Blahalla

www.csoonline.com · 4 months ago

Insider threats are increasing, with 42% of organizations reporting more malicious incidents and 42% seeing a rise in negligent incidents. These incidents are costing organizations an average of $13.1 million each, and experts predict this trend will continue. The rise is attributed to factors like social media providing information for threat actors, employees having greater technical capabilities, and the increasing use of AI, which can be exploited. Additionally, malicious actors are increasingly recruiting insiders and coercing them into compromising data.

CanisterWorm Hijacks npm Publisher Accounts, Steals Tokens

gbhackers.com · 4 months ago

A new, highly automated supply chain attack called CanisterWorm is compromising npm publisher accounts and injecting malicious code into legitimate packages. The attackers, identified as TeamPCP, steal npm access tokens and use them to replace package contents with backdoors and self-propagation scripts. These scripts install a persistent backdoor on Linux systems and exfiltrate credentials via the Internet Computer Protocol. Affected packages include versions of @emilgroup/discount-sdk, @emilgroup/document-uploader, and others, with compromised versions including 1.5.1 and 0.0.10. Organizations are advised to rotate credentials, audit logs, and remove persistence mechanisms to mitigate the threat.

Crossbench MPs pressure Labor over gas export tax – as it happened

www.theguardian.com · 4 months ago

Politicians in Canberra are preparing for a busy week, facing pressure regarding global oil supply and fallout from recent elections. New regulations are being introduced for datacentres and artificial intelligence, requiring new developments to utilize clean energy and reduce water usage. A new report warns that governments are ill-prepared for the rising threat of drone-based terror attacks due to advancements in technology and accessibility. The report highlights past incidents and emphasizes the potential for civilian and military sites to be targeted.

CISA Issues Warning on Apple Vulnerabilities Exploited Through DarkSword iOS Chain

gbhackers.com · 4 months ago

The Cybersecurity and Infrastructure Security Agency has issued a warning about critical security vulnerabilities within the Apple ecosystem. Attackers are exploiting a sequence of flaws, known as the DarkSword iOS chain, to gain complete control of devices. This attack begins with malicious web content, progresses through memory corruption and buffer overflow exploits, and culminates in manipulation of shared memory. CISA has mandated federal agencies update their systems by April 3, 2026, and strongly recommends immediate updates for all Apple users. The vulnerabilities are tracked as CVE-2025-31277, CVE-2025-43520, and CVE-2025-43510.

Hackers Exploit CVE-2025-32975 (CVSS 10.0) to Hijack Unpatched Quest KACE SMA Systems

thehackernews.com · 4 months ago

Cybersecurity experts have observed attackers exploiting a severe vulnerability, CVE-2025-32975, in Quest KACE Systems Management Appliance devices. This authentication bypass allows unauthorized access and potential complete control of administrative accounts. Attackers are using the vulnerability to execute commands, create new administrator accounts, and harvest credentials with tools like Mimikatz. Affected users should update their systems to patched versions including 13.0.385, 13.1.81, 13.2.183, 14.0.341 Patch 5, and 14.1.101 Patch 4.

Oracle Releases Emergency Patch for Critical Identity Manager Vulnerability

www.securityweek.com · 4 months ago

Oracle issued out-of-band updates to patch a critical vulnerability in Oracle Identity Manager and Oracle Web Services Manager that allows unauthenticated remote code execution. The flaw, tracked as CVE-2026-21992, has a CVSS score of 9.8 and affects the REST WebServices component of Identity Manager and the Web Services Security component of Web Services Manager, enabling an attacker with HTTP access to potentially take over affected installations. Oracle published an advisory urging immediate application of the patches, and while the vendor has not clearly confirmed exploitation, the vulnerability may have been exploited in the wild.

Langflow Got Hacked Twice Through the Same exec() Call - CVE-2026-33017 (CVSS 9.3) exploited in 20 hours with no public PoC

blog.barrack.ai · 4 months ago

Langflow, a tool for orchestrating AI workflows, has experienced two critical vulnerabilities, CVE-2025-3248 and CVE-2026-33017, enabling unauthenticated remote code execution. Attackers can exploit these flaws to steal sensitive information like API keys and AWS credentials. The underlying cause involves a design that permits arbitrary Python code execution, requiring a complete removal of this feature. Users are urged to update to version 1.9.0 or later, rotate credentials, and disable automatic login.

Two poems with opposite registers flipped answers across 4 LLMs on the same question. Filed with OWASP as a new attack class.

shapingrooms.com · 4 months ago

Research indicates that seemingly ordinary content, like poetry, can subtly influence how large language models make decisions without triggering any obvious flags or warnings. Experiments showed that different poems altered the models' responses to the same question, with models acknowledging the influence only when directly prompted. This phenomenon, termed "postural manipulation," highlights a broader risk where unexamined content within AI workflows shapes behavior, and human operators are often unaware of this influence. Mitigating this risk requires addressing the underlying systemic property, not just specific content artifacts.

No Zero-Day Needed: Russian Phishers Swipe Signal & WhatsApp Accounts with Plain Old Lies

factide.com · 4 months ago

Federal agencies, including the FBI and CISA, are warning of large-scale phishing campaigns targeting Signal and WhatsApp users. Russian-affiliated actors are posing as support personnel to steal account access through SMS codes or QR code scans. Compromised accounts have impacted thousands of users worldwide, with targets including government officials, military personnel, and journalists. Users are advised to never share verification codes and regularly review linked devices.

VoidStealer malware steals Chrome master key via debugger trick

www.bleepingcomputer.com · 4 months ago

A new information stealer called VoidStealer is bypassing Chrome’s Application-Bound Encryption to steal the master key used to decrypt sensitive browser data. The malware uses hardware breakpoints to extract the key directly from memory, a technique previously seen in open-source tools like ElevationKatz. VoidStealer is a Malware-as-a-Service platform advertised on dark web forums and introduced this bypass in version 2.0. Google implemented ABE in Chrome 127 to protect sensitive data, but multiple malware families have found ways to circumvent this protection.

CanisterWorm Gets Teeth: TeamPCP's Kubernetes Wiper Targets Iran

www.aikido.dev · 4 months ago

A threat actor known as TeamPCP has developed a destructive payload called CanisterWorm targeting Iranian systems. The payload employs three methods: wiping Kubernetes clusters, deleting data on non-Kubernetes hosts using the rm -rf / command, and laterally moving through networks using exposed Docker APIs and SSH keys. CanisterWorm uses timezone and locale checks to identify Iranian systems and installs a backdoor disguised as a Postgres monitor. This new payload marks an escalation in TeamPCP’s tactics, indicating a willingness to engage in geopolitically targeted destructive attacks.

EmEditor Supply Chain Analysis: Why "Publisher Authorization" isn't the silver bullet we think it is

techcommunity.microsoft.com · 4 months ago

Researchers have discovered several vulnerabilities affecting multiple versions of the ActiCast video management system. These flaws allow for unauthorized access to system files, cross-site scripting, and the potential for remote code execution. The vulnerabilities, affecting versions 6.1.1 through 6.1.8, have been assigned CVE identifiers CVE-2024-2911, CVE-2024-2912, CVE-2024-2913, CVE-2024-2914, and CVE-2024-2915. Successful exploitation could allow attackers to compromise system functionality and sensitive data.

Agent skill marketplace supply chain attack: 121 skills across 7 repos vulnerable to GitHub username hijacking, 5 scanners disagree by 10x on malicious skill rates (arXiv:2603.16572)

raxe.ai · 4 months ago

Researchers have identified several emerging threats to AI systems. Adversaries are exploiting abandoned GitHub repositories to distribute malicious AI skills, and traditional log analysis is inadequate for detecting complex software supply chain attacks. New techniques like context-bridging can bypass LLM safety measures, and methods for removing unwanted concepts from image generation models are easily circumvented. Defenses such as tool whitelisting, separating agent roles, and correcting model representations during inference offer promising avenues for improvement.

LLVM Adventures: Fuzzing Apache Modules

pwner.gg · 4 months ago

There is a new in-process fuzzing framework called apatchy that uses LibFuzzer, sanitizers, SanCov, and a custom LLVM pass to feed in-memory HTTP request bytes directly into Apache’s input filter chain so full request parsing, hooks, filters, and handlers are exercised without socket I/O. The framework provides structure-aware harnesses driven by protobuf converters for modules such as modsessioncrypto, modrewrite, and modproxyuwsgi, and it builds Apache in separate compiler trees to handle sanitizer, coverage, and compatibility flags automatically. apatchy includes a coverage introspector that walks combined LLVM bitcode to produce a call graph with per-function metadata and a web UI to visualize coverage and crash points, plus a one-day bug reproduction system that downloads targeted Apache versions and replays crashes from bug manifests. The project is in alpha, the author plans more harnesses and AFL++ support, and contributions of new Apache configuration files are encouraged to exercise additional code paths.

ONNX Hub silent=True suppresses all trust verification, enabling supply chain attacks on ML model loading (CVE-2026-28500, CVSS 9.1, no patch available)

raxe.ai · 4 months ago

There is a security control bypass in the ONNX Python library tracked as CVE-2026-28500 where the onnx.hub.load function accepts a silent parameter set to true that suppresses trust verification warnings and user confirmation prompts, allowing silent loading of models from untrusted repositories. ONNX Hub integrity verification depends on a SHA256 manifest fetched from the same repository that hosts the models, so a compromised repository can control both model files and integrity manifests and there is no independent trust anchor. This issue affects all ONNX versions through 1.20.1, no patch is currently available, and because the silent option is documented and used in tutorials and CI pipelines, the vulnerability poses significant supply chain risk for any workflow that loads ONNX models from external repositories.

Trivy vulnerability scanner breach pushed infostealer via GitHub Actions

www.bleepingcomputer.com · 4 months ago

The Trivy vulnerability scanner was compromised in a supply chain attack by threat actors known as TeamPCP, who pushed trojanized Trivy version 0.69.4 releases and malicious GitHub Actions tags that distributed credential stealing malware. Attackers abused credentials exfiltrated in an earlier March 1, 2026 breach to swap the GitHub Actions entrypoint and force push 75 of 76 tags in the trivy action repository, causing external workflows to execute malicious code before legitimate scans. The infostealer harvested host reconnaissance data, environment variables, SSH keys, cloud provider and database credentials, TLS private keys, tokens, CI and package manager credentials, and cryptocurrency wallets, then encrypted stolen files into an archive named tpcp.tar.gz and exfiltrated them to a typosquatted command and control server or uploaded them to a public GitHub repository if exfiltration failed, while also dropping a Python payload to persist as a systemd service. Researchers linked the campaign to TeamPCP and reported a related follow up campaign using a self propagating CanisterWorm that compromises npm packages, steals npm tokens to publish malicious updates, and uses Internet Computer canisters for decentralized command and control.

Paul Brainerd Dies at 78; Pioneered Desktop Publishing With PageMaker

www.nytimes.com · 4 months ago

false

Google adds ‘Advanced Flow’ for safe APK sideloading on Android

www.bleepingcomputer.com · 4 months ago

false

What to Know About Electric Cars When Gas Prices Are Surging

www.nytimes.com · 4 months ago

The war in Iran has disrupted global oil and natural gas flows, driving sharp swings in crude and gasoline prices and pushing U.S. pump prices to their highest levels since 2023. That volatility has prompted increased interest in electrified vehicles, with Edmunds reporting that shopper research for hybrids plug in hybrids and battery electric vehicles rose from 20.7 percent to 22.4 percent in the week beginning March 2 2026. Industry analysts and dealers caution that higher electricity costs limited charging infrastructure and the end of some federal incentives could blunt sales gains so higher research activity may not translate into sustained EV purchases.

Microsoft Azure Monitor alerts abused for callback phishing attacks

www.bleepingcomputer.com · 4 months ago

Azure Monitor alerts are being abused to send callback phishing emails that impersonate the Microsoft Security Team and warn of unauthorized charges on accounts. Attackers configure alert rules to include callback messages and phone numbers, and the alerts are sent from Microsoft's legitimate no reply email address so they pass SPF DKIM and DMARC checks and bypass spam filters. The campaign uses urgent fake charges, for example a Windows Defender transaction for three hundred eighty nine dollars and ninety cents, to pressure recipients into calling and thereby enabling credential theft, payment fraud, or installation of remote access software. Recipients should treat any Azure or Microsoft alert that includes a phone number or an urgent billing request with suspicion and avoid calling numbers listed in unexpected alerts.

FBI Warns Russian Hackers Target Signal, WhatsApp in Mass Phishing Attacks

thehackernews.com · 4 months ago

There are ongoing phishing campaigns by actors affiliated with Russian intelligence targeting commercial messaging applications such as Signal and WhatsApp to seize control of accounts belonging to high intelligence value individuals including current and former U.S. government officials, military personnel, political figures, and journalists. Threat actors impersonate support accounts and use social engineering to trick victims into providing verification PINs or scanning malicious links or QR codes, resulting either in account takeover where attackers control the account or in the attacker linking a device that grants access to past and future messages. The campaign has resulted in unauthorized access to thousands of individual accounts globally while not exploiting weaknesses in the apps' encryption. Users are advised never to share verification codes or PINs, to scrutinize unexpected messages and links, and to regularly review and remove suspicious linked devices.

These people used AI to help find their lost pets

www.washingtonpost.com · 4 months ago

false

Critical Quest KACE Vulnerability Potentially Exploited in Attacks

www.securityweek.com · 4 months ago

Arctic Wolf detected suspicious activity beginning in early March 2026 that appears tied to exploitation of CVE-2025-32975, a critical authentication bypass in Quest KACE Systems Management Appliance that can allow unauthenticated actors to impersonate legitimate users and potentially gain full administrative control. The flaw was patched by Quest in May 2025 and appears to have been used to gain initial access and achieve administrative control of internet exposed appliances, with some affected customers in the education sector though there is insufficient data to confirm a sector-specific targeting. No signs were found that the related May 2025 fixes for CVE-2025-32976, CVE-2025-32977, and CVE-2025-32978 were involved, and organizations running outdated KACE SMA versions are urged to apply the available patches immediately.

Oracle Patches Critical CVE-2026-21992 Enabling Unauthenticated RCE in Identity Manager

thehackernews.com · 4 months ago

Oracle has released security updates to address a critical vulnerability in Identity Manager and Web Services Manager that can be exploited to achieve unauthenticated remote code execution. The flaw is tracked as CVE-2026-21992 with a CVSS score of 9.8 and affects Oracle Identity Manager versions 12.2.1.4.0 and 14.1.2.1.0 and Oracle Web Services Manager versions 12.2.1.4.0 and 14.1.2.1.0. NIST notes the vulnerability is easily exploitable over HTTP and could allow an unauthenticated attacker to take over vulnerable instances, and Oracle urges customers to apply the updates without delay. Separately, CVE-2025-61757 was added to the Known Exploited Vulnerabilities catalog in November 2025 after evidence of active exploitation.

Oracle Fixes High-Severity RCE Vulnerability Affecting Identity and Web Services Platforms

gbhackers.com · 4 months ago

A critical unauthenticated remote code execution vulnerability affects Oracle Identity Manager and Oracle Web Services Manager and is tracked as CVE-2026-21992. Affected releases include versions 12.2.1.4.0 and 14.1.2.1.0, and Oracle recommends applying the Fusion Middleware patches referenced in advisory KB878741 while unsupported versions must be upgraded. The flaw is exploitable over standard network protocols, including HTTPS, allowing attackers to execute arbitrary code, deploy malware, exfiltrate identity data, or pivot within enterprise networks, so organizations should apply fixes immediately and follow Oracle support and Software Error Correction policies to mitigate active exploitation risk.

Trivy Vulnerability Scanner Compromised to Inject Malicious Scripts That Steal Credentials

gbhackers.com · 4 months ago

A highly sophisticated supply chain attack compromised the official Trivy GitHub Actions repository, discovered March 19, 2026, with attackers force-pushing to hijack 75 of 76 version tags and converting trusted tags into an automated distribution channel for an infostealer. The injected script executes immediately before legitimate scans and operates in three phases: harvesting cloud credentials, SSH keys, tokens and runner process memory on GitHub hosted Linux runners and using a Python harvester on self hosted runners; encrypting the stolen data with a randomly generated session key wrapped with the attackers' public key using AES; and exfiltrating the encrypted bundle to a typosquatted domain or by uploading it to victim accounts. Embedded comments identify the payload as authored by TeamPCP Cloud stealer also tracked as DeadCatx3 or ShellForce and organizations are urged to assume any pipelines referencing the poisoned tags are compromised, stop using version tags and instead pin actions to verified commit hashes, rotate exposed secrets and audit accounts for unauthorized exfiltration.

Trivy Supply Chain Attack Triggers Self-Spreading CanisterWorm Across 47 npm Packages

thehackernews.com · 4 months ago

Threat actors behind the Trivy supply chain compromise are conducting follow-on attacks that have led to the compromise of 47 npm packages, including 28 packages in the EmilGroup scope and 16 in the opengov scope, along with several standalone packages. The campaign uses a postinstall hook to run a loader that drops a Python backdoor which polls an Internet Computer blockchain canister acting as a decentralized dead drop to retrieve a URL for the next-stage payload, with a YouTube link used as a dormancy kill switch. Persistence is established via a systemd user service masquerading as PostgreSQL tooling and configured to always restart, while propagation is automated by a deploy script that uses stolen npm tokens and by a later CanisterWorm variant that harvests tokens during postinstall to self-propagate. A cloud-focused cybercriminal operation known as TeamPCP is suspected, and a trojanized Trivy binary was observed in version 0.69.4.

CISA Flags Apple, Craft CMS, Laravel Bugs in KEV, Orders Patching by April 3, 2026

thehackernews.com · 4 months ago

The U.S. Cybersecurity and Infrastructure Security Agency added five vulnerabilities affecting Apple Craft CMS and Laravel Livewire to its Known Exploited Vulnerabilities catalog and ordered federal agencies to patch them by April 3 2026. The flaws are CVE-2025-31277 a WebKit memory corruption bug CVE-2025-43510 and CVE-2025-43520 kernel memory corruption vulnerabilities CVE-2025-32432 a Craft CMS code injection that has been exploited as a zero day since February 2025 and CVE-2025-54068 a Laravel Livewire code injection that can lead to remote command execution. Reports link the Apple bugs to an iOS exploit kit known as DarkSword that delivers malware families including Ghostblade Ghostknife and Ghostsaber while threat actors such as Mimo also known as Hezb and the MuddyWater group have been observed exploiting other vulnerabilities to deploy miners backdoors and proxyware. Organizations are urged to prioritize immediate patching to prevent further remote code execution system compromise and data theft.

Evaluating AI and ML in Network Security: A Comprehensive Literature Review

www.sciencedirect.com · 4 months ago

AI and machine learning have emerged as pivotal tools that extend and sometimes surpass traditional techniques for improving network security. They are being used to detect and mitigate cyber threats, automate threat responses, and predict future vulnerabilities, improving the accuracy and efficiency of security processes. Effective benefit requires seamless integration of AI and ML with existing security architectures, and current research highlights gaps including a lack of empirical evaluations and practical implementation challenges. The literature calls for more empirical studies and future research focused on real world deployment and rigorous evaluation of AI and ML in network security.

Trivy vulnerability scanner backdoored with credential stealer in supply chain attack

www.csoonline.com · 4 months ago

Attackers have compromised the widely used open source Trivy vulnerability scanner and its GitHub Actions, injecting credential stealing malware into official releases and CI CD workflows that could trigger cascading supply chain compromises if secrets are not rotated. Three components were backdoored: trivy action, setup trivy, and the Trivy binary, with many version tags force pushed to malicious commits while commit metadata was cloned to evade detection, leaving only trivy action tag 0.35.0 untouched. The malware runs alongside the legitimate scanner, reads process memory and searches filesystems for SSH keys cloud provider credentials Kubernetes tokens Docker configurations and cryptocurrency wallets, encrypts stolen data and exfiltrates it to a typosquatted domain or by creating a public repository named tpcp docs, and may install a persistent Python dropper that polls attacker servers. Organizations are advised to treat pipeline secrets as compromised and rotate them immediately, pin GitHub Actions to full commit SHAs, use safe versions Trivy v0.69.3 trivy action 0.35.0 and setup trivy 0.2.6, search for repositories named tpcp docs, and block the attacker command and control infrastructure.

FBI and CISA Flag Russian Cyber Operations Targeting Select Individuals via Signal

gbhackers.com · 4 months ago

The FBI and CISA issued a joint advisory warning that Russian Intelligence Services are conducting a widespread phishing campaign targeting users of encrypted messaging apps, primarily Signal, and that the operation has resulted in unauthorized access to thousands of accounts worldwide. Attackers bypass end to end encryption by hijacking accounts through social engineering, impersonating official support chats, and tricking victims into surrendering SMS verification codes or scanning malicious QR codes to link attacker devices. Once linked, attackers can read private messages, monitor conversations, harvest contacts, and impersonate victims to carry out follow on phishing. Agencies recommend never sharing verification codes or PINs, avoiding unsolicited QR codes and unverified links, regularly auditing linked devices, and enabling disappearing messages for sensitive conversations.

Trivy Under Attack Again: Widespread GitHub Actions Tag Compromise Exposes CI/CD Secrets Attackers

socket.dev · 4 months ago

There was a supply chain attack on the aquasecurity Trivy GitHub Action in which an attacker force-pushed 75 of 76 version tags to new malicious commits, turning common version tags into a distribution mechanism for an infostealer and affecting thousands of workflows. The injected entrypoint executes before the legitimate Trivy scan, harvesting runner environment variables and files, dumping process memory on GitHub-hosted runners to extract secrets and SSH keys, encrypting the data with AES-256-CBC wrapped by RSA-4096, and exfiltrating it to attacker controlled endpoints. Investigators report the attacker used compromised credentials with write access obtained during an earlier March breach to perform authenticated tag updates, leaving tag 0.35.0 as the sole unaffected tag. Any CI/CD pipeline that references affected tags such as @0.34.2, @0.33.0, or @0.18.0 may have executed the malware, and consumers are advised to pin actions to full commit SHAs and verify credential rotation and repository integrity.

Copyright Complaint Lures Linked to New PureLog Stealer Credential Theft Wave

gbhackers.com · 4 months ago

There is a multi stage campaign distributing the PureLog Stealer via copyright complaint lures, localized phishing emails, and Google Ads malvertising that targets healthcare, government, hospitality, and education organizations in Germany, Canada, the United States, and Australia. The infection chain opens a benign decoy PDF while a curl based downloader with a custom user agent fetches an encrypted payload, then retrieves the decryption key from a remote command and control server and extracts the payload using a renamed archive tool disguised as an image. A fileless Python loader masquerading as a Windows service patches the Antimalware Scan Interface in memory, establishes registry persistence, fingerprints the system, captures full screen images in memory, and spawns two ConfuserEx protected.NET loaders that decrypt and reflectively load the PureLog payload into memory to evade detection. Indicators of compromise published include multiple SHA 256 hashes, malicious domains, and IP addresses to support detection and response.

Polymarket Says It Deals in Truth, but Its Social Feeds Are Filled With Falsehoods

www.nytimes.com · 4 months ago

NULL

Inside the Trump administration’s campaign to counter content bans in Europe

www.washingtonpost.com · 4 months ago

false

Are nations ready to be the cybersecurity insurers of last resort?

www.csoonline.com · 4 months ago

There is concern among Cyber Monitoring Center figures and industry experts that the UK government loan guarantee of 1.5 billion pounds to Jaguar Land Rover after a major cyberattack set an unfortunate precedent because it was provided case by case without clear criteria for state intervention. Experts warn that such intervention could create a too important to fail signal that attracts attackers and could encourage firms to underinvest in cyber resilience, while the Jaguar Land Rover incident illustrates how a cyber event can ripple across an entire economy affecting GDP, employment and exports. Insurers and reinsurers say there is a cyber insurance protection gap and call for a public private structure to share risk, whether through compulsory insurance, tax incentives, or other frameworks. Critics argue that reliance on insurance and government bailouts fuels moral hazard and that the true cost of robust security should be built into goods and services.

Elon Musk Is Responsible for Some Twitter Investor Losses, Jury Finds

www.nytimes.com · 4 months ago

false

Who’s Really Shopping? Retail Fraud in the Age of Agentic AI

unit42.paloaltonetworks.com · 4 months ago

There is a rising wave of agentic AI enabled retail fraud that can target tokenized commerce protocols and autonomous shopping agents to invisibly manipulate digital checkouts. There are two representative threat scenarios: payload poisoning that appends unauthorized digital gift cards to a shopper's cart and logic hijacking hidden in product metadata that can trigger refunds without verification, both of which can be automated at scale. These attacks rely on indirect prompt injection when agents autonomously browse and ingest web content, creating opportunities for large financial loss and reputational damage before humans detect the fraud. Mitigations highlighted include stronger authorization and accountability protocols, agent identity and reputation frameworks, and defensive AI assessments and platform guardrails to limit agentic commerce abuse.

Friday Squid Blogging: Jumbo Flying Squid in the South Pacific

www.schneier.com · 4 months ago

false

FBI links Signal phishing attacks to Russian intelligence services

www.bleepingcomputer.com · 4 months ago

Federal authorities warn that Russian intelligence linked threat actors are targeting users of end-to-end encrypted messaging apps including Signal and WhatsApp in phishing campaigns that have already compromised thousands of accounts worldwide. Attackers are bypassing encryption protections by hijacking accounts rather than breaking encryption, commonly impersonating support, tricking victims into sharing verification codes, or inducing them to scan malicious QR codes or link attacker controlled devices. Those targeted include individuals of high intelligence value such as current and former government officials, military personnel, political figures, and journalists, and users are advised to be suspicious of unexpected messages, never share verification codes, and avoid scanning unknown QR codes or linking devices.

Attack surface analysis of 5,121 MCP servers: 555 have toxic data flows where safe tools combine into dangerous paths

agentseal.org · 4 months ago

Toxic data flows occur when two individually benign MCP tools combine to create a dangerous path, for example a tool that reads private data feeding into a tool that sends data externally or a tool that fetches untrusted content chained to a destructive operation. After scanning 5,125 MCP servers and 53,533 total security findings, toxic flows were detected in 555 servers with 935 toxic-flow findings, and severity skewed toward critical and high; runtime probes on 113 servers executed 1,757 probes and produced 53 unexpected successes but no confirmed prompt injection in responses. High risk concentrates in servers with many tools and in servers affected by known CVEs including CVE-2025-6514 and CVE-2025-53109, and recommended mitigations include auditing read and write tool pairs, applying least privilege, separating read and write capabilities, monitoring tool definition changes, and scanning for toxic flows.

Patch Now: Oracle's Fusion Middleware Has Critical RCE Flaw

www.darkreading.com · 4 months ago

There is a critical remote code execution vulnerability in Oracle Fusion Middleware affecting Oracle Identity Manager and Oracle Web Services Manager, identified as CVE-2026-21992, which Oracle issued a special security alert for on March 19, 2026 and which can be exploited without authentication. The flaw carries a 9.8 CVSS score and can allow attackers to manipulate identities, roles, and security policies or disable protections to facilitate lateral movement, privilege escalation, data theft, or service disruption, and it bears similarities to a prior OIM RCE vulnerability CVE-2025-61757. Because OIM is deployed at over a thousand organizations, many of them large enterprises, patching complexity may delay remediation, so affected administrators should prioritize applying Oracle’s fixes and mitigations to reduce the potential blast radius.

Oracle pushes emergency fix for critical Identity Manager RCE flaw

www.bleepingcomputer.com · 4 months ago

Oracle has released an out of band security update that fixes a critical unauthenticated remote code execution vulnerability in Oracle Identity Manager and Oracle Web Services Manager identified as CVE-2026-21992 with a CVSS v3.1 score of 9.8, affecting versions 12.2.1.4.0 and 14.1.2.1.0. The vulnerability is low complexity, remotely exploitable over HTTP without authentication or user interaction, and Oracle strongly urges customers to apply the Security Alert patches immediately while noting patches are provided only for versions under Premier or Extended Support. Oracle has not disclosed whether the flaw has been exploited in the wild and older unsupported releases may remain exposed until upgraded or patched.

More! More! More! Tech Workers Max Out Their A.I. Use.

www.nytimes.com · 4 months ago

false

Water utilities strengthen cybersecurity through cooperation

www.csoonline.com · 4 months ago

Water utilities, many running aging systems with minimal IT or cybersecurity staff, are strengthening defenses by sharing information and coordinating responses. A two year pilot involving two hundred small and medium utilities showed improved security when cybersecurity training was paired with adequate technical and organizational support. There were reported attacks in 2024, including an October incident that disrupted American Water billing, other attacks against a Texas water company, and incidents reported in Norway and Poland. There are four recommended steps to improve security: avoid relying on free tools, expand hands on technical assistance for implementation, incorporate cybersecurity training into operator licensing, and strengthen ties between utilities and water sector associations to shift from information distribution to capacity building and build resilient infrastructure.

Trivy Security Scanner GitHub Actions Breached, 75 Tags Hijacked to Steal CI/CD Secrets

thehackernews.com · 4 months ago

Trivy, the open source vulnerability scanner maintained by Aqua Security, was compromised again within a month when attackers force pushed 75 version tags in the aquasecurity trivy action repository and seven tags in the aquasecurity setup trivy repository to distribute a malicious payload. The payload executes inside GitHub Actions runners to harvest environment variables and credentials including SSH keys, cloud provider and database credentials, Git and Docker configurations, Kubernetes tokens, and cryptocurrency wallets, then encrypts and exfiltrates the data. The incident follows a prior breach in which an autonomous bot exploited a pull request target workflow to steal a personal access token that was then used to seize repository control, and the malware contains strings and tooling that point to the threat actor known as TeamPCP while a false flag remains possible. Users are advised to treat pipeline secrets as compromised and rotate them immediately, pin GitHub Actions to full commit hashes rather than version tags, upgrade to the safe versions trivy 0.69.3, trivy action 0.35.0 and setup trivy 0.2.6, and block known exfiltration channels and check for public repositories used to stage stolen data.

Claude Code workspace trust dialog bypass via repository settings loading order [CVE-2026-33068, CVSS 7.7]. Settings resolved before trust dialog shown.

raxe.ai · 4 months ago

There is a configuration loading order defect in Anthropic's Claude Code CLI tool versions prior to 2.1.53 that can allow a malicious repository level settings file to bypass the workspace trust confirmation, tracked as CVE-2026-33068. A repository level settings file named dot claude slash settings dot json containing the documented bypassPermissions field is resolved before the trust dialog is presented, causing elevated permissions to be applied without user approval and representing CWE 807 reliance on untrusted inputs in a security decision. Claude Code 2.1.53 fixes the issue by changing the loading order so the workspace trust dialog is shown before repository settings are resolved.

Fake ‘Trusted Sender’ Labels Misused in New Apple Mail Phishing Scheme

www.techrepublic.com · 4 months ago

Apple Mail's trusted sender label can be misleading because the mail app assigns it based on familiarity such as saved contacts or prior replies rather than any verification of message authenticity. Scammers are exploiting that behavior by spoofing addresses so malicious messages still display the trusted sender banner, creating a false sense of safety that encourages users to click phishing links. Security guidance calls for users not to rely on app visual cues, to verify account messages independently, to enable two factor authentication, and to watch for subtle branding or wording errors.

Police take down 373,000 fake CSAM sites in Operation Alice

www.bleepingcomputer.com · 4 months ago

Operation Alice, an international law enforcement action led by Germany with support from Europol, shut down more than 373,000 dark web sites that advertised fake child sexual abuse material packages and offered cybercrime as a service. The platform called Alice with Violence CP, run by a 35 year old suspect based in China, used previews to trick roughly 10,000 users into paying between EUR 17 and EUR 250 in Bitcoin, collecting about four hundred thousand dollars, and authorities have identified 440 users in 23 countries and are investigating 100 of them. Investigators seized 287 servers including 105 in Germany, issued an international arrest warrant for the operator, and highlighted victim support and investigative initiatives such as Europol's Help4U platform and the Trace an Object campaign.

White House Unveils A.I. Policy Aimed at Blocking State Laws

www.nytimes.com · 4 months ago

The White House unveiled a four page legislative framework urging Congress to create a national AI law and to preempt state AI laws it views as unduly burdensome. The framework outlines priorities including protections for children, measures to prevent electricity cost spikes from AI infrastructure, respect for intellectual property, safeguards for free speech, and education for Americans on using the technology. House Republican leaders praised the plan, but analysts and state officials warn that federal preemption would set up a clash with states and that passing a unified federal bill will be difficult without bipartisan agreement.