Hackers hijacked 4,837 GitHub accounts

Hi there, this is your daily ☕️ Cyberpresso.

Stolen GitHub logins hijack 4,837 accounts LINK

• A threat actor used stolen GitHub credentials to push secret-stealing malware into tens of thousands of code repositories, hijacking 4,837 accounts to open probe pull requests across 61,764 repositories starting on Tuesday, researchers at Aikido found.
• At least 1,806 repositories ran the "wfx" payload, a shell script that grabs GitHub Actions secrets, cloud keys, and tokens for AI tools like OpenAI and Anthropic, then swaps OIDC tokens for AWS, Google Cloud and Azure sessions.
• Teams should search for the listed branches and pull request titles, rotate every exposed secret and cloud role, revoke hijacked tokens, and treat the Packagist package neuronai-studio 3.14.3 and affected machines as compromised.

Worm hijacks ScreenConnect at 48 firms LINK

• A worm has broken into ScreenConnect at 48 organizations using Huntress, letting an attacker push commands through cloud-hosted panels to install rogue clients and a toolkit called SCKit, hitting hundreds of hosts in 90 minutes.
• The attack traces to the RMM+ extension, whose latest versions accepted a blank access key, letting callers skip ScreenConnect's accounts, roles, and MFA; ConnectWise disabled it and four other extensions across hundreds of partner environments.
• The attack was seen in the wild on Wednesday, pushing SCKit through the Run command feature as SYSTEM; the worm's designed spread from endpoints to technician workstations and other panels failed in every host log Huntress checked.

Palo Alto flaw gives attackers root access LINK

• Palo Alto Networks has revealed a flaw in PAN-OS on its Panorama management tool that lets an authenticated administrator with network access run arbitrary code with root privileges, tracked as CVE-2026-0330.
• The company says it has seen no attacks in the wild and an internal researcher found the bug; it only hits Panorama, not Cloud NGFW, Prisma Access, or standalone PAN-OS firewalls.
• Rated medium severity, the flaw has no real workaround, so admins should install the patched releases across the five supported PAN-OS branches and keep management access off the open internet.

Critical Apache flaw allows code execution LINK

• A critical bug in Apache HTTP Server's HTTP/2 module, mod_http2, lets a remote attacker with no login run code and take over the server process, Red Hat warns, rating it 9.8 out of 10.
• Tracked as CVE-2026-119618, the flaw hits versions 2.4.55 through 2.4.70 and is fixed in 2.4.71, released yesterday; only servers with HTTP/2 turned on through the Protocols directive are at risk.
• Red Hat says public research shows a reliable exploit against a default setup, but Wiz reported no available exploit and no attacks in the wild; admins who can't patch should drop h2 and h2c, leaving http/1.1.

Hackers abuse AI tool to steal AWS keys LINK

• Hackers are abusing AntiFox, an AI-driven automatic penetration testing tool hosted on GitHub, to steal Amazon Web Services credentials and move deeper into victims' cloud environments, according to a report from Arctic Wolf Labs shared with The Hacker News.
• In some attacks, the intruders broke in through a compromised internet-facing IIS application, planted an ASPX web shell via the "w3wp.exe" worker process, then read config files and database connection strings; the exact vulnerability behind the code execution is unknown.
• Using valid AWS identities, the attackers enumerated dozens of services including IAM, S3, and Secrets Manager; while many requests returned "AccessDenied," some succeeded, including a GetSecretValue request that pulled a production secret.

Strategies & Tactics

> Blind trust: stealing credentials through CSP: Automates finding Content Security Policy misconfigurations in Burp so attackers can steal credentials even when the policy blocks all JavaScript, via self-submitting forms.

> Microsoft Finds Hackers Using GSocket to Hide Linux Backdoors as Legitimate System Processes: Attackers hide a remote-access tool as fake system processes on Linux servers, keeping backdoor access even after admins patch the original flaw.

> Stolen Claude API keys can turn Claude Code into a remote-controlled implant: cking tool, so scope and protect those keys like production credentials.

> Detecting credential-driven attacks on Databricks: Correlate Databricks login and audit records to spot stolen credentials by the unusual activity that follows a valid sign-in, since attackers lack an employee's normal behavior.

> New Windows SignDrop Zero-Day Exploit Allows Attackers to Gain SYSTEM Privileges: A publicly released zero-day exploit turns Microsoft's own anti-tampering fix into a way for low-privilege users to seize full SYSTEM control.

> Citrus Ransomware: VLESS for traffic, Sn00py for encryption, and LLMs lending a hand: Citrus mixes stolen contractor logins, disguised web traffic, and AI-generated attack scripts to breach networks and deploy ransomware while slipping past defenders.

Other news you might like

• Node.js Fixes 16 Flaws, Including HTTP/2 Bugs That Crash Servers and Leak Other Connections' Data LINK

• Who Runs the Ransomware Group Qilin? LINK

• Shipping giant Maersk probes 'IT security incident,' data theft LINK

• Critical Pre-Auth MongoDB Flaw (CVE-2026-118383) Can Crash Lagging Servers on Failover LINK

• PoC Exploit Released for Critical Cisco Expressway RCE Vulnerability With CVSS 9.8 Score LINK

• Critical Red Hat Quay Vulnerability Could Expose Private Container Images LINK

Trending tools

VibeDefend: embeds security and business rules into AI coding agents like Cursor and Copilot, catching vulnerabilities and malicious packages before code reaches your PR. LINK

Ghostifier: scans inbox headers to find companies holding your data, then auto-unsubscribes, opts out of sales, and sends tracked deletion requests privately. LINK

Off the Record: emits inaudible interference that blocks AI notetakers from transcribing your meetings, including hidden ones running silently on other people's devices. LINK

AuthMonster: manages two-factor authentication with biometric locking, easy migration from Google Authenticator and LastPass, and a donation-based model with no paywalls. LINK

Termaxa: open-source Rust safety gate for AI coding agents that previews deletions, backs up files, and enforces policy with a tamper-evident audit log. LINK

Strac Comply: automates SOC 2, ISO 27001, HIPAA, and PCI compliance by connecting your cloud stack and running 100+ continuous tests. LINK

Trending papers & reports

Backdoor detection in shared-model training spots malicious participants who secretly poison a collaboratively trained AI, using just 10 clean reference images and matching the accuracy that rival methods need far more data to reach. LINK

BypassGym is a testing ground measuring whether AI agents can write malware that slips past antivirus defenses, finding even cheap models succeed often, with one evading detection in 17 of 18 attempts. LINK

RAG access control that reshapes document embeddings keeps unauthorized users from retrieving restricted files but still leaks the originals from a few hundred examples, meaning isolation alone does not guarantee confidentiality in enterprise search systems. LINK

Keyboard eavesdropping can reconstruct what someone types just from the sound of their keystrokes recorded by an ordinary microphone, and now works reliably across unfamiliar keyboards, noisy rooms, and compressed audio, not just in the lab. LINK

See you tomorrow for a new dose of ☕️ Cyberpresso!