Blahalla

Musk Lawyer’s Question for Sam Altman on the Stand: Are You Trustworthy?

www.nytimes.com · 2 months ago

Sam Altman testified that he feared Elon Musk wanted control of OpenAI and pushed to make a for profit arm that Musk would dominate, a proposal that made Altman extremely uncomfortable. Altman denied betraying the companys founding mission and rejected claims that he misled Musk about OpenAI moving toward commercial partnerships. The trial has produced internal texts, emails and diary entries that attorneys say reveal the founders conflict and have raised broader questions about AI safety, corporate control and whether chatbot and other digital records can be used as courtroom evidence.
Accelerating detection engineering using AI-assisted synthetic attack logs generation

www.microsoft.com · 2 months ago

Researchers describe methods that translate attacker behaviors and techniques into AI-assisted synthetic attack telemetry that can be generated on demand. These synthetic logs produce realistic, labeled telemetry able to trigger detections at scale while avoiding the use of sensitive production data. The capability is intended to accelerate detection engineering by enabling large-scale testing, benchmarking, and tuning of detection rules and models.
Defense at AI speed: Microsoft’s new multi-model agentic security system tops leading industry benchmark

www.microsoft.com · 2 months ago

A new multi-model agentic security system codenamed MDASH discovered 16 new vulnerabilities across the Windows networking and authentication stack, including four Critical remote code execution flaws in components such as the Windows kernel TCP/IP stack, the IKEv2 service, Netlogon, and the DNS API. The newly reported CVEs are CVE-2026-33827, CVE-2026-40413, CVE-2026-40405, CVE-2026-33824, CVE-2026-40406, CVE-2026-35422, CVE-2026-32209, CVE-2026-35424, CVE-2026-35423, CVE-2026-40414, CVE-2026-40401, CVE-2026-40415, CVE-2026-33096, CVE-2026-40399, CVE-2026-41089, and CVE-2026-41096. Ten of the vulnerabilities are kernel mode and six are user mode, the majority are reachable from a network position with no credentials, and the findings were included in the May 12, 2026 Patch Tuesday cohort. MDASH orchestrates more than 100 specialized AI agents across an ensemble of frontier and distilled models to discover, debate, and prove exploitable bugs end to end, achieved top scores on public benchmarks, and is being used by Microsoft security teams and a limited set of preview customers.
Defense at AI speed: Microsoft’s new multi-model agentic security system tops leading industry benchmark

www.microsoft.com · 2 months ago

There is a new multi-model agentic scanning harness called codename MDASH that orchestrates more than 100 specialized AI agents to discover, debate, validate, and prove vulnerabilities end-to-end and is being used by Microsoft security engineering teams in a limited private preview. There are 16 vulnerabilities reported in the May 12 2026 Patch Tuesday cohort that were found using MDASH, including CVE-2026-33827, CVE-2026-33824, CVE-2026-41089, CVE-2026-41096, CVE-2026-40413, CVE-2026-40405, CVE-2026-40406, CVE-2026-35422, CVE-2026-32209, CVE-2026-35424, CVE-2026-35423, CVE-2026-40414, CVE-2026-40401, CVE-2026-40415, CVE-2026-33096, and CVE-2026-40399. The harness found all 21 injected vulnerabilities in a private test driver with zero false positives, achieved 96 percent recall on five years of CLFS MSRC cases and 100 percent recall on tcpip.sys, and scored 88.45 percent on the public CyberGym benchmark, the top result on the leaderboard at the time of writing.
Patch Tuesday, May 2026 Edition

krebsonsecurity.com · 2 months ago

There are near record volumes of security fixes this month as AI driven code analysis accelerated discovery and prompted major vendors including Microsoft, Apple, Google, Mozilla and Oracle to speed patch releases. Microsoft released updates on May 12, 2026 addressing at least 118 vulnerabilities, including sixteen rated critical, and did not ship fixes for any already exploited zero day flaws this month. Three particularly concerning Microsoft issues are CVE-2026-41089, a critical stack based buffer overflow in Windows Netlogon that can yield SYSTEM on domain controllers without user interaction; CVE-2026-41096, a critical remote code execution in the Windows DNS client; and CVE-2026-41103, a critical elevation of privilege that allows forged credentials to bypass Entra ID. Other notable fixes include Apple shipping iOS updates that addressed at least 52 vulnerabilities, Chrome fixing 127 flaws, Firefox 150 resolving 271 vulnerabilities, and Oracle increasing its patch cadence after a quarterly update that fixed at least 450 flaws.
It's Patch Tuesday for Microsoft & Not a Zero-Day In Sight

www.darkreading.com · 2 months ago

There are no actively exploited zero day vulnerabilities for the first time in nearly two years, but Microsoft’s May 12, 2026 security update fixed 137 CVEs, 13 of which it flagged as likely candidates for exploitation and nine of which were rated critical. Two Microsoft Office Word flaws exploitable via the Preview Pane are CVE-2026-40361 and CVE-2026-40364, and CVE-2026-40464 is another remote code execution bug stemming from a type confusion issue; none require user interaction and can be triggered by a crafted document. Three near maximum CVSS 9.9 vulnerabilities are CVE-2026-42898, CVE-2026-42823, and CVE-2026-33109, with CVE-2026-42898 enabling code injection in Dynamics 365 on premises and Microsoft noting the Azure managed instance RCE was fully mitigated so no user action is required for that service. A severe Netlogon remote code execution, CVE-2026-41089, was highlighted for prioritization and administrators were advised to patch promptly and monitor Netlogon traffic and domain controller behavior for anomalous requests and crashes.
UK fines water supplier $1.3M for exposing data of 664k customers

www.bleepingcomputer.com · 2 months ago

The Information Commissioner’s Office fined South Staffordshire Water Plc and its parent South Staffordshire Plc £963,900 after a cyberattack exposed the personal data of hundreds of thousands of customers and employees. The compromise began in September 2020, the attackers remained undetected for about 20 months, and between May and July 2022 they escalated privileges to gain domain administrator access before the breach was discovered in July 2022. Exposed records published on the dark web included full names, addresses, email addresses, phone numbers, dates of birth, customer account credentials, bank account details, and employee human resources data such as National Insurance numbers, and the ICO found failures including insufficient controls to prevent privilege escalation, monitoring that covered only about five percent of the IT environment, use of obsolete software including Windows Server 2003, poor vulnerability management and missing security patches.
Here’s the latest.

www.nytimes.com · 2 months ago

NULL
Webinar: Fixing the gaps in network incident response

www.bleepingcomputer.com · 2 months ago

IT teams frequently see network incidents escalate not from lack of visibility but because manual triage and coordination across disparate monitoring, infrastructure, identity, and security systems slow response during high pressure events. On June 2, 2026 a webinar with Tines will show how automation and AI assisted workflows can reduce response times, prevent outages, and coordinate actions across systems to move incidents from alert to resolution faster. Topics include how incidents evolve from initial alert to service impact, where triage, enrichment, and routing break down, automatic enrichment of alerts with network identity and threat context, and techniques to prioritize and route incidents without manual intervention.
Signal adds security warnings for social engineering, phishing attacks

www.bleepingcomputer.com · 2 months ago

Signal has introduced new in-app confirmations and warning messages to help users detect and block phishing and social engineering attempts that can lead to account takeovers, chat and contact theft, and other fraud. The protections label unfamiliar direct message contacts with a Name not verified note and a No groups in common indicator, prompt users to confirm new requests while reminding them that Signal will never ask for registration codes, PINs, or recovery keys, and expand in-app safety tips and reminders not to respond to messages claiming to be Signal Support. Authorities including the FBI and Dutch and German agencies have linked recent attacks to Russian state sponsored actors who abused the Linked Device feature by tricking victims into scanning QR codes or sharing one time codes to link devices and steal account data.
Sam Altman has been targeted in a backlash against A.I.

www.nytimes.com · 2 months ago

There was an incident in April in which an incendiary device was thrown at Sam Altman’s San Francisco home, igniting an exterior gate but causing no injuries. Police arrested a 20-year-old suspect who allegedly made threats at OpenAI’s headquarters and was found with incendiary materials and a document opposing artificial intelligence that named tech executives. Officials and news outlets describe the attack as part of a growing backlash against artificial intelligence that has heightened scrutiny and security concerns for AI leaders and companies.
Sam Altman has been targeted in a backlash against A.I.

www.nytimes.com · 2 months ago

There is a growing backlash against artificial intelligence and OpenAI CEO Sam Altman was targeted when an incendiary device was thrown at his San Francisco home on April 10, 2026. The device ignited a small fire at an exterior gate, no one was injured, and authorities arrested a suspect who faces charges including attempted murder and attempted arson along with related federal counts. The suspect is also accused of making threats at OpenAI’s headquarters. The attack has intensified scrutiny and safety concerns around company leaders amid the ongoing trial involving OpenAI and Elon Musk.
Microsoft releases Windows 10 KB5087544 extended security update

www.bleepingcomputer.com · 2 months ago

Microsoft has released the Windows 10 KB5087544 extended security update to address the May 2026 Patch Tuesday fixes and to correct Remote Desktop warning dialog rendering issues. After installing the update Windows 10 is updated to build 19045.7291 and Windows 10 Enterprise LTSC 2021 is updated to build 19044.7291, and the release contains security updates that fix 120 vulnerabilities plus bug fixes including Secure Boot status reporting and a daylight savings time update for Egypt. Microsoft warns of a known issue that may prompt BitLocker recovery keys on systems using a specific Group Policy that includes PCR7 with certain Secure Boot and boot manager conditions tied to the Windows UEFI CA 2023 certificate and advises removing the affected Group Policy setting and then suspending and resuming BitLocker as a temporary workaround.
Over 1 Million Baby Monitors, Security Cameras Exposed Through Meari Flaws

www.techrepublic.com · 2 months ago

There are more than 1.1 million baby monitors and security cameras whose live feeds, stored images, and device data were accessible because Meari Technology's infrastructure and cloud storage lacked basic access controls. The root causes include a shared hardcoded MQTT key, weak default credentials, and publicly reachable internal servers and object storage that allowed trivial access to footage and employee and device data. Five CVEs were filed against the platform, including CVE-2026-33356, CVE-2026-33357, CVE-2026-33359, and CVE-2026-33361.
Fortinet warns of critical RCE flaws in FortiSandbox and FortiAuthenticator

www.bleepingcomputer.com · 2 months ago

There are two critical remote code execution vulnerabilities tracked as CVE-2026-44277 and CVE-2026-26083, with CVE-2026-44277 impacting FortiAuthenticator and patched in FortiAuthenticator versions 6.5.7, 6.6.9, and 8.0.3, and CVE-2026-26083 affecting FortiSandbox, FortiSandbox Cloud, and FortiSandbox PaaS WEB UI. Both flaws may allow unauthenticated attackers to execute unauthorized commands or arbitrary code via crafted requests, the first stemming from improper access control and the second from a missing authorization weakness. Fortinet indicated FortiAuthenticator Cloud is not impacted and did not report active exploitation, but similar Fortinet vulnerabilities have frequently been abused in ransomware and cyber-espionage campaigns and have drawn attention from US cybersecurity authorities.
Windows 11 KB5089549 & KB5087420 cumulative updates released

www.bleepingcomputer.com · 2 months ago

There are Windows 11 cumulative updates KB5089549 and KB5087420 for versions 25H2, 24H2, and 23H2 released May 12, 2026 that install the May 2026 Patch Tuesday security patches for 120 vulnerabilities and are marked mandatory. The updates add an Xbox mode to the desktop, expand File Explorer archive format support, introduce haptic feedback for compatible input devices, rename Drag Tray to Drop Tray, and include fixes for the taskbar, Windows Hello, printing, storage, and general reliability and performance. After installation build numbers change to 26200.8457 for 25H2, 26100.8457 for 24H2, and 22631.7079 for 23H2, and Microsoft reports no new known issues with this release.
Microsoft May 2026 Patch Tuesday fixes 120 flaws, no zero-days

www.bleepingcomputer.com · 2 months ago

There are 120 security flaws fixed in Microsoft's May 12, 2026 Patch Tuesday and no zero day disclosures this month. Seventeen are rated Critical, including 14 remote code execution, two elevation of privilege, and one information disclosure, with an overall breakdown of 61 elevation of privilege, 31 remote code execution, 14 information disclosure, eight denial of service, 13 spoofing, and six security feature bypass vulnerabilities. Noteworthy fixes include a Windows GDI remote code execution via Enhanced Metafile files CVE-2026-35421, a SharePoint Server remote code execution CVE-2026-40365, a Windows DNS Client remote code execution CVE-2026-41096, and a Microsoft 365 Copilot information disclosure CVE-2026-26164, as well as multiple Word and Excel remote code execution and Click To Run elevation vulnerabilities exploitable by opening or previewing malicious documents. Administrators are urged to deploy the updates promptly, especially Office updates, to mitigate the risk of remote code execution from malicious attachments or previewed files.
Microsoft Patches 137 Vulnerabilities

www.securityweek.com · 2 months ago

Microsoft released patches addressing 137 vulnerabilities across Azure, Windows, Dynamics 365, and the SSO Plugin for Jira and Confluence, and none have been flagged as exploited in the wild. Roughly a dozen of the fixes carry an exploitability rating of exploitation more likely, and the most severe is CVE-2026-41103, a critical elevation of privilege in the Microsoft SSO Plugin for Jira and Confluence caused by an incorrect implementation of the authentication algorithm. Two high-severity remote code execution flaws in Microsoft Word, CVE-2026-40364 and CVE-2026-40361, both with CVSS scores of 8.4, involve a type confusion and a use after free respectively, and additional high-severity issues affect Remote Desktop, the Windows kernel and drivers, Azure services, Dynamics 365, and multiple Office components.
Dead.Letter (CVE-2026-45185) How XBOW found an unauthenticated RCE on Exim

xbow.com · 2 months ago

There is CVE-2026-45185, a critical unauthenticated remote code execution vulnerability in Exim discovered by XBOW. The flaw is a use after free during TLS shutdown when GnuTLS is used, where an ungetc writes a single newline into a freed transfer buffer, corrupting allocator metadata and allowing escalation to remote code execution on default Exim 4.97 installations such as Debian and Ubuntu 24.04 LTS. XBOW used the disclosure window to test and compare human and autonomous exploit development while detailing the technical exploitation path to achieve RCE.
The Algorithm Goes to War: Inside the AI Cyberweapon Revolution That Governments Cannot Stop

novarapress.net · 2 months ago

There are four load bearing pillars of agentic AI architecture: persistent memory, tool use, execution chains, and recursive self correction, and when applied to cyber operations these agents can autonomously perform reconnaissance, CVE cross referencing, exploit selection, privilege escalation, and lateral movement, compressing red team work from days or weeks into hours. There are observable cases of diffusion into criminal ecosystems, including WormGPT and FraudGPT for phishing and malware generation, documented AI assisted zero day discovery including a DeepMind discovery in SQLite, and academic demonstrations that GPT 4 based agents can exploit one day vulnerabilities when provided with CVE descriptions, which lowers the skill floor for attackers and multiplies attack velocity. There are state level programs and persistent intrusions pre positioned inside critical infrastructure alongside institutional responses such as the NSA AI Security Center, DARPA AIxCC, and industry investments like Copilot for Security, but near term scenarios range from managed escalation to cascading infrastructure disruptions and autonomous escalation, making AI enabled cyberattack a present condition that demands immediate defensive adaptation.
Exaforce Raises $125 Million for Agentic SOC Platform

www.securityweek.com · 2 months ago

Exaforce has raised 125 million dollars in a Series B round, bringing its total funding to 200 million dollars, with participation from HarbourVest, Peak XV, Mayfield, Khosla Ventures, Seligman Ventures, and AICONIC. There is an agentic security operations platform built around autonomous AI agents called Exabots that perform detection, triage, investigation, and response for cloud and SaaS environments by ingesting and unifying high volume telemetry into a correlated data view to enable real time analysis without relying on traditional SIEM rules or manual query languages. The platform uses a multi model AI engine combining data semantics, machine learning, behavioral baselining, and large language models together with a real time knowledge graph to provide context rich alert triage, natural language search, and automated workflows such as user verification and access revocation, and the new funding will be used to enhance the product and expand operations in Japan and Europe.
Mistral AI SDK, TanStack Router hit in npm software supply chain attack

www.csoonline.com · 2 months ago

On May 11 TeamPCP executed a large software supply chain attack that rapidly compromised roughly 170 npm and PyPI packages, including the entire TanStack Router ecosystem, Mistral AI’s SDKs, Guardrails AI on PyPI, and multiple vendor namespaces such as squawk, uipath, tallyui, and beproduct. Attackers exploited GitHub Actions weaknesses and the pullrequesttarget trigger to harvest short lived OIDC tokens and hijack legitimate release pipelines, then injected the automated Mini Shai Hulud malware which propagated like a worm. The malware is designed to steal developer credentials including GitHub and npm tokens, cloud credentials, API keys, Kubernetes service accounts, and SSH keys, and it deploys a destructive dead man’s switch that attempts to delete a developer’s home directory if a stolen GitHub token is revoked. Recommended responses include auditing lockfiles for known compromised versions, pinning dependencies to known good releases, scanning for malware files, and rotating any credentials that were in use at the time of an infected import.
This is Sam Altman’s first time testifying in court.

www.nytimes.com · 2 months ago

NULL
Škoda warns of customer data breach after online shop hack

www.bleepingcomputer.com · 2 months ago

Skoda warned that attackers exploited an unspecified vulnerability in the software of its online shop, gaining unauthorized access to the store system. The breach exposed customer personal information including names, addresses, contact details, order information, and login credentials containing password hashes, while full credit card details were not stored on the compromised systems. Skoda said the vulnerability has been resolved, the incident was reported to authorities and a forensic investigation is underway, and customers were advised to watch for phishing and monitor their accounts and payment statements.
Android 17 to expand banking scam call and privacy protections

www.bleepingcomputer.com · 2 months ago

Android 17, expected to roll out in June 2026, will introduce new security and privacy features focused on device theft protection, enhanced threat detection, and defenses against banking scam calls. Banks and apps will be able to verify call authenticity via app level queries and internal bank number lists so Android can detect and automatically terminate spoofed scam calls, with initial coverage for Revolut, Itaú Unibanco, and Nubank and a promise to make the feature available on Android 11 and later. Live Threat Detection will be expanded to detect abuse techniques such as SMS forwarding misuse, concealed accessibility overlays, hidden app icons, and malicious background launches, while Advanced Protection will restrict accessibility service access, disable device to device unlocking, disable Chrome WebGPU support, and add scam detection for chat notifications. The Mark as lost feature will allow biometric locking, hide Quick Settings, and disable Wi Fi and Bluetooth to prevent thieves from disabling tracking, and other improvements include APK scanning in Chrome before install, reduced PIN guessing attempts, temporary precise location sharing, hiding SMS one time passwords from most apps for three hours, AISeal with pKVM, Pixel verification of official builds via a public ledger, carrier options to ship devices with 2G disabled, and post quantum cryptography protections.
Adobe Patches 52 Vulnerabilities in 10 Products

www.securityweek.com · 2 months ago

Adobe released patches addressing 52 vulnerabilities across 10 products, with more than half potentially enabling arbitrary code execution and application denial of service being the second most common impact. The Adobe Connect update fixes two critical flaws tracked as CVE-2026-34659 with a CVSS score of 9.6 that could allow arbitrary code execution and CVE-2026-34660 with a CVSS score of 9.3 that could allow privilege escalation. Adobe Commerce received the largest number of fixes, including ten high severity and five medium severity issues, Content Authenticity SDK was patched for 14 flaws mainly leading to denial of service, and Adobe reports no awareness of these vulnerabilities being exploited in the wild.
New Exim BDAT Vulnerability Exposes GnuTLS Builds to Potential Code Execution

thehackernews.com · 2 months ago

A use-after-free vulnerability tracked as CVE-2026-45185 and nicknamed Dead.Letter in Exim's BDAT message body parsing when using GnuTLS can cause heap corruption and enable remote code execution. The flaw is triggered when a client sends a TLS closenotify before a BDAT body transfer completes and then sends a final cleartext byte on the same TCP connection, which can write into a freed buffer and corrupt allocator metadata. All Exim versions from 4.97 through 4.99.2 built with USEGNUTLS=yes are affected, there are no known mitigations, the issue was reported May 1, 2026 by Federico Kirschbaum of XBOW, and the defect is fixed in Exim 4.99.3 so affected users should upgrade immediately.
Malicious Coding Agent Skills and the Risk of Dynamic Context | Datadog Security Labs

securitylabs.datadoghq.com · 2 months ago

There are supply chain risks in coding agents where skills bundled with a repository can introduce attacker controlled instructions that execute before the model inspects them, enabling credential exfiltration, source code reconnaissance, and other harmful actions. Claude Code can load skills from enterprise policies, personal and project directories, plugins, nested folders, and added directories, and its dynamic context feature runs shell commands during preprocessing, which can bypass model level prompt injection defenses. Datadog Security Labs demonstrated a malicious skill that used the GitHub command line to steal authentication tokens and upload them, noted that models sometimes block such prompts while dynamic context commands may already have executed, and recommended disabling shell execution in managed settings, reviewing skill directories before use, restricting allowed tools, and monitoring developer workstations and cloud workloads for suspicious activity.
Is Anthropic’s Claude Mythos Really a Cybersecurity Risk?

www.nytimes.com · 2 months ago

Claude Mythos Preview is being withheld from general release because it can find and exploit high severity software vulnerabilities at a level that could threaten widely used systems and critical infrastructure. Access is being limited to select organizations under a defensive program intended to help harden systems, but reports of unauthorized access through a third party and independent evaluations showing expert level hacking ability have reopened debate about whether withholding or releasing such tools better serves cybersecurity. Security researchers and policymakers are divided, with some warning that concentration of powerful cyber tools in private hands reduces accountability while others argue a staged defensive rollout gives defenders a necessary head start to patch vulnerabilities.
Defending consumer web properties against modern DDoS attacks

www.microsoft.com · 2 months ago

There has been a dramatic upswing in distributed denial of service attacks, with volumes rising beginning in mid March of 2024 and a shift toward stealthy application layer techniques that can reach thousands of incidents per day. Modern attacks are multi vector, often combining network floods with application layer abuse and leveraging botnets made from IoT and misconfigured cloud workloads, which makes simple network level blocklists insufficient and increases reliance on fingerprinting such as JA3 and JA4, layered controls, and strong operational visibility. Defenders are advised to adopt a defense in depth posture ranging from basic CDN and volumetric shielding to advanced edge filtering, resilient architecture with graceful degradation and load shedding, and ultimately AI powered autonomous defenses, while prioritizing core user flows and treating DDoS as a normal operating condition for internet facing services.
AI Vulnerability Research and the Fuzzer Era Déjà Vu

voidsec.com · 2 months ago

NULL
White Circle Raises $11 Million for AI Control Platform

www.securityweek.com · 2 months ago

White Circle, a Paris based startup founded in 2025, raised eleven million dollars in seed funding to build an AI control platform. The platform serves as a control layer that monitors AI inputs and outputs to identify risks such as hallucinations, harmful content, prompt injection attacks, abuse and model drift, to detect sensitive customer data leakage, to block agents from malicious actions, and to flag and ban abusive users while enforcing organization defined policies. With support for one hundred fifty languages, the company says its models improve over time and it will use the funding to accelerate product development, hire talent across the United States, the United Kingdom and Europe, and expand its customer base.
China Sought Access to Anthropic’s Newest A.I. The Answer Was No.

www.nytimes.com · 2 months ago

The latest models from Anthropic and OpenAI are extending the United States lead in frontier artificial intelligence and intensifying the strategic rivalry with China. Representatives from China reportedly sought access to Anthropic’s newest Mythos model, but Anthropic refused and informed U.S. officials, signaling that access to cutting edge AI is being treated as a national security issue. Anthropic has restricted Mythos to a small set of trusted companies because of cybersecurity and safety concerns, amid accusations that Chinese developers have been extracting capabilities from U.S. models using techniques like distillation.
EBay Rejects GameStop’s $55 Billion Takeover Bid

www.nytimes.com · 2 months ago

eBay rejected GameStop's unsolicited cash-and-stock proposal, valued at roughly fifty five to fifty six billion dollars, calling the offer "neither credible nor attractive." eBay's board said it weighed eBay's standalone prospects and, after review with legal and financial advisers, flagged uncertainty over financing, leverage and operational risks, valuation concerns, and issues with GameStop's governance and executive incentives. GameStop had proposed about one hundred twenty five dollars per share composed of fifty percent cash and fifty percent stock, a roughly twenty percent premium, but eBay indicated it would remain focused on executing its own strategy.
Undermining the trust boundary: Investigating a stealthy intrusion through third-party compromise

www.microsoft.com · 2 months ago

Microsoft Incident Response investigated a stealthy intrusion that leveraged a compromised third party IT services provider and legitimate HPE Operations Manager and HPE Operations Agent administrative tooling to execute VBScript and PowerShell payloads and deploy web shells so malicious actions blended with routine operations. Attackers used web shells Errors.aspx and a modified Signoff.aspx to maintain internet facing persistence, registered a malicious network provider mslogon DLL that captured cleartext credentials via NPLogonNotify and NPPasswordChangeNotify, and later installed a password filter passms DLL on domain controllers with an msupdate DLL module to encode and exfiltrate harvested credentials over SMB and by email. The campaign prioritized long term access through credential theft, lateral movement, and reestablishing persistence rather than exploiting software vulnerabilities, and defenders were provided mitigation and detection guidance to identify abuse of trusted operational relationships and authentication mechanisms.
RubyGems Suspends New Signups After Hundreds of Malicious Packages Are Uploaded

thehackernews.com · 2 months ago

There is a major malicious attack on RubyGems that prompted the repository to temporarily disable new account registrations after hundreds of malicious packages were uploaded. The uploaded packages mostly target the Ruby ecosystem and some include exploits capable of stealing credentials, and the security firm Mend.io said it will release further details once the incident is contained. Google reported that credentials stolen from affected environments have been monetized through partnerships with ransomware and data theft extortion groups.
BWH Hotels Says Hackers Had Access to Reservation Data for 6 Months

www.securityweek.com · 2 months ago

Threat actors had access to a web application holding BWH Hotels guest reservation data from October 14, 2025 until the intrusion was discovered on April 22, 2026. The exposed data included guest names, email addresses, phone numbers, and reservation details, and payment and other financial information were not stored in the affected system and therefore were not accessed. BWH Hotels took the compromised application offline, engaged external security experts, said the number of affected guests is unclear, and warned the stolen information could be used for scams and phishing.
Start-Up Raises $1.3 Billion for an A.I. ‘Grid’

www.nytimes.com · 2 months ago

As tech giants come to dominate the specialized hardware needed to train and run advanced artificial intelligence, a startup named Amp is developing alternative hardware to challenge that concentration. Amp aims to offer a different source of AI compute so organizations can reduce dependence on a handful of large suppliers and increase competition in the market.
Hugging Face Packages Weaponized With a Single File Tweak

www.darkreading.com · 2 months ago

Researchers at HiddenLayer found that editing a model's tokenizer.json file can take direct control of model outputs and be used to exfiltrate data by intercepting tool call arguments and redirecting URL tokens through attacker infrastructure, revealing every URL a model accesses, API parameters, and any credentials embedded in those requests. The attack was demonstrated against locally run open source models using SafeTensors, ONNX, and GGUF and could affect other platforms that run open source models such as LlamaCPP and Ollama. Models served through remote inference services are not impacted because the exploit requires modifying local files, but poisoned tokenizer files can be uploaded to public repositories and pass normal distribution pipelines because they are structurally identical to legitimate files. Recommended mitigations include scanning third party models, using cryptographic checksums and model signing for production deployments, and treating tokenizer files as part of the trusted codebase rather than as configuration.
Free OnlyFans Lure Used to Spread Cross-Platform CRPx0 Malware

www.securityweek.com · 2 months ago

CRPx0 is a complex, stealthy and persistent multi platform malware campaign targeting macOS and Windows with Linux support in development, delivered via a social engineering lure offering free OnlyFans accounts that uses a malicious zip containing a shortcut to install the malware. Its operators perform clipboard cryptocurrency theft, selective data exfiltration controlled by command and control, and a ransomware phase that generates a Fernet based AES key to encrypt files with the extension.crpx0 while excluding critical system directories, replacing the desktop wallpaper and dropping ransom notes in English, Russian and Chinese. The campaign is self updating, can deliver additional payloads, maintains a leaks site claiming dozens of victims and over ten thousand terabytes of stolen data, and includes IoCs and a MITRE ATT&CK mapping for detection and response.
Zoom Rooms and Workplace Flaws Expose Users to Elevated Access Attacks

gbhackers.com · 2 months ago

There are three newly disclosed vulnerabilities in Zoom's software suite that can allow local authenticated attackers to escalate privileges or cause information disclosure. Two high severity Windows flaws are tracked as CVE-2026-30906, an untrusted search path issue affecting Zoom Rooms for Windows before version 7.0.0 with a CVSS score of 7.8, and CVE-2026-30905, an external control of file name or path vulnerability in the Zoom Workplace VDI Plugin version 6.6.10 with a CVSS score of 7.8 that requires updating to version 6.6.11 or newer. A lower severity iOS bug, CVE-2026-30904 with a CVSS score of 1.8, is a protection mechanism failure that can lead to unauthorized information disclosure when an attacker has physical access, and users and administrators are urged to apply the latest patches immediately.
Deal Reached With Hackers to Delete Data Stolen From the Canvas Educational Platform

www.securityweek.com · 2 months ago

Instructure said it reached an agreement with the unauthorized actor involved in the Canvas cyberattack and that the data was returned with digital confirmation in the form of shred logs indicating remaining copies were destroyed, while acknowledging there is no way to be completely certain the data was erased. A hacking group calling itself ShinyHunters claimed responsibility and had threatened to leak data tied to nearly nine thousand schools and about two hundred seventy five million individuals, extending its ransom deadline after some schools entered negotiations. The breach appeared to involve student identification numbers, names, email addresses and messages on the Canvas platform, and the company reported no evidence that passwords, dates of birth, government identification or financial information were compromised. Instructure temporarily took the system offline during the investigation, which disrupted access for students and faculty during finals, and said it is working with expert vendors to perform forensic analysis, harden systems and review the data involved.
Threat Actors Abuse Vercel AI Tools to Mass-Produce Realistic Phishing Sites

gbhackers.com · 2 months ago

Threat actors are rapidly adopting generative AI platforms to scale phishing operations, and Vercel's GenAI tool v0 dev can generate fully functional websites from simple text prompts, enabling attackers to mass produce highly convincing spoofed sign in pages for services such as Microsoft, Spotify, and Facebook. Vercel's hosting and deployment model, together with a free tier and token based access to GenAI features, allows instant deployment and rapid iteration that helps evade detection, while integrations like serverless functions and Telegram bots automate credential capture and delivery. Security researchers have documented campaigns that include fake recruitment pages impersonating brands such as Adidas, Nike, and Ferrari, and warn that generative AI is lowering the technical barrier to phishing, making attacks more automated, personalized, and harder for defenders to detect.
SAP Releases Patch for Critical SQL Injection Flaw in S/4HANA

gbhackers.com · 2 months ago

On May 12, 2026, SAP released its monthly security update addressing 15 newly discovered vulnerabilities across its software portfolio, including a critical SQL injection in the ABAP enterprise search component rated 9.6 out of 10 that could allow attackers to execute arbitrary database queries. The highest severity issues are tracked as CVE-2026-34260 and CVE-2026-34263, both rated 9.6, and the update also remediates CVE-2026-34259, CVE-2026-40135, CVE-2026-40133, CVE-2026-40137, CVE-2026-0502, CVE-2026-40132, CVE-2025-68161, CVE-2026-34258, CVE-2026-27682, CVE-2026-40136, CVE-2026-40134, CVE-2026-40129, and CVE-2026-40131.
OpenAI introduces Daybreak cyber platform, takes on Anthropic Mythos

www.csoonline.com · 2 months ago

OpenAI has unveiled Daybreak, a cyber defense platform that combines its large language models and Codex agent capabilities to accelerate threat prioritization, validate likely vulnerabilities in isolated environments, generate and test patches within enterprise repositories, and provide audit ready evidence of remediation. The rollout uses a three tier model stack with GPT 5.5 for general enterprise use, GPT 5.5 with Trusted Access for defensive security workflows, and GPT 5.5 Cyber for preview access to specialized tasks such as authorized red teaming and penetration testing. Daybreak is being developed with industry partners including Cisco, Oracle, CrowdStrike, Palo Alto Networks, Cloudflare, Fortinet, Akamai, and Zscaler, and OpenAI is in discussions with the European Commission about regulated access while Anthropic is taking a more restricted approach with its Mythos project.
West Pharmaceutical Services Hit by Disruptive Ransomware Attack

www.securityweek.com · 2 months ago

West Pharmaceutical Services experienced a ransomware attack on May 4, 2026, that forced a global shutdown and isolation of affected on-premise infrastructure. Attackers exfiltrated data before deploying file-encrypting ransomware, prompting the company to restrict access to enterprise systems, activate crisis protocols, retain Palo Alto Networks Unit 42 for response, and notify law enforcement. Core enterprise systems have been restored and shipping, receiving, and manufacturing have restarted at some sites, but a timeline for full restoration has not been finalized and the company has not disclosed the type or extent of data taken or whether the incident will have a material financial impact. No ransomware group has publicly claimed responsibility and available reporting indicates the company has taken steps intended to mitigate dissemination of the exfiltrated data, suggesting possible negotiations with the attackers.
Google Says Hackers Used AI to Build Zero-Day Exploit

www.techrepublic.com · 2 months ago

Google Threat Intelligence Group disrupted a cybercrime campaign that it assesses produced a zero day exploit with likely AI assistance capable of bypassing two factor authentication in a popular open source web based system administration tool, implemented as a Python script and requiring valid credentials. The exploit code contained educational docstrings, a hallucinated CVSS score, and a structured textbook Python format that gave researchers high confidence an AI model helped discover and weaponize the flaw, and Google stated it does not believe its Gemini model was used. The vulnerability stemmed from a high level semantic logic error caused by a hard coded trust assumption that traditional scanners may miss, prompting guidance for defenders to test authentication behavior after credential compromise rather than only verifying that two factor authentication is enabled.
New TrickMo Variant Uses TON C2 and SOCKS5 to Create Android Network Pivots

thehackernews.com · 2 months ago

A new TrickMo Android banking trojan variant uses The Open Network for command and control and adds SSH tunneling and authenticated SOCKS5 proxying to turn infected phones into programmable network pivots and traffic exit nodes. ThreatFabric observed the variant between January and February 2026 actively targeting banking and cryptocurrency wallet users in France, Italy, and Austria and found it relies on a runtime loaded APK module that enables network reconnaissance commands such as curl, dnslookup, ping, telnet, and traceroute. Dropper apps masquerade as adult versions of a short video app while the malware impersonates Google Play Services and embeds a native TON proxy that resolves adnl hostnames to evade takedowns and network blocking. The trojan also contains dormant hooks for the Pine hooking framework and extensive NFC permissions, suggesting potential future capability expansion.
Cushman and Wakefield Confirms Data Breach Impacting Over 310,000 Accounts

gbhackers.com · 2 months ago

Cushman and Wakefield confirmed a data breach impacting over 310,000 accounts after the ShinyHunters extortion group posted hundreds of thousands of corporate records online following a pay or leak demand. The exposed records include corporate and personal information that could be used to facilitate identity based attacks against clients and employees. The incident highlights the growing threat of corporate extortion where attackers pressure organizations to pay to prevent public data leaks.
Apple Patches Dozens of Vulnerabilities in macOS, iOS

www.securityweek.com · 2 months ago

Apple published 11 security advisories and released updates for iOS, iPadOS, macOS, watchOS, tvOS, and visionOS, with iOS and iPadOS 26.5 addressing more than 60 CVEs including 20 WebKit issues. macOS Tahoe 26.5 resolves nearly 80 vulnerabilities and Apple also patched macOS Sequoia 15.7.7 and macOS Sonoma 14.8.7, with fixes covering flaws that can lead to information disclosure, privilege escalation, sandbox escape, Gatekeeper bypass, crashes, denial of service, and arbitrary code execution. A patch for a flaw that can be used to recover deleted chats was backported to older iOS versions after reports the FBI exploited the weakness to recover Signal messages, and Apple said it has not flagged other vulnerabilities as being exploited in the wild.