artificial intelligence | Breaking Cybersecurity News | The Hacker News

A recent analysis of enterprise data suggests that generative AI tools developed in China are being used extensively by employees in the US and UK, often without oversight or approval from security teams. The study, conducted by Harmonic Security, also identifies hundreds of instances in which sensitive data was uploaded to platforms hosted in China, raising concerns over compliance, data residency, and commercial confidentiality. Over a 30-day period, Harmonic examined the activity of a sample of 14,000 employees across a range of companies. Nearly 8 percent were found to have used China-based GenAI tools, including DeepSeek, Kimi Moonshot, Baidu Chat, Qwen (from Alibaba), and Manus. These applications, while powerful and easy to access, typically provide little information on how uploaded data is handled, stored, or reused. The findings underline a widening gap between AI adoption and governance, especially in developer-heavy organizations where time-to-output often trumps policy ...

By 2025, Zero Trust has evolved from a conceptual framework into an essential pillar of modern security. No longer merely theoretical, it's now a requirement that organizations must adopt. A robust, defensible architecture built on Zero Trust principles does more than satisfy baseline regulatory mandates. It underpins cyber resilience, secures third-party partnerships, and ensures uninterrupted business operations. In turn, more than 80% of organizations plan to implement Zero Trust strategies by 2026, according to a recent Zscaler report .  In the context of Zero Trust, artificial intelligence (AI) can assist greatly as a tool for implementing automation around adaptive trust and continuous risk evaluation. In a Zero Trust architecture, access decisions must adapt continuously to changing factors such as device posture, user behavior, location, workload sensitivity, and more. This constant evaluation generates massive volumes of data, far beyond what human teams can process alo...

Security operations centers (SOCs) are under pressure from both sides: threats are growing more complex and frequent, while security budgets are no longer keeping pace. Today's security leaders are expected to reduce risk and deliver results without relying on larger teams or increased spending. At the same time, SOC inefficiencies are draining resources. Studies show that up to half of all alerts are false positives, with some reports citing false positive rates as high as 99 percent . This means highly trained analysts spend a disproportionate amount of time chasing down harmless activity, wasting effort, increasing fatigue, and raising the chance of missing real threats. In this environment, the business imperative is clear: maximize the impact of every analyst and every dollar by making security operations faster, smarter, and more focused. Enter the Agentic AI SOC Analyst The agentic AI SOC Analyst is a force multiplier that enables organizations to do more with the team an...

Popular messaging platform WhatsApp has added a new artificial intelligence (AI)-powered feature that leverages its in-house solution Meta AI to summarize unread messages in chats. The feature, called Message Summaries, is currently rolling out in the English language to users in the United States, with plans to bring it to other regions and languages later this year. It "uses Meta AI to privately and quickly summarize unread messages in a chat, so you can get an idea of what is happening, before reading the details in your unread messages," WhatsApp said in a post. Message Summaries is optional and is disabled by default. The Meta-owned service said users can also enable " Advanced Chat Privacy " to choose which chats can be shared for providing AI-related features. Most importantly, it's made possible by Private Processing , which WhatsApp launched back in April as a way to enable AI capabilities in a privacy-preserving manner. Private Processing is de...

Google has revealed the various safety measures that are being incorporated into its generative artificial intelligence (AI) systems to mitigate emerging attack vectors like indirect prompt injections and improve the overall security posture for agentic AI systems. "Unlike direct prompt injections, where an attacker directly inputs malicious commands into a prompt, indirect prompt injections involve hidden malicious instructions within external data sources," Google's GenAI security team said . These external sources can take the form of email messages, documents, or even calendar invites that trick the AI systems into exfiltrating sensitive data or performing other malicious actions. The tech giant said it has implemented what it described as a "layered" defense strategy that is designed to increase the difficulty, expense, and complexity required to pull off an attack against its systems. These efforts span model hardening, introducing purpose-built mac...

Cybersecurity researchers have discovered a novel attack technique called TokenBreak that can be used to bypass a large language model's (LLM) safety and content moderation guardrails with just a single character change. "The TokenBreak attack targets a text classification model's tokenization strategy to induce false negatives, leaving end targets vulnerable to attacks that the implemented protection model was put in place to prevent," Kieran Evans, Kasimir Schulz, and Kenneth Yeung said in a report shared with The Hacker News. Tokenization is a fundamental step that LLMs use to break down raw text into their atomic units – i.e., tokens – which are common sequences of characters found in a set of text. To that end, the text input is converted into their numerical representation and fed to the model.  LLMs work by understanding the statistical relationships between these tokens, and produce the next token in a sequence of tokens. The output tokens are detokeni...

AI is changing everything — from how we code, to how we sell, to how we secure. But while most conversations focus on what AI can do, this one focuses on what AI can break — if you're not paying attention. Behind every AI agent, chatbot, or automation script lies a growing number of non-human identities — API keys, service accounts, OAuth tokens — silently operating in the background. And here's the problem: 🔐 They're invisible 🧠 They're powerful 🚨 They're unsecured In traditional identity security, we protect users. With AI, we've quietly handed over control to software that impersonates users — often with more access, fewer guardrails, and no oversight. This isn't theoretical. Attackers are already exploiting these identities to: Move laterally through cloud infrastructure Deploy malware via automation pipelines Exfiltrate data — without triggering a single alert Once compromised, these identities can silently unlock critical systems. You don't get a second cha...

A novel attack technique named EchoLeak has been characterized as a "zero-click" artificial intelligence (AI) vulnerability that allows bad actors to exfiltrate sensitive data from Microsoft 365 (M365) Copilot's context sans any user interaction. The critical-rated vulnerability has been assigned the CVE identifier CVE-2025-32711 (CVSS score: 9.3). It requires no customer action and has been already addressed by Microsoft. There is no evidence that the shortcoming was exploited maliciously in the wild. "AI command injection in M365 Copilot allows an unauthorized attacker to disclose information over a network," the company said in an advisory released Wednesday. It has since been added to Microsoft's Patch Tuesday list for June 2025, taking the total number of fixed flaws to 68. Aim Security, which discovered and reported the issue, said it's an instance of a large language model (LLM) Scope Violation that paves the way for indirect prompt injecti...

OpenAI has revealed that it banned a set of ChatGPT accounts that were likely operated by Russian-speaking threat actors and two Chinese nation-state hacking groups to assist with malware development, social media automation, and research about U.S. satellite communications technologies, among other things. "The [Russian-speaking] actor used our models to assist with developing and refining Windows malware, debugging code across multiple languages, and setting up their command-and-control infrastructure," OpenAI said in its threat intelligence report. "The actor demonstrated knowledge of Windows internals and exhibited some operational security behaviors." The Go-based malware campaign has been codenamed ScopeCreep by the artificial intelligence (AI) company. There is no evidence that the activity was widespread in nature. The threat actor, per OpenAI, used temporary email accounts to sign up for ChatGPT, using each of the created accounts to have one conversat...

When generative AI tools became widely available in late 2022, it wasn't just technologists who paid attention. Employees across all industries immediately recognized the potential of generative AI to boost productivity, streamline communication and accelerate work. Like so many waves of consumer-first IT innovation before it—file sharing, cloud storage and collaboration platforms—AI landed in the enterprise not through official channels, but through the hands of employees eager to work smarter. Faced with the risk of sensitive data being fed into public AI interfaces, many organizations responded with urgency and force: They blocked access. While understandable as an initial defensive measure, blocking public AI apps is not a long-term strategy—it's a stopgap. And in most cases, it's not even effective. Shadow AI: The Unseen Risk The Zscaler ThreatLabz team has been tracking AI and machine learning (ML) traffic across enterprises, and the numbers tell a compelling story. In 2024 ...

India's Central Bureau of Investigation (CBI) has revealed that it has arrested six individuals and dismantled two illegal call centers that were found to be engaging in a sophisticated transnational tech support scam targeting Japanese citizens. The law enforcement agency said it conducted coordinated searches at 19 locations across Delhi, Haryana, and Uttar Pradesh on May 28, 2025, as part of an initiative called Operation Chakra V, which aims to combat cyber-enabled financial crimes. The cybercrime syndicates, per the CBI, defrauded foreign nationals, mainly Japanese citizens, by masquerading as technical support personnel from various multinational corporations, including Microsoft.  "The syndicate operated call centers designed to appear as legitimate customer service centers, through which victims were deceived into believing that their electronic devices were compromised," the agency said . "Under this pretext, victims were coerced into transferring funds ...

A critical infrastructure entity within Ukraine was targeted by a previously unseen data wiper malware named PathWiper, according to new findings from Cisco Talos. "The attack was instrumented via a legitimate endpoint administration framework, indicating that the attackers likely had access to the administrative console, that was then used to issue malicious commands and deploy PathWiper across connected endpoints," researchers Jacob Finn, Dmytro Korzhevin, and Asheer Malhotra said in an analysis published Thursday. The attack is assessed to be the work of a Russia-nexus advanced persistent threat (APT) actor based on the tradecraft observed and the overlapping capabilities with destructive malware used in attacks against Ukraine. Talos said the commands issued by the administrative tool's console were received by its client running on the victim endpoints and then executed as a batch (BAT) file. The BAT file, in turn, consisted of a command to run a malicious Visu...

Meta on Thursday revealed that it disrupted three covert influence operations originating from Iran, China, and Romania during the first quarter of 2025. "We detected and removed these campaigns before they were able to build authentic audiences on our apps," the social media giant said in its quarterly Adversarial Threat Report. This included a network of 658 accounts on Facebook, 14 Pages, and two accounts on Instagram that targeted Romania across several platforms, including Meta's services, TikTok, X, and YouTube. One of the pages in question had about 18,300 followers. The threat actors behind the activity leveraged fake accounts to manage Facebook Pages, direct users to off-platform websites, and share comments on posts by politicians and news entities. The accounts masqueraded as locals living in Romania and posted content related to sports, travel, or local news. While a majority of these comments did not receive any engagement from authentic audiences, Met...

Fake installers for popular artificial intelligence (AI) tools like OpenAI ChatGPT and InVideo AI are being used as lures to propagate various threats, such as the CyberLock and Lucky_Gh0$t ransomware families, and a new malware dubbed Numero. "CyberLock ransomware, developed using PowerShell, primarily focuses on encrypting specific files on the victim's system," Cisco Talos researcher Chetan Raghuprasad said in a report published today. "Lucky_Gh0$t ransomware is yet another variant of the Yashma ransomware, which is the sixth iteration of the Chaos ransomware series, featuring only minor modifications to the ransomware binary." Numero, on the other hand, is a destructive malware that impacts victims by manipulating the graphical user interface (GUI) components of their Windows operating system, thereby rendering the machines unusable. The cybersecurity company said the legitimate versions of the AI tools are popular in the business-to-business (B2B) sal...