data exfiltration | Breaking Cybersecurity News | The Hacker News

A malicious tool server connected to an AI coding assistant can quietly walk off with SSH keys, environment secrets, source code, and customer data without ever sending one obviously harmful instruction. The trick can work even after a blunt version of the same theft is refused: split the request into fragments that each look routine, place them in channels the assistant already uses, and let the agent stitch them together and send the data back. The attack targets coding tools that connect to outside servers over the Model Context Protocol (MCP), the open standard that lets AI assistants call external tools. A malicious MCP server can put one fragment in a tool description and another in a tool result; some setups also support server-initiated sampling. MCP does preserve structured tool and result boundaries. But ASSET Research Group's tests show agents can still combine instructions across them in the same working context, so no single fragment has to contain the whole mal...

Microsoft has disclosed that Storm-1175 , a financially motivated threat actor linked to China, has deployed a previously undocumented ransomware strain called StormEncryptor . The use of StormEncryptor marks a shift from the adversary's previous use of Medusa ransomware, the Microsoft Threat Intelligence Team said. "StormEncryptor is written in C++ and appends the file name extension .encrypted to files it encrypts," Microsoft noted in a series of posts on Bluesky. "It then drops a ransom note named !!!README_FIRST!!!.txt to every scanned directory." Although the exact vulnerability exploited by the threat actor as part of this campaign is unclear, the tech giant said it likely involves the exploitation of CVE-2026-18577 , a newly disclosed security flaw in N-able N‑central, to obtain initial access. The vulnerability is assessed to be a patch bypass for CVE-2026-18556, both of which allow authentication bypass and account takeover in susceptible vers...

New research shows content inside an email can escape its message boundary and interfere with the webmail interface. Across attack chains spanning Outlook, Gmail, Fastmail, Proton Mail, Yahoo Mail, and AOL Mail, the techniques can capture passwords, take over third-party accounts, leak tokens, hijack trusted UI actions, and manipulate AI tools that read email. PortSwigger researcher Gareth Heyes presented the work at Black Hat USA 2026. One Outlook/Firefox chain spoofs a Microsoft sign-in screen and captures the password a recipient types. A Yahoo/AOL paste race can expose a Medium email-login token and let an attacker sign in as the victim. A Gmail/Cowork chain can exfiltrate a Slack token after prompt injection and user interaction. The paper presents proof-of-concept research and does not report malicious exploitation. Public PoCs remain available as of August 8. The researcher said Fastmail fixed two CSS mutation bugs and a Proton Mail proxy bypass stopped working when he re...

A GitHub issue opened by an account with no repository privileges was enough to execute code on the CI runners behind Anthropic's and Google's own coding-agent repositories. On OpenAI's, it was enough to hijack the next agent run. Novee Security ran the attack against each vendor's agent in the configuration that the vendor ships by default, and presented the work at Black Hat USA on August 5. Two CVEs came out of it. Both are patched. Gemini CLI carries the worst of the two. CVE-2026-12537 (CVSS 4 score: 10.0) is an OS command injection in the container launcher, reached through a crafted .gemini/.env file, which lets an unprivileged attacker run code on the host of a headless CI platform before the sandbox starts. It is fixed in Gemini CLI 0.39.1 and run-gemini-cli 0.1.22. In Claude Code, CVE-2026-54316 turned Hugging Face's public download counter into an exfiltration channel that leaked an API key one character at a time, and is fixed in 2.1.163. Every ...

A cluster of 77 extensions on the Open VSX marketplace has been found to impersonate legitimate developer tools while transmitting information about the systems and development environments on which they were installed. The "evil twin" extensions were uploaded to the repository between July 26 and August 1, 2026, according to Manifold Security. The packages have been removed from Open VSX as of August 3, 2026. "In most of the packages it sends little more than the machine's hostname," security researchers Ax Sharma and Cody Nash said . "In nineteen of them it sends a detailed description of the machine, the repository open in the editor, and the CI system the editor is running inside." Of the identified extensions, 58 have been described as lightweight tools designed to exfiltrate the hostname and, in some cases, the workspace folder name or editor version. The rest are reconnaissance payloads that transmit the developer-related information: l...

The Russian threat actors recently linked to the exploitation of a now-patched vulnerability in Zimbra have been observed exploiting another vulnerability, this time in Microsoft Outlook Web Access (OWA), to target U.S. and European government entities, as well as the telecommunications, financial, hospitality, and aerospace sectors. The activity, which began on July 22, 2026, involves the weaponization of CVE-2026-42897 (CVSS score: 8.1), a cross-site scripting (XSS) vulnerability in OWA. It was flagged by Microsoft as having been exploited in attacks as far back as May 2026. Enterprise security company Proofpoint has attributed the activity to Laundry Bear (aka CL-STA-1114, TA488, UNK_PitStop, and Void Blizzard), which was recently attributed to the zero-day exploitation of CVE-2025-66376, an XSS flaw in Zimbra's Classic UI, since at least July 2025 before it was patched four months later. In these attacks, the threat actors sent out messages from adversary-controlled P...

A newly discovered espionage implant has been using a hijacked Microsoft 365 calendar as its command channel, planting operator instructions and smuggling out stolen files as attachments on calendar events dated to the year 2050. Group-IB , which named the malware HollowGraph , says the approach moves tasking and stolen data through legitimate Microsoft Graph API traffic, so the activity looks like ordinary Microsoft 365 chatter and network controls keyed to attacker-owned destinations have nothing to flag. The implant is a .NET DLL that supports just two commands, get and send , and it never reaches out to an attacker-owned server for payloads. Instead, it treats the compromised mailbox's calendar as a two-way dead drop. To pull tasking, it queries the calendar for an event the operator has planted, buried at 2050-05-13 , so the mailbox owner is unlikely to ever scroll to it, and reads its instructions from an attached file. To exfiltrate, it does the reverse: encrypts th...

Cybersecurity researchers have flagged an intrusion in which an unknown threat actor leveraged a vibe-coded PowerShell script for Active Directory (AD) enumeration. "The script looked for the Domain Controller (DC) and mapped users, computers, and domains, before creating a directory and exporting out a number of files, and finally creating AD_Report.html to measure the success of the enumeration attempt," Huntress researchers Jevon Ang and Dray Agha said . The attack chain involved the threat actor establishing Remote Desktop Protocol (RDP) access onto a domain-joined Windows Server with a set of pre-compromised credentials, followed by staging the tools in the "C:\ProgramData\" folder. The incident took place in early June 2026. This included an artificial intelligence (AI)-generated payload to map the Active Directory environment. The assessment is based on various telltale signs, such as the prompt iteration title, placeholder strings, over-engineered cod...

Researchers at  Shandong University  have shown a fast new way to pull data off computers that are cut off from every network. The technique, called  TrojPix , tweaks on-screen pixels in ways the eye cannot see, so that the video cable carrying them radiates a faint radio signal a nearby receiver can decode. But TrojPix works only once malware is already on the target machine, so it is a way for stolen data to get out, not a way in. In the researchers' tests, TrojPix hit a peak throughput of 8.1 Mbps and reached as far as 208 meters, the two measured separately rather than together. Most air-gap covert channels crawl along at bits or kilobits per second; at 8.1 megabits, roughly a megabyte a second, TrojPix could move a 100 MB file in under two minutes. That turns the threat from leaking a password into moving whole files while the monitor looks switched off. Real-world range is another matter: a receiver still has to fight through walls, shielding, and noise. Th...

Threat actors associated with the Anubis ransomware operation have been observed exploiting the Citrix Bleed 2 (CVE-2025-5777) vulnerability to obtain initial access. "Although tactics differ between affiliates, common patterns emerged in tradecraft through use of legitimate Remote Management and Monitoring (RMM) tooling, credential access, and hands-on-keyboard procedures used for lateral movement," Arctic Wolf said in a report published this week. "Anubis affiliates repeatedly abused legitimate remote access and administration tools, including ScreenConnect, Zoho Assist, MeshAgent, Remotely, UltraVNC, and Total Software Deployment, to blend in with normal IT activity while maintaining control of victim systems." Anubis is a ransomware-as-a-service (RaaS) group that first emerged in late 2024 as a rebrand of Sphinx ransomware. The ransomware operation was formally announced on the Ransomware and Advanced Malware Protection (RAMP) underground forum in Febr...

New Microsoft research shows how attackers can hijack AI agents that act on a user's behalf, using nothing more than a poisoned tool description to make the agent quietly hand over company data to an outsider. The trick is that the agent never breaks a rule. Every step looks routine, so in a default setup no alarm may fire. The work comes from Microsoft Incident Response and its Defender security research team, and it lands as companies start letting AI do more than read and summarize. What changes when an agent can act Until recently, the workplace AI risk was mostly framed around what a model read and wrote. A poisoned document could skew an answer, and that was mostly where it ended. Agents are different. Microsoft 365 Copilot can send email, create files, and change calendars. Custom agents built in Copilot Studio or Azure AI Foundry can reach into business systems and run multi-step jobs on their own. The same injection trick that biases a summary now trigger...

Cybersecurity researchers have disclosed details of four vulnerabilities in Dify , an open-source agentic workflow platform with more than 146,000 GitHub stars , that could allow attackers to stealthily read artificial intelligence (AI) conversions from other customers' applications without requiring authentication. The vulnerabilities have been collectively codenamed DifyTap by Zafran Security. "Two were critical severity, two required no authentication, and three carried cross-tenant impact on Dify's multi-tenant cloud service, allowing one customer's data to be exposed to another," researchers Ido Shani and Gal Zaban said . The security defects could have allowed attackers to read private AI chats from other customers' applications, creating a covert exfiltration channel for every message and model response. They also made it possible to traverse Dify's internal Plugin Daemon API from unauthenticated requests and trigger cross-tenant internal ...

A single click on a trusted Microsoft link could have let an attacker pull emails, calendar details, and indexed files out of Microsoft 365 Copilot Enterprise Search. Researchers at Varonis Threat Labs chained three bugs into a one-click exfiltration path they call SearchLeak . Because the link pointed to a real microsoft.com domain, traditional anti-phishing and URL filtering tools were unlikely to flag it. No prompt, no password, no second click. Microsoft assigned CVE-2026-42824 and marked it critical; the CVSS scores ran lower and disagreed, 6.5 from Microsoft and 7.5 from the National Vulnerability Database . The company mitigated the flaw on its backend, so customers have nothing to worry about, and Varonis presented a proof-of-concept, not observed exploitation. Three bugs, one click Microsoft's advisory describes the flaw as a command injection that can expose information over a network. In practice, SearchLeak stacks one AI-specific weakness on two old web bugs, ...

Two security teams have shown, in separate research published this week, that OpenClaw , the popular self-hosted AI agent, can be driven to run attacker-controlled code or hand over sensitive data through ordinary-looking inputs. Imperva buried instructions inside shared contacts, vCards, and location pins that the agent executed without the victim ever seeing them. Varonis built a test agent on the platform, gave it a mailbox full of synthetic business data, and watched a single plain email talk it into forwarding mock AWS keys and a fake customer export to an outside address. The flaw Imperva found is patched in OpenClaw 2026.4.23, so update if you run it. The phishing weakness Varonis found is not something a patch fixes; it comes down to limiting what the agent can do on its own. Different doors into the same room: the agent trusts what reaches it, and its access becomes the attacker's. Hidden commands in a shared contact Imperva researcher Yohann Sillam looked at ...

OpenAI has begun rolling out a new Lockdown Mode to ChatGPT for eligible personal accounts to reduce the risk of data exfiltration arising from prompt injection attacks . The feature is primarily designed for people and organizations that handle sensitive data and require stricter protection guarantees. Lockdown Mode is available to logged-in users across Free, Go, Plus, and Pro, and self-serve ChatGPT Business plans. "Lockdown Mode is an optional advanced security setting that limits many tools and capabilities in OpenAI products that can connect to the web or external services," OpenAI said . "It is designed to reduce the risk of data exfiltration from prompt injection attacks by limiting outbound network requests, at the expense of disabling or limiting some useful features." The safeguards are aimed at hardening the attack surface against prompt injections, which continues to be a "frontier" problem impacting all large language models (LLMs...

Unknown attackers spent at least five months inside the Outlook mailbox of a senior executive at a major global stock exchange, copying the inbox out in small, repeated batches and routing it through Dropbox and OneDrive so the traffic blended into normal cloud activity. Symantec and Carbon Black's Threat Hunter Team reported the campaign this week. This points to espionage, not a money grab: Symantec said the commands indicate intelligence collection, not theft for profit. Neither the executive nor the exchange was named. The value is plain enough: an exchange executive's inbox can hold non-public listing details, enforcement matters, deal terms, market-moving plans, plus the executive's calendar and contacts. Five months of quiet access handed the attacker a detailed read on the executive's dealings and where the organization was heading, without needing broad access to other business systems. The first malicious activity showed up on October 10, 2025. By th...

Cybersecurity researchers are sounding the alarm about what has been described as "malicious activity" in newly published versions of node-ipc. According to Socket and StepSecurity , three different versions of the npm package have been confirmed as malicious - node-ipc@9.1.6 node-ipc@9.2.3 node-ipc@12.0.1 "Early analysis indicates that node-ipc@9.1.6, node-ipc@9.2.3, and node-ipc@12.0.1 contain obfuscated stealer/backdoor behavior," Socket said. "The malware appears to fingerprint the host environment, enumerate and read local files, compress and chunk collected data, wrap the payload in a cryptographic envelope, and attempt exfiltration through a network endpoint selected via DNS/address logic." StepSecurity said the heavily obfuscated payload is triggered when the package is required at runtime, and attempts to exfiltrate a broad set of developer and cloud secrets to an external command-and-control (C2) server. This includes 90 categories ...

Cybersecurity researchers are calling attention to a new campaign dubbed GemStuffer that has targeted the RubyGems repository with more than 150 gems that use the registry as a data exfiltration channel rather than for malware distribution. "The packages do not appear designed for mass developer compromise," Socket said . "Many have little or no download activity, and the payloads are repetitive, noisy, and unusually self-contained." "Instead, the scripts fetch pages from U.K. local government democratic services portals, package the collected responses into valid .gem archives, and publish those gems back to RubyGems using hardcoded API keys." The development comes as RubyGems temporarily disabled new account registration following what has been described as a major malicious attack. While it's not clear if the two sets of activities are related, the application security company said GemStuffer fits the "same abuse pattern," which invo...

Cybersecurity researchers have disclosed details of a new credential theft framework dubbed PCPJack that targets exposed cloud infrastructure and ousts any artifacts linked to TeamPCP from the environments. "The toolset harvests credentials from cloud, container, developer, productivity, and financial services, then exfiltrates the data through attacker-controlled infrastructure while attempting to spread to additional hosts," SentinelOne security researcher Alex Delamotte said in a report published today. PCPJack is specifically designed to target cloud services like Docker, Kubernetes, Redis, MongoDB, RayML, and vulnerable web applications, allowing the operators to spread in a worm-like fashion, aswell as move laterally within the compromised networks. It's assessed that the end goal of the cloud attack campaign is to generate illicit revenue for the threat actors through credential theft, fraud, spam, extortion, or resale of stolen access. The  What makes thi...

The threat actor known as Harvester has been attributed to a new Linux version of its GoGra backdoor deployed as part of attacks likely targeting entities in South Asia. "The malware uses the legitimate Microsoft Graph API and Outlook mailboxes as a covert command-and-control (C2) channel, allowing it to bypass traditional perimeter network defenses," the Symantec and Carbon Black Threat Hunter Team said in a report shared with The Hacker News. The cybersecurity company said it identified artifacts uploaded to the VirusTotal platform from India and Afghanistan, suggesting that the two countries may be the target of the espionage activity. Harvester was first publicly documented by Symantec in late 2021, linking it to an information-stealing campaign aimed at telecommunications, government, and information technology sectors in South Asia since June 2021, using a bespoke implant called Graphon that used the Microsoft Graph API for C2. Subsequent activity flagged in Au...