April 24, 2025

Algorithm can make AI responses increasingly reliable with less computational overhead

The algorithm uses the structure according to which the language information is organized in the AI's large language model (LLM) to find related information. The models divide the language information in their training data into word parts. The semantic and syntactic relationships between the word parts are then arranged as connecting arrows—known in the field as vectors—in a multidimensional space. The dimensions of space, which can number in the thousands, arise from the relationship parameters that the LLM independently identifies during training using the general data. ... Relational arrows pointing in the same direction in this vector space indicate a strong correlation. The larger the angle between two vectors, the less two units of information relate to one another. The SIFT algorithm developed by ETH researchers now uses the direction of the relationship vector of the input query (prompt) to identify those information relationships that are closely related to the question but at the same time complement each other in terms of content. ... By contrast, the most common method used to date for selecting the information suitable for the answer, known as the nearest neighbor method, tends to accumulate redundant information that is widely available. The difference between the two methods becomes clear when looking at an example of a query prompt that is composed of several pieces of information.

Bring Your Own Malware: ransomware innovates again

The approach taken by DragonForce and Anubis shows that cybercriminals are becoming increasingly sophisticated in the way they market their services to potential affiliates. This marketing approach, in which DragonForce positions itself as a fully-fledged service platform and Anubis offers different revenue models, reflects how ransomware operators behave like “real” companies. Recent research has also shown that some cybercriminals even hire pentesters to test their ransomware for vulnerabilities before deploying it. So it’s not just dark web sites or a division of tasks, but a real ecosystem of clear options for “consumers.” We may also see a modernization of dark web forums, which currently resemble the online platforms of the 2000s. ... Although these developments in the ransomware landscape are worrying, Secureworks researchers also offer practical advice for organizations to protect themselves. Above all, defenders must take “proactive preventive” action. Fortunately and unfortunately, this mainly involves basic measures. Fortunately, because the policies to be implemented are manageable; unfortunately, because there is still a lack of universal awareness of such security practices. In addition, organizations must develop and regularly test an incident response plan to quickly remediate ransomware activities.

Phishing attacks thrive on human behaviour, not lack of skill

Phishing draws heavily from principles of psychology and classic social engineering. Attacks often play on authority bias, prompting individuals to comply with requests from supposed authority figures, such as IT personnel, management, or established brands. Additionally, attackers exploit urgency and scarcity by sending warnings of account suspensions or missed payments, and manipulate familiarity by referencing known organisations or colleagues. Psychologs has explained that many phishing techniques bear resemblance to those used by traditional confidence tricksters. These attacks depend on inducing quick, emotionally-driven decisions that can bypass normal critical thinking defences. The sophistication of phishing is furthered by increasing use of data-driven tactics. As highlighted by TechSplicer, attackers are now gathering publicly available information from sources like LinkedIn and company websites to make their phishing attempts appear more credible and tailored to the recipient. Even experienced professionals often fall for phishing attacks, not due to a lack of intelligence, but because high workload, multitasking, or emotional pressure make it difficult to properly scrutinise every communication. 

What Steve Jobs can teach us about rebranding

Humans like to think of themselves as rational animals, but it comes as no news to marketers that we are motivated to a greater extent by emotions. Logic brings us to conclusions; emotion brings us to action. Whether we are creating a poem or a new brand name, we won’t get very far if we treat the task as an engineering exercise. True, names are formed by putting together parts, just as poems are put together with rhythmic patterns and with rhyming lines, but that totally misses what is essential to a name’s success or a poem’s success. Consider Microsoft and Apple as names. One is far more mechanical, and the other much more effective at creating the beginning of an experience. While both companies are tremendously successful, there is no question that Apple has the stronger, more emotional experience. ... Different stakeholders care about different things. Employees need inspiration; investors need confidence; customers need clarity on what’s in it for them. Break down these audiences and craft tailored messages for each group. Identifying the audience groups can be challenging. While the first layer is obvious—customers, employees, investors, and analysts—all these audiences are easy to find and message. However, what is often overlooked is the individuals in those audiences who can more positively influence the rebrand. It may be a particular journalist, or a few select employees. 

Coaching AI agents: Why your next security hire might be an algorithm

Like any new team member, AI agents need onboarding before operating at maximum efficacy. Without proper onboarding, they risk misclassifying threats, generating excessive false positives, or failing to recognize subtle attack patterns. That’s why more mature agentic AI systems will ask for access to internal documentation, historical incident logs, or chat histories so the system can study them and adapt to the organization. Historical security incidents, environmental details, and incident response playbooks serve as training material, helping it recognize threats within an organization’s unique security landscape. Alternatively, these details can help the agentic system recognize benign activity. For example, once the system knows what are allowed VPN services or which users are authorized to conduct security testing, it will know to mark some alerts related to those services or activities as benign. ... Adapting AI isn’t a one-time event, it’s an ongoing process. Like any team member, agentic AI deployments improve through experience, feedback, and continuous refinement. The first step is maintaining human-in-the-loop oversight. Like any responsible manager, security analysts must regularly review AI-generated reports, verify key findings, and refine conclusions when necessary. 

Cyber insurance is no longer optional, it’s a strategic necessity

Once the DPDPA fully comes into effect, it will significantly alter how companies approach data protection. Many enterprises are already making efforts to manage their exposure, but despite their best intentions, they can still fall victim to breaches. We anticipate that the implementation of DPDPA will likely lead to an increase in the uptake of cyber insurance. This is because the Act clearly outlines that companies may face penalties in the event of a data breach originating from their environment. Since cyber insurance policies often include coverage for fines and penalties, this will become an increasingly important risk-transfer tool. ... The critical question has always been: how can we accurately quantify risk exposure? Specifically, if a certain event were to occur, what would be the financial impact? Today, there are advanced tools and probabilistic models available that allow organisations to answer this question with greater precision. Scenario analyses can now be conducted to simulate potential events and estimate the resulting financial impact. This, in turn, helps enterprises determine the appropriate level of insurance coverage, making the process far more data-driven and objective. Post-incident technology also plays a crucial role in forensic analysis. When an incident occurs, the immediate focus is on containment. 

Read more here ...