News archive · 40 stories
August 2026
- BreachMcKesson breach started with vishing against Okta SSO, then reached Salesforce and SnowflakeShinyHunters social-engineered McKesson employees from a lookalike domain, took their Okta credentials, and used the SSO session to reach Salesforce and Snowflake. The group claims roughly 284 million records and demanded 55 million dollars.
- Two PaperCut flaws chain into unauthenticated code execution, and the patch has bypassesCVE-2026-81578 bypasses authentication on the PaperCut web management interface, and CVE-2026-82078 turns a config edit into remote code execution. Exploitation started August 27, and new bypasses affect fully patched instances.
- StandardsA loose PHP comparison let attackers sign in as WordPress admin through SAMLTwo unauthenticated bypasses in the miniOrange SAML 2.0 Single Sign On plugin, CVE-2026-61979 and CVE-2026-15981, treat OpenSSL's error return as a valid signature. Both are under active exploitation.
- Mirage2FA reached 4,500 organizations by stealing sessions instead of breaking MFAANY.RUN traced the Mirage2FA phishing-as-a-service kit across 4,532 organization domains from 2024 to 2026. It abuses legitimate Microsoft 365 login flows to lift passwords and session cookies, then rides the authenticated session into every SSO-connected app.
- PasskeysWhatsApp lets one account hold several passkeys, closing the cross-platform gapMeta announced on August 25, 2026 that WhatsApp accounts can register multiple passkeys, so a user carrying both iOS and Android no longer has to bind to one ecosystem. Two-step verification also moves from a six-digit PIN to a full password.
- VulnerabilityKeycloak password reset flaw let anyone skip the email token and take over any accountCVE-2026-18963 is improper state validation in Keycloak's reset-credentials flow. A crafted request jumped the authentication session straight to the password-update step, no verification token needed. CVSS 9.1, fixed in 26.7.2.
- AI agentsOkta ships Agent SSO, making AI agents first-class identities instead of static API keysOkta made Agent SSO generally available on August 24, 2026, registering AI agents in Universal Directory with short-lived governed tokens and pushing Cross App Access into the Model Context Protocol as its enterprise authorization extension.
- VulnerabilityMicrosoft patched a CVSS 10.0 Entra ID flaw, then corrected the exploitation flag from yes to noCVE-2026-69836 was an unauthenticated deserialization flaw in Entra ID scoring a perfect 10.0. Microsoft fixed it service-side with no customer action, but first published it marked as exploited, then reversed that a day later.
- Machine identityPoisoned Rust crates turned routine builds into credential theft for about 90 minutes eachMalicious versions of arrayref, internment, and append-only-vec shipped an infostealer targeting browser profiles and crypto wallets. crates.io pulled them within two hours, but lockfiles and registry caches keep them around.
- Three Russian clusters move from password phishing to OAuth token theftGoogle Threat Intelligence Group tracked UNC6293, UNC7005, and UNC5976 abusing device-code flow, verification-code relay, and fake Continue with Google pages to take authenticated sessions rather than passwords.
- VulnerabilityNetScaler ships a critical authentication bypass affecting Gateway and AAA virtual serversCVE-2026-19490 (CVSS 9.3) bypasses authentication on NetScaler ADC and Gateway appliances running a Gateway or AAA virtual server, with a SAML action configured on current builds. Fixed in 14.1-73.32 and 13.1-63.21.
- Operation CameraSwarm compromised 14,500 Dahua devices, mostly by guessing passwordsHunt.io found an exposed operator directory documenting 14,530 compromised Dahua cameras over five weeks. Password attacks from 12,324 IPs did most of the work; two 2021 auth bypasses accounted for 1,923.
- PhishingRecruitTrap ran 3,000 fake recruitment logins with browser-in-the-browser popups and live MFA relayCTM360 found over 3,000 phishing URLs impersonating recruiters at 50-plus organizations across 14 sectors. The pages fake a browser window, complete with address bar and padlock, and relay MFA codes in real time.
- Breach1.6 million RingCentral records leaked, and the entry point was one phone callShinyHunters voice-phished a RingCentral employee out of their password in July, took 623GB, and dumped 280GB after the company refused to pay. Have I Been Pwned confirmed 1.6 million accounts including names, emails, phone numbers and addresses.
- VulnerabilityFortinet's January SSO bypass hit boxes already patched for December's SAML bugCVE-2026-24858 is the follow-on FortiCloud SSO SAML bypass. Devices patched for CVE-2025-59718 and 59719 were still exploitable. Actively exploited. CISA guidance 28 January 2026.
- VulnerabilityKerberLoss: invisible Unicode lets an attacker twin a Kerberos SPNCVE-2026-25177, CVSS 8.8. Active Directory treated look-alike SPNs as unique. Semperis and Shai Laron showed how that becomes service hijack and NTLM downgrade. Patched March 2026.
- VulnerabilityResetNightmare: a low-priv UPN write can reset a Domain Admin via Kerberos kpasswdCVE-2026-27912 lets a user who can write their own UPN aim a Kerberos change-password at a Domain Admin. Microsoft rated it Important. Identity teams should not.
- Token theftSharePoint JWT bypass went from proof of concept to exploitation in under 48 hoursRapid7 published a working PoC for CVE-2026-55040 in mid-August. Exploitation telemetry spiked from one attempt to eight the following day, from eight IPs across five countries. Microsoft patched it in July.
- AI agentsOpening a GitHub issue was enough to reach CI secrets in Claude Code and Gemini CLINovee Security showed at Black Hat that an unprivileged GitHub user could open an issue that reached workflow credentials on the coding-agent repositories of Anthropic, Google, and OpenAI. Fixes shipped in Gemini CLI 0.39.1 and Claude Code 2.1.163.
- Machine identityOver 1,000 AI-named npm typosquats deliver a cross-platform dropperOpenSourceMalware and Sonatype tracked Flooding Dropper, a campaign of AI-generated typosquat package names carrying WEL1DROPPER, which fetches Sliver on Linux and encrypted payloads on Windows and macOS.
- N-able confirms attackers used an N-central auth bypass to reach managed customer networksCVE-2026-18577 gave unauthenticated attackers full administrative control of N-central, which they used through the Take Control feature to reach downstream managed endpoints and plant Cloudflare Tunnels for persistence. A second mandatory hotfix landed August 7.
- PasskeysMalware can drive a Windows Hello key for Entra ID persistence without a PIN promptDirk-jan Mollema showed that code running in a signed-in Windows session can use the victim's TPM-bound Windows Hello for Business key as a FIDO2 credential, satisfying phishing-resistant Conditional Access and acquiring a Primary Refresh Token. No admin rights, no CVE.
- Kali365 phishes Microsoft's own device login page for durable Microsoft 365 tokensANY.RUN documented Kali365, a kit that lures victims through fake SharePoint, OneDrive, and DocuSign pages to Microsoft's genuine device login portal, where approving an attacker-supplied code hands over access and refresh tokens.
- Machine identityGitGuardian found 4,576 leaked n8n tokens, and a third of reachable instances accepted themExposed n8n API tokens in public GitHub commits gave researchers read access to workflows, stored credentials, and data tables, plus a path to exfiltrate the raw secrets of every connected service.
- PasskeysPass-the-Passkey: a Black Hat researcher found the WebAuthn implementation bugs, not the standardAt Black Hat USA 2026, SpecterOps principal security researcher Michael Grafnetter presented a family of passkey attacks including cleartext YubiKey signatures readable by any authenticated user and a major cloud passkey implementation vulnerable to the exact attack it was built to stop.
- AI agentsA CVSS 10.0 bug let one user's Terraform token serve another user's requestHashiCorp's Terraform MCP Server failed to assign unique session identifiers in stateless HTTP mode, so a token supplied by one user could be reused for later requests from others. Veeam and Django patched serious flaws the same week.
- ProductCrowdStrike's Falcon Fund backs Above Security, folding AI-native insider risk into FalconAbove Security, a Tel Aviv-based insider-threat platform, announced a strategic investment from the CrowdStrike Falcon Fund at Black Hat USA 2026, alongside an integration that gives Falcon customers ready-made insider risk investigations built on Falcon Next-Gen SIEM telemetry.
- AI agentsZero Networks ties AI agent identity to the network layer, with just-in-time MFA for the sensitive protocolsZero Networks launched Least Agency Enforcement at Black Hat USA 2026, implementing OWASP's emerging Least Agency principle with identity-based microsegmentation and default-deny network access for AI agents, plus MFA prompts on RDP, SMB, and WinRM even when an agent presents valid credentials.
- N-able's first fix for an N-central auth bypass missed a second exploitation pathAttackers used an authentication bypass in N-able's N-central RMM platform to gain administrative access and register persistent Cloudflare tunnels on managed endpoints. N-able's initial patch blocked only one exploitation route; a second CVE covers the one it missed.
- PasskeysUnit 42 finds malware can extract Google's synced passkey keys straight out of Chrome's memoryPalo Alto Networks Unit 42 disclosed three attacks, Pass-ta-key, Silver Pass-ta-key, and Golden Pass-ta-key, that extract device identity keys and the master secret behind Google's synced passkeys on Windows, undermining the claim that passkey private keys cannot be copied.