Blahalla

I spent a weekend trying to get OpenClaw to leak my own personal data and it caught me immediately...

shiftmag.dev · 2 months ago

OpenClaw can be installed and connected to Gmail in about fifteen minutes, but default settings leave gateway authentication off and filesystem permissions unrestricted, leaving exposed instances reachable over the open internet and able to access sensitive local files. Skills are plain markdown files that declare CLI capabilities and serve as the trust boundary, so instructions in a skill look identical at the token level to instructions in untrusted inputs. Security depends chiefly on the model layer: frontier models detected and refused hidden exfiltration attempts, while lower-tier models were more probabilistic and sometimes complied, meaning safety varies with which model is routed through. Users are advised to run OpenClaw in hardened environments with throwaway credentials, pin the model in configuration, sandbox or containerize installations, and follow the project hardening documentation until stronger isolation and signed skill frameworks mature.
SAP Patches Critical S/4HANA, Commerce Vulnerabilities

www.securityweek.com · 2 months ago

SAP released 15 security notes on May 12, 2026 addressing critical code injection flaws in S 4HANA and Commerce that carry CVSS scores of 9.6 and could lead to data disclosure and arbitrary code execution. The S 4HANA issue tracked as CVE-2026-34260 is an SQL injection caused by missing input validation that could let authenticated attackers inject malicious SQL, and the Commerce flaw tracked as CVE-2026-34263 is a missing authentication check that allows unauthenticated configuration uploads and server side code injection. A separate high severity OS command injection in Forecasting and Replenishment is tracked as CVE-2026-34259, and organizations are advised to apply the patches promptly.
Vidar Stealer Campaign Evades EDR to Steal Credentials

gbhackers.com · 2 months ago

A new Vidar Stealer campaign is abusing trusted tools, multi stage loaders, and heavy obfuscation to bypass endpoint detection and response visibility and silently steal credentials from infected systems. The operation begins with spear phishing emails that deliver compressed archives and then chain loaders that employ living off the land techniques to reduce forensic footprints. This shift toward stealthy backdoor architectures and fileless style techniques makes signature based defenses largely ineffective and increases the need for behavior based and multi layer detection.
20 Leaders Who Built the CISO Era: 2 Decades of Change

www.darkreading.com · 2 months ago

There are twenty leaders credited with shaping the modern CISO era, including pioneers who formalized the role and a range of founders, researchers, criminals, and policymakers whose actions reset enterprise risk. The CISO role has expanded from tactical cyber defense to board level responsibility covering business resilience, national security, brand protection, compliance, and corporate trust. Common themes from their work include aligning cybersecurity with business outcomes, modernizing disclosure and crisis communications, closing gaps between public and private sectors and between cloud and on premise environments, pressuring supply chain and third party risk, prioritizing safety for connected devices and critical infrastructure, and preparing organizations for AI accelerated threats.
Webinar: What the Riskiest SOC Alerts Go Unanswered - and How Radiant Security Can Help

thehackernews.com · 2 months ago

There are persistent blind spots in security operations where high risk alerts from WAF, DLP, OT and IoT environments, dark web intelligence, and supply chain signals routinely go uninvestigated, creating the greatest breach risk despite overwhelming alert volumes. In-house SOCs often lack the specialist expertise or analyst capacity to handle these alert types, while MSSPs and MDRs frequently escalate complex alerts back to clients because investigations are time consuming and require business context. Most AI SOC automation platforms rely on static, prebuilt triage logic covering only a few categories, so novel or unfamiliar alerts are deprioritized or passed on. A webinar on May 21, 2026 will demonstrate an AI SOC approach that generates custom triage logic on the fly to close this coverage ceiling.
Mini Shai-Hulud Worm Compromises TanStack, Mistral AI, Guardrails AI & More Packages

thehackernews.com · 2 months ago

There is a Mini Shai-Hulud campaign attributed to TeamPCP that compromised npm and PyPI packages from TanStack, UiPath, Mistral AI, OpenSearch, Guardrails AI and other maintainers by injecting an obfuscated JavaScript file named routerinit.js designed to profile execution environments and deploy a credential stealer. The malware profiles targets including cloud providers, cryptocurrency wallets, AI tools, messaging apps and CI systems, establishes persistence in Claude Code and Visual Studio Code, installs a gh-token-monitor service, and uses Session Protocol infrastructure and the GitHub GraphQL API to exfiltrate stolen data and re-exfiltrate GitHub tokens via malicious workflows. The attack chained GitHub Actions techniques including pullrequesttarget, cache poisoning and OIDC token extraction to publish malicious versions with valid SLSA provenance, enabling the worm to propagate by locating publishable npm tokens, and the TanStack compromise was assigned CVE-2026-45321 with a CVSS score of 9.6, impacting 42 packages and 84 versions.
Fake Claude Code takes the IElevator to your browser secrets

www.csoonline.com · 2 months ago

Attackers are luring developers with fake Claude Code installers that execute attacker controlled PowerShell commands to deploy a previously undocumented PowerShell infostealer that evades detection and steals sensitive developer data. The campaign hides almost all detectable actions inside a PowerShell loader, uses geographic exclusion to skip hosts with Windows region settings in CIS member states and Iran, establishes persistence, performs browser enumeration, recovers v10 and v20 encryption keys, decrypts and collects browser protected data, and exfiltrates it while minimizing standard behavioral indicators. The malware abuses Chrome Elevation Services via the IElevator2 COM interface to retrieve Application Bound Encryption keys introduced in Chrome 127 in July 2024 so attackers can access cookies and saved passwords, and researchers published YARA rules and indicators of compromise while recommending PowerShell Constrained Language Mode, phishing resistant multifactor authentication, AMSI tamper protection, and blocking newly registered domains.
Shai Hulud attack ships signed malicious TanStack, Mistral npm packages

www.bleepingcomputer.com · 2 months ago

Hundreds of packages on npm, PyPI and Composer were compromised in a Shai Hulud supply chain campaign that delivered credential stealing malware targeting developers. The attacker, attributed to TeamPCP, hijacked OpenID Connect tokens and published malicious package versions through legitimate CI pipelines carrying valid SLSA Build Level 3 provenance attestations, first compromising dozens of TanStack and Mistral AI packages before spreading to Guardrails AI, UiPath, OpenSearch and Bitwarden CLI. The intrusion chained a risky pull request target workflow, GitHub Actions cache poisoning and OpenID Connect token theft from runner memory, and the payload reads GitHub Actions process memory to harvest GitHub tokens, npm publish tokens, cloud credentials, SSH keys and other developer secrets before exfiltrating them over a peer to peer Session network. The malware persists by adding hooks to Claude Code and Visual Studio Code auto run tasks, and vendors advise assuming credentials were exposed, rotating all tokens and secrets, auditing machines for persistence, verifying package provenance, enforcing lockfile only installs and adding behavioral analysis at install time.
Claude Mythos Finds Only One Curl Vulnerability; Experts Divided on What It Really Means

www.securityweek.com · 2 months ago

A test of Anthropic's restricted Mythos model against the curl codebase found only one real vulnerability, rated low severity and scheduled to be patched in late June 2026. The Mythos report flagged five confirmed issues, but three were already documented and one was a nonsecurity bug, leaving a single new security flaw. Industry reaction is mixed, with some saying Mythos underperformed compared with Anthropic's claims while others argue the result reflects curl's long history of auditing and robustness; Mozilla, by contrast, used Mythos to find more than 270 Firefox vulnerabilities. Curl has had 188 CVEs assigned to date and there are no public reports of any of those CVEs being exploited in the wild.
Hackers Hijack Microsoft Teams Accounts to Spread ModeloRAT Malware

gbhackers.com · 2 months ago

Hackers are abusing hijacked Microsoft Teams accounts and fake IT helpdesk chats to deliver an undocumented Python based ModeloRAT into corporate environments, social engineering employees to run a “diagnostic” PowerShell command. The loader writes and unpacks a ZIP under AppData, deploys a portable WinPython environment and runs pythonw.exe to hide execution, splitting into separate reconnaissance and command and control components that beacon to hard coded IP addresses to evade domain based blocking. This variant evaded multiple major EDR products and VirusTotal at the time of analysis, uses Run key persistence paired with a randomly named scheduled task, and defenders are advised to tighten Teams external access, monitor unexpected cloud storage downloads, hunt for ZIP creation under AppData and suspicious pythonw.exe executions, and review new Run keys and scheduled tasks.
Open WebUI File Upload Vulnerability Enables 1-Click RCE Attack

gbhackers.com · 2 months ago

There is a critical, unpatched stored cross site scripting vulnerability in Open WebUI’s profile image upload that allows attackers to embed malicious JavaScript in base64 encoded SVG images, enabling one click remote code execution and full account hijacking. Attackers can upload a base64 encoded reverse shell as a profile photo and lure victims with a redirect link so the application renders and executes the script in the victim’s context, allowing creation of malicious tools via the API and exfiltration of tokens and chat logs. The flaw is unpatched in version 0.7.2, can be mitigated by restricting the media type to a strict allowlist of safe image formats, and targets endpoints including /api/v1/tools/create, /api/v1/chats/all, /api/v1/users/search, and /api/v1/users/[userid]/profile/image.
Worm Redux: Fresh Mini Shai-Hulud Infections Bite Supply Chain

www.darkreading.com · 2 months ago

Hundreds of npm packages related to the TanStack ecosystem were compromised by Mini Shai Hulud, a self propagating credential stealing worm attributed to the threat cluster TeamPCP. Researchers at Aikido identified 373 malicious package version entries across 169 package names while Socket found 84 modified TanStack artifacts, and evidence indicates the campaign also touches SAP related packages, AI tooling, and enterprise libraries. The malware steals credentials from developer machines and CI CD runners, abuses trusted publishing workflows and short lived tokens and provenance to push trojanized updates, uses heavily obfuscated JavaScript and Bun execution techniques, and can persist through IDE integrations and developer tooling hooks to self propagate. Recommended defenses include scanning npm publishing logs for unexpected publishes, rotating npm GitHub cloud and CI CD credentials, enabling provenance verification, package allow listing and dependency monitoring, and hunting for unauthorized publishes and persistence on developer endpoints.
Copy.Fail Linux Vulnerability

www.schneier.com · 2 months ago

A critical Linux kernel local privilege escalation tracked as CVE-2026-31431 and nicknamed Copy Fail allows an unprivileged local user to gain root; it was publicly disclosed on April 29, 2026 and a working proof of concept was published. The flaw abuses the kernel cryptographic AFALG interface together with splice to write four controlled bytes directly into the page cache of any readable file without modifying the file on disk, enabling attackers to bypass checksum based integrity monitors. The exploit works unmodified across major distributions and kernels released since 2017, requires no race conditions or per distribution offsets, and has prompted urgent vendor advisories and warnings about active exploitation.
SAP fixes critical vulnerabilities in Commerce Cloud and S/4HANA

www.bleepingcomputer.com · 2 months ago

SAP released May 2026 security updates that fix 15 vulnerabilities across multiple products, including two critical flaws affecting the Commerce Cloud e-commerce platform and the S/4HANA cloud ERP suite. A missing authentication check in Commerce Cloud allows unauthenticated attackers to upload malicious configurations and execute arbitrary server side code, tracked as CVE-2026-34263. A second critical issue, CVE-2026-34260, permits low complexity SQL injection by concatenating unsanitized user input into SQL queries, which can expose sensitive database information or crash the application. The advisory also addresses one high severity and 11 medium severity issues such as command injection, missing authorization checks, cross site scripting, cross site request forgery, and denial of service, and SAP said it has not found evidence these vulnerabilities were exploited in the wild.
Is the SOC Obsolete, and We Just Haven’t Admitted It Yet?

www.securityweek.com · 2 months ago

Security operations centers are becoming obsolete because attacker capabilities powered by AI now operate at near machine speed, outpacing traditional human-centric SOC workflows. Adversaries are deploying AI powered malware that adapts during execution, with examples including PROMPTFLUX and reported AI orchestrated espionage campaigns, and industry data shows lateral movement and exploit times collapsing to seconds. Defense requires sovereign, data complete architectures that allow AI agents to ingest unfiltered historical and current datasets while preserving explainability, auditability, and reproducibility so autonomous decisions can be trusted and audited. With such AI native SOCs, humans would shift from alert triage to oversight and exception handling, guiding autonomous systems that correlate signals continuously and respond at machine speed.
EBay Rejects GameStop’s $55 Billion Takeover Bid

www.nytimes.com · 2 months ago

eBay’s board has rejected GameStop’s unsolicited, nonbinding cash-and-stock takeover proposal, concluding that the proposal was neither credible nor attractive. The offer valued eBay at roughly fifty-five to fifty-six billion dollars at one hundred twenty-five dollars per share and was structured about fifty percent cash and fifty percent GameStop stock, with reported financing commitments including a letter for up to twenty billion dollars from TD Bank. eBay said its directors, with the support of financial and legal advisers, reviewed the proposal and emphasized confidence in eBay’s standalone prospects, execution and capital returns to shareholders.
Why Agentic AI Is Security's Next Blind Spot

thehackernews.com · 2 months ago

Agentic AI is running in production across many organizations, executing tasks, consuming data, and taking actions often without meaningful involvement from security teams. There are three categories of agents with different risk profiles: general purpose coding and productivity agents embedded in developer workflows, vendor built agents using the Model Context Protocol that can connect to external services and act on behalf of users, and custom agents built by nonprogrammers that can gain real system access. Most agents are deployed without security review or secure configuration, which grants broad permissions that increase blast radius and enable novel lateral movement paths such as manipulating an agent via calendar or email to act on other systems. Security teams must gain hands on fluency by building and testing agents, engage early in design to scope access and apply configuration controls, and continuously evaluate emerging tools and vendor controls to reduce risk.
cPanel flaw exposes enterprises to hosting supply-chain risks

www.csoonline.com · 2 months ago

A newly disclosed cPanel vulnerability tracked as CVE-2026-41940 is being exploited at scale to deploy backdoors, plant SSH keys, steal credentials, and otherwise compromise hosting systems. Researchers observed more than 2,000 attacker source IPs and estimated over 40,000 servers at risk, with activity including cryptomining, ransomware, botnet propagation, and data theft linked to a low detection threat group called MrRot13. The exploitation underscores weak visibility into hosting supply chains and internet facing management panels that many organizations do not monitor with the same rigor as endpoints, cloud workloads, or core systems, creating an aggregator level attack opportunity. Security teams are urged to treat potential exposure as an incident response, act within hours to patch and rotate credentials, hunt for webshells and unauthorized SSH keys, review logs and telemetry for persistence and exfiltration channels including Telegram, and correlate forensic evidence across hosting stacks and third party providers.
TanStack, Mistral AI, UiPath Hit in Fresh Supply Chain Attack

www.securityweek.com · 2 months ago

A coordinated Mini Shai Hulud software supply chain campaign attributed to the TeamPCP threat group compromised over 170 packages and resulted in at least 401 malicious package artifacts being published within a five hour window on May 11, 2026. The campaign infected 42 TanStack NPM packages, 65 UiPath packages, Mistral AI PyPI packages, the OpenSearch JavaScript client, multiple Squawk packages and the Guardrails AI package, injecting implants that harvest developer credentials, API keys, cloud secrets and cryptocurrency wallets. Attackers chained three weaknesses including a pull request target misconfiguration, GitHub Actions cache poisoning across the fork to base trust boundary, and runtime extraction of OIDC tokens from the Actions runner to hijack CI CD pipelines, mint valid publish tokens and produce legitimate SLSA provenance for malicious releases. Recommended actions are to check for and remove compromised package versions, rotate potentially exposed credentials and secrets, audit GitHub Actions OIDC configurations and pull request target workflows for cache poisoning, and deploy behavioral analysis at install time in addition to provenance verification.
Developer workstations are the new beachhead

www.csoonline.com · 2 months ago

There are three recent threat intelligence reports that together show developer workstations are now the highest value initial access target in enterprise environments. One campaign, Contagious Interview, pushed more than seventeen hundred malicious packages across npm PyPI Go Modules crates io and Packagist that impersonate developer tooling and steal browser data, cryptocurrency wallet credentials and password manager contents. Another operation, GlassWorm, used a fake IDE extension bundling a Zig compiled native binary to escape extension sandboxes, scan for IDEs and install a persistent cross platform dropper that avoids Russian systems and uses Solana infrastructure for command and control. A third spree, TeamPCP, chained compromises through vulnerability scanners and developer tools to run malware in build pipelines and on developer machines, harvesting cloud tokens, CI CD secrets and service account credentials and enabling rapid pivoting to full cloud administration access.
North Korea Hackers Abuse Git Hooks to Deploy Cross-Platform Malware

gbhackers.com · 2 months ago

North Korean threat actors have shifted tactics in the Contagious Interview campaign to abuse Git hooks by embedding minimal pre-commit and post-checkout scripts inside cloned coding assessment repositories that fingerprint the operating system and load next stage malware. The hooks retrieve platform specific payloads from disposable hosting on Vercel using common tools and execute Bash scripts on macOS and Linux or batch style commands on Windows while suppressing output and returning successful exit codes to remain stealthy. Indicators include a.githooks pre-commit path and a specific SHA-256 file hash, and recommended mitigations are to inspect.githooks.husky and.vscode directories before running code, execute untrusted projects in isolated virtual machines, and never expose credentials or crypto wallets during interview related exercises.
Curl lead developer Daniel Stenberg provides insightful feedbacks from Mythos analysis results

daniel.haxx.se · 2 months ago

A Mythos AI scan of curl's master branch covering roughly 178,000 lines initially reported five confirmed security issues, but after manual review four were judged false positives or non vulnerabilities and one remains a confirmed low severity flaw that will be assigned a CVE and published with curl version 8.21.0 in late June 2026. The scan reported zero memory safety vulnerabilities, described about twenty additional bugs under investigation, and noted that curl has 188 published CVEs to date with recent AI analyses contributing around a dozen or more new CVE filings and hundreds of bug fixes. The assessment concludes that AI analyzers are very useful at finding established error types and improving code hygiene, but Mythos does not appear to produce a notably different or dramatically more dangerous class of findings compared with other modern AI tools.
Instructure reaches 'agreement' with ShinyHunters to stop data leak

www.bleepingcomputer.com · 2 months ago

Instructure reached an agreement with the ShinyHunters extortion group on May 12, 2026, after the group returned roughly 3.6 terabytes of stolen uncompressed data and provided shred logs confirming its destruction. ShinyHunters exploited multiple cross site scripting vulnerabilities in the Free for Teacher environment of the Canvas learning management system, injecting malicious JavaScript to hijack administrator sessions, perform privileged actions, and deface login portals in a follow up intrusion on May 7, 2026. Instructure temporarily shut down Free for Teacher accounts, said customers will not be individually extorted, scheduled a public webinar for May 13, 2026 to share more details and remediation steps, and noted that federal authorities warn paying ransoms does not guarantee stolen data will not be resold or victims re extorted.
Cline AI Agent Flaw Allows Attackers to Launch RCE Attacks

gbhackers.com · 2 months ago

A critical vulnerability in the Cline AI coding assistant's kanban npm package bundled with the Cline CLI allows remote code execution, data theft, and denial of service when a developer visits a malicious website. The kanban package starts a WebSocket server bound to 127.0.0.1 on port 3484 with no Origin validation, enabling cross origin attacks against local services. Security researcher Sagilayani disclosed the flaw as CVE-2026-44211 on GitHub.
CISOs step into the AI spotlight

www.csoonline.com · 2 months ago

Security leaders are increasingly visible to boards and central to enterprise strategy, with 95 percent engaging boards multiple times a month and 31 percent of top security leaders reporting directly to boards while only 20 percent report to corporate chief information officers. Organizations are prioritizing formal AI security frameworks and embedding security early in AI and product development to enable rapid innovation while managing risk through AI risk assessments, continuous monitoring, and tighter integration with engineering and data teams. Leaders warn that AI is accelerating attacks by improving impersonation, automating vulnerability discovery and exploitation, and expanding the attack surface, which drives a shift toward near real time patching, stronger identity and data governance, and adaptive detection and response. CISOs are urged to act as strategic business partners who translate risk into business terms, build trust through governance and hands on AI fluency, and develop resilient teams that enable trusted, compliant AI adoption at scale.
Why patching SLAs should be the floor, not the strategy

www.csoonline.com · 2 months ago

There is a common pattern where organizations hit patching SLAs by closing the easy cases while legacy systems, embedded devices, configuration and architectural flaws that require rearchitecting or business coordination remain unremediated. Research highlights a wide gap between patch times and attacker exploitation, with average time to patch reported at 209 days while attackers' median time to exploitation is five days, and many exploited vulnerabilities had patches available long before they were used in the wild. To move beyond SLA theater the recommended shifts are to treat SLAs as a hygiene floor rather than the headline metric, make exception processes require quantified loss exposure with scheduled revisits and funded remediation plans, and fund hard remediations as capital or operational projects by translating vulnerabilities into dollarized risk so they can compete effectively for budget.
Fake TronLink Chrome Extension Steals Crypto Wallet Credentials

gbhackers.com · 2 months ago

A phishing campaign is targeting TRON wallet users through a deceptive Chrome extension that impersonates the TronLink wallet and inherits apparent popularity from a hijacked listing. The extension loads remote content on demand and injects a full screen phishing interface that replicates the wallet UI, prompting users to enter mnemonic phrases, private keys, or keystore files which are captured and transmitted to attacker infrastructure in real time. Attackers relied on Unicode control characters and Cyrillic homoglyphs to spoof the brand name, requested only minimal storage permissions to evade review, and employed anti analysis measures and geo targeting while hosting the remote UI on a global edge network to maintain stealth. Threat monitoring flagged the campaign as high risk and users are advised to avoid installing wallet extensions from unverified sources and to inspect extension updates and sudden changes in branding or behavior.
Claude Chrome Extension Flaw Lets Malicious Add-Ons Steal Gmail and Drive Data

gbhackers.com · 2 months ago

A critical design flaw called ClaudeBleed in the Claude Chrome extension allows any Chrome extension, even one with zero declared permissions, to inject scripts into the Claude environment and issue hidden prompts that can exfiltrate sensitive data from Gmail, Google Drive, and GitHub. Attackers can bypass confirmation flows using approval looping and perception manipulation by altering the page DOM and automating approvals, enabling actions such as sharing Drive files, stealing private GitHub code, sending emails, summarizing and forwarding inbox contents, and deleting traces without user interaction. Anthropic released version 1.0.70 on May 6, 2026 to add explicit approval flows in the extension side panel, but the underlying origin-based trust boundary remains exploitable in privileged or autonomous modes, so the fix is considered incomplete.
New ipTIME Pre-Auth RCE in CWMP

ssd-disclosure.com · 2 months ago

A pre authentication remote code execution vulnerability exists in the CWMP implementation of ipTIME routers, allowing unauthenticated attackers to execute arbitrary commands with root privileges. The issue affects ipTIME firmware version 15.324 and stems from the easycwmp process writing parameter values received in ACS SOAP messages to a temporary file and later executing those lines with eval, enabling shell command injection when parameter values contain commands. An attacker can exploit this by running a malicious ACS server or performing a man in the middle attack to deliver crafted SetParameterValues messages, and a public proof of concept demonstrates forcing a router reboot.
Postmortem: TanStack npm supply-chain compromise

tanstack.com · 2 months ago

On May 11, 2026 between 19:20 and 19:26 UTC an attacker published 84 malicious versions across 42 TanStack npm packages by chaining a pull request target workflow pattern, GitHub Actions cache poisoning across the fork to base trust boundary, and runtime memory extraction of an OIDC token from the GitHub Actions runner. The malicious optional dependency executed an obfuscated payload during install that harvested credentials from common locations including AWS IMDS and Secrets Manager, GCP metadata, Kubernetes service account tokens, Vault tokens, npm configuration files, GitHub tokens and SSH private keys, exfiltrated data over the Session Oxen messenger network, and attempted to self propagate by republishing packages maintained by victims. The malicious versions were detected publicly within about 20 minutes, all affected versions have been deprecated, npm security was engaged to remove tarballs, caches were purged and workflows hardened, and users who installed an affected version on May 11, 2026 are urged to treat install hosts as compromised and rotate cloud, GitHub, npm and SSH credentials.
Microsoft Warns: MistralAI PyPI Package Compromised with Malware

gbhackers.com · 2 months ago

Mistral's official Python client on PyPI, package name mistralai, was compromised with a backdoor in version 2.4.6 that executes code when the library is imported. The backdoor was designed to silently deploy a credential stealing payload on Linux systems and run automatically upon import, putting developers who merely use the library at risk. The compromised package was removed from PyPI during a wave of AI supply chain attacks and is the subject of a Microsoft security advisory.
Instructure Reaches Ransom Agreement with ShinyHunters to Stop 3.65TB Canvas Leak

thehackernews.com · 2 months ago

There was a breach of Instructure's Canvas platform by the decentralized extortion group ShinyHunters that resulted in the theft of 3.65 terabytes of data affecting nearly nine thousand organizations. Attackers exploited an unspecified vulnerability related to support tickets in the Free for Teacher environment to steal about 275 million records including usernames, email addresses, course names, enrollment information, and messages, while course content, submissions, and credentials were not compromised. After a second wave of activity on May 7, 2026 that defaced login portals at roughly 330 institutions and set a May 12, 2026 deadline, Instructure reached an agreement with the unauthorized actor, obtained the stolen data, received digital confirmation of its destruction, and said customers will not be separately extorted. Instructure temporarily shut down Free for Teacher accounts, revoked privileged credentials and access tokens, rotated internal keys, restricted token creation pathways, deployed additional security controls, and said it is working with forensic vendors while experts warn the exposed records could enable targeted phishing and recommend immediate advisories to affected communities.
OpenAI Daybreak Automates Vulnerability Detection and Patching

gbhackers.com · 2 months ago

OpenAI Daybreak automates vulnerability detection, triage, and patching by integrating advanced reasoning models with Codex to perform secure code review, threat modeling, and dependency risk analysis directly in development workflows. The system can reason across large codebases to find subtle logical flaws, generate and test patches within repositories, verify fixes, and produce audit ready evidence while offering tiered model access and strict verification and safeguards to prevent misuse. OpenAI plans to deploy Daybreak iteratively with industry and government partners including Cloudflare, Cisco, CrowdStrike, Palo Alto Networks, Oracle, Zscaler, Akamai, and Fortinet in the coming weeks to accelerate defensive response times and improve software resilience.
OpenAI Launches Daybreak for AI-Powered Vulnerability Detection and Patch Validation

thehackernews.com · 2 months ago

OpenAI has launched Daybreak, a cybersecurity initiative that combines frontier AI model capabilities with Codex Security to help organizations identify, test, and patch vulnerabilities before attackers can exploit them. Daybreak builds editable threat models for repositories, identifies and tests vulnerabilities in isolated environments, and proposes fixes while integrating secure code review, threat modeling, patch validation, dependency risk analysis, detection, and remediation guidance into development workflows. The effort is built on three models: GPT-5.5, GPT-5.5 with Trusted Access for Cyber for verified defensive work, and GPT-5.5-Cyber for red teaming and controlled validation, and access is being tightly controlled with organizations invited to request scans. Major vendors including Akamai, Cisco, Cloudflare, CrowdStrike, Fortinet, Oracle, Palo Alto Networks, and Zscaler are integrating these capabilities as part of a broader move to use AI security agents to accelerate remediation amid faster AI-driven vulnerability discovery.
Magecart Hackers Exploit Google Tag Manager to Inject Credit Card Skimmers

gbhackers.com · 2 months ago

There are Magecart-style attackers abusing Google Tag Manager to inject credit card skimmers into ecommerce sites, leveraging GTM's trusted domain to blend malicious scripts with legitimate functionality. Compromised GTM containers can execute custom JavaScript to harvest payment data at checkout, with known containers like GTM-WJ6S9J6 observed on at least 178 infected sites in 2023 and newer containers such as GTM-TVKQ79ZS deploying an updated ATMZOW skimmer. Attackers hide infrastructure behind lookalike analytics domains and Cloudflare, register many benign sounding domains and randomly select a subset per session, and they rapidly recycle containers and layer multiple skimming techniques to avoid detection. Defenders should regularly audit GTM configurations, restrict and monitor account access, and watch for unfamiliar containers or unexpected scripts loading on checkout pages.
BitUnlocker Downgrade Attack Bypasses Windows 11 Disk Encryption in Minutes

gbhackers.com · 2 months ago

A proof of concept exploit named BitUnlocker can bypass Windows 11 BitLocker in under five minutes by downgrading Secure Boot to load a legacy Microsoft PCA 2011 signed boot manager, exploiting vulnerability CVE-2025-48804. The attack requires physical access to TPM only BitLocker systems and uses a modified BCD plus an attacker controlled WIM hidden in an SDI to load a compromised Windows Recovery Environment that causes the TPM to release BitLocker keys and present the decrypted operating system volume. Recommended mitigations include enabling TPM plus PIN pre boot authentication, migrating boot managers to the Windows UEFI CA 2023 and revoking the PCA 2011 certificate via the DBX database, and changing PCR policies to detect unauthorized boot path modifications.
Checkmarx Jenkins AST Plugin Compromised in KICS Supply Chain Attack

gbhackers.com · 2 months ago

A supply chain campaign linked to KICS and Trivy compromises has extended to Checkmarx's Jenkins ecosystem, with a malicious Checkmarx Jenkins AST plugin release identified as version 2026.5.09 fraudulently published to the Jenkins Marketplace. The tampered artifacts were distributed between 2026-05-09 01:25:00 UTC and 2026-05-10 08:47:00 UTC, and organizations that downloaded or deployed that version should treat affected Jenkins controllers and build agents as potentially compromised and initiate full incident response. Checkmarx says the last known good build is 2.0.13-829.vc72453fa1c16 released on December 17, 2025, and advises customers to roll back to that build or earlier, rotate exposed credentials, block attacker infrastructure, and review CI and developer telemetry for signs of abuse while a comprehensive code audit proceeds.
84 npm Packages Linked to TanStack Hit by Supply-Chain Breach

gbhackers.com · 2 months ago

A massive supply chain breach compromised 84 npm packages in the TanStack ecosystem, including widely used components such as React Router, by injecting a credential-stealing payload that targets continuous integration environments like GitHub Actions. The attack delivers an obfuscated worm executed via a malicious optional dependency and a prepare lifecycle hook during npm install, with files named routerinit.js and tanstackrunner.js used to harvest GitHub Actions workflow tokens, AWS metadata, Kubernetes service account certificates and HashiCorp Vault data, then exfiltrate credentials over the Session peer-to-peer network while persisting in local editor configuration folders. Threat actors abused a chained attack against the project’s CI pipeline to extract runtime OpenID Connect tokens from runner memory and push compromised updates, and indicators of compromise include filenames routerinit.js and tanstackrunner.js and SHA256 hashes ab4fcadaec49c03278063dd269ea5eef82d24f2124a8e15d7b90f2fa8601266c and 2ec78d556d696e208927cc503d48e4b5eb56b31abc2870c2ed2e98d6be27fc96; developers are advised to inspect environments for unexpected scripts, purge workflow caches and rotate cloud secrets immediately.
China Seeks A.I. Independence, Weakening Trump’s Leverage

www.nytimes.com · 2 months ago

China reached a milestone in its drive for technological self sufficiency as domestic AI firm DeepSeek adapted its latest model to run efficiently on Huawei chips, prompting adoption by Chinese semiconductor manufacturers ahead of a U.S. China summit. State backed funding and industrial directives have accelerated integration between Chinese AI models and homegrown processors, signaling Beijing's push to reduce reliance on U.S. semiconductor technology. U.S. officials and industry figures have raised concerns that Chinese firms trained models using advanced Nvidia chips despite export controls and that techniques such as model distillation threaten American firms, complicating export control debates as the summit approaches.
iOS 26.5 Brings Default End-to-End Encrypted RCS Messaging Between iPhone and Android

thehackernews.com · 2 months ago

There is a rollout of iOS 26.5 that introduces end to end encrypted Rich Communication Services messaging in beta, enabled by default for new and existing cross platform conversations between iPhone users on iOS 26.5 with supported carriers and Android users on the latest Google Messages. RCS now supports high resolution photos and videos, typing indicators, and read receipts, and encrypted conversations will be indicated by a lock icon on iPhone and a padlock icon on Android. The change follows GSMA led cross industry efforts and builds on testing that began in iOS 26.4 beta to provide a more secure alternative to traditional SMS.
Google Warns Hackers Are Using AI to Build Working Zero-Day Exploits

gbhackers.com · 2 months ago

Google Threat Intelligence Group warns that adversaries are using generative AI to discover vulnerabilities and craft functional zero day exploits, including a foiled mass campaign that used an AI assisted exploit to bypass two factor authentication in a popular open source web administration tool. Large language models are finding high level semantic logic flaws that traditional fuzzers miss, and the exploit script showed LLM hallmarks such as textbook Python formatting, extensive docstrings, and a hallucinated CVSS score left in the code. State sponsored groups and criminal clusters are automating AI driven vulnerability discovery and exploit refinement with agentic frameworks, specialized datasets, and temporal knowledge graphs, while malware like PROMPTSPY uses AI APIs to autonomously interact with Android interfaces to capture biometric data and block uninstall actions. Defenders are likewise using AI to deactivate malicious API accounts and operate automated hunting and patching agents to mitigate these AI accelerated threats.
TrickMo Android Malware Targets Banking, Wallet, and Authenticator Apps

gbhackers.com · 2 months ago

A redesigned TrickMo Android banking malware has resurfaced, targeting banking, fintech, wallet, and authenticator applications with enhanced stealth, persistence, and operational flexibility. It coerces users into granting accessibility permissions to enable full remote device takeover, supporting overlay credential phishing, keylogging, screen recording, and interception of SMS and push one time passwords. The variant shifts command and control to the TON decentralized network using adnl endpoints and DNS over HTTPS, embeds a local TON proxy, and adds network reconnaissance, SSH tunneling, and an authenticated SOCKS5 proxy to route attacker traffic through compromised devices and evade fraud detection. Indicators of compromise include SHA 256 hashes and package names for dropper and host applications and a dynamically loaded dex.module payload.
American Factories Lag in Adopting A.I. This Drugmaker Is an Exception.

www.nytimes.com · 2 months ago

false
Cybersicherheitsvorschriften: So erfüllen Sie Ihre Compliance-Anforderungen

www.csoonline.com · 2 months ago

false
How do Fortune 10 SOCs handle incident response with 15 people instead of 150? Energy-Based Models.

www.securesql.info · 2 months ago

SentinelMesh is a global autonomous incident response architecture that replaces autoregressive large language models with energy based models that score competing threat hypotheses as energy landscapes rather than predicting text. It deploys distributed agents across six continents and low earth orbit, uses cryptographic proofs and explainable confidence scores, and claims to produce court admissible forensic evidence in 47 seconds while enforcing a ten layer safety stack and a five minute undo window. The platform auto translates indicators across major SIEMs, auto tunes confidence thresholds as alert volumes change, and reports 99.9997 percent uptime and more than 99.95 percent accuracy on routine incidents. Detailed EBM training architectures and orbital node specifications are redacted under non disclosure agreements and available only to pre authorized partners.
Linux kernel maintainers suggest a ‘kill switch’ to protect systems until a zero-day vulnerability is patched

www.csoonline.com · 2 months ago

There is a proposal from a Linux kernel maintainer to add a kill switch that lets privileged operators disable specific vulnerable kernel functions until a patched kernel can be built, distributed and systems rebooted. The suggestion was prompted by recent high severity bugs including Copy Fail CVE-2026-31431 and the Dirty Frag exploit that combines CVE-2026-43284 and CVE-2026-43500. Security experts are divided, with critics warning the kill switch could be misused, encourage delaying patches and cause self inflicted outages because its impact is hard to assess, while supporters including a Red Hat official say it could provide immediate non disruptive mitigation in high availability environments.
Microsoft’s C.E.O. Intervened When OpenAI Fired Sam Altman, Musk’s Lawyer Claims

www.nytimes.com · 2 months ago

Elon Musk’s lawyer told a federal trial that Microsoft Chief Executive Satya Nadella helped secure Sam Altman’s return to the OpenAI chief executive role after Altman was briefly fired in November 2023. Trial exhibits including text messages and emails show executives such as Mira Murati discussing Nadella’s involvement and Microsoft’s communications with OpenAI during the chaotic week. Nadella testified to clarify Microsoft’s financial ties and internal debates over the company’s influence as the case examines whether OpenAI strayed from its original nonprofit mission.
OpenAI announces Daybreak, "frontier AI for defenders"

openai.com · 2 months ago

OpenAI Daybreak is a frontier AI initiative to change how software is built and defended by surfacing risk earlier and making software resilient by design, enabling defenders to reason across codebases, identify subtle vulnerabilities, validate fixes, analyze unfamiliar systems, and accelerate remediation. Daybreak pairs expanded defensive capabilities with trust, verification, proportional safeguards, and accountability, combining OpenAI models and Codex as an agentic harness to support secure code review, threat modeling, vulnerability triage, malware analysis, detection engineering, patch generation and validation, and authorized red teaming. Access tiers include GPT‑5.5 as a default, GPT‑5.5 with Trusted Access for Cyber for verified defensive workflows, and GPT‑5.5‑Cyber for specialized authorized work, and the program is being deployed with industry and government partners and is already used or endorsed by major security vendors including Cloudflare, Cisco, CrowdStrike, and Palo Alto Networks.
GM agrees to $12.75M California settlement over sale of drivers’ data

www.bleepingcomputer.com · 2 months ago

There is a proposed twelve point seven five million dollar settlement requiring General Motors to pay civil penalties and accept injunctive measures after officials found it sold hundreds of thousands of Californians’ precise location and driving behavior data to data brokers Verisk and LexisNexis between 2020 and 2024. The settlement is the largest California Consumer Privacy Act penalty to date and requires GM to stop selling driving data to consumer reporting agencies for five years, delete retained driving data within 180 days absent express consent, and ask the brokers to delete the data. Officials said the data was collected through OnStar and the Smart Driver program, that GM made roughly twenty million dollars nationwide from those sales, and that the company must implement stronger privacy compliance and submit regular assessments. The settlement is subject to court approval and follows a separate Federal Trade Commission order restricting GM’s disclosure of geolocation and driver behavior data.
Official CheckMarx Jenkins package compromised with infostealer

www.bleepingcomputer.com · 2 months ago

A rogue version of Checkmarx's Jenkins Application Security Testing plugin was published to the Jenkins Marketplace on May 9, 2026 and included credential-stealing malware. The TeamPCP hacker group claimed responsibility after using credentials obtained in the March Trivy supply-chain compromise to access Checkmarx's GitHub repositories and publish backdoored developer tools to GitHub, Docker, and VSCode distributions. Checkmarx advised running plugin version 2.0.13-829.vc72453fa1c16 published December 17, 2025 or an older release, to assume credentials may be compromised, rotate all secrets, investigate for lateral movement and persistence, and use published indicators of compromise to hunt for infections while noting the company says its GitHub repositories are isolated from customer production environments.