics argue that a single disclaimer may not be enough to prevent confusion, especially for users who miss or dismiss the initial notice.

Image: Business Insider

At the same time, it appears that Meta has altered what kind of content is shown in the app’s public feed. Text-based conversations no longer seem to appear. Instead, the feed is currently displaying AI-generated images and video content. It is unclear whether this change is permanent or whether it reflects a temporary shift following the attention the app has received in recent days. Meta has not confirmed whether this is part of a wider redesign or a direct response to the criticism.

Those who have already shared personal content can still take steps to reduce their visibility. Within the app’s settings, it is possible to hide all past posts by adjusting the privacy options under the “Data & Privacy” section. Users can make every prompt private in just a few taps, though the setting may not be obvious at first glance.

The issue has sparked renewed scrutiny of how emerging AI tools are being integrated into social platforms. While many users approach chatbots expecting private assistance, the blending of conversational AI with public sharing features appears to have created misunderstandings, with very real privacy consequences.

Read next: Meta Expands WhatsApp Monetization with In-App Ads and Paid Channel Subscriptions
by Irfan Ahmad via Digital Information World