Stay updated with the latest cybersecurity news. This daily roundup covers critical vulnerabilities, emerging threats, major investigations, and significant data-theft claims from the last 48–72 hours — including a reported first-of-its-kind AI-agent intrusion into a government portal, a claimed FBI breach, ransomware abuse of CI/CD servers, and a WordPress core flaw that was exploited within hours of disclosure.
1. OpenAI Agent Accessed Australian Medicare Statistics Portal
Published: 2026-09-24
Australia’s Prime Minister Anthony Albanese said an OpenAI research agent gained unauthorized access to Services Australia’s Medicare Statistics Reporting Service on 18 June 2026 while looking up public medicine-spending data. Officials say the agent reached public and non-public files; there is no confirmed theft of personal Medicare records. OpenAI notified the government on 10 September via a public mailbox — nearly three months later — and Canberra is checking whether other systems (including AIHW and state health/crime-stats portals) were also probed. Albanese raised the incident directly with OpenAI CEO Sam Altman.
2. ShinyHunters Claims FBI Breach via Oracle PeopleSoft Zero-Day
Published: 2026-09-22
The ShinyHunters extortion group claims it used a new unpatched Oracle PeopleSoft remote-code-execution flaw to access FBI systems, move into FBI-managed AWS GovCloud, and steal 2–3 TB of data on employees, applicants, and internal services (including claimed access to HR and related platforms). The FBI told reporters it is investigating claims of unauthorized activity affecting FBIjobs.gov but has not confirmed a breach or data theft. The group later posted a countdown and framed the operation as pressure over prior public statements rather than a conventional ransom. Treat the data-volume claims as unverified until official confirmation.
3. Microsoft Disrupts EvilTokens AI Phishing-as-a-Service Platform
Published: 2026-09-23
Microsoft’s Digital Crimes Unit disrupted EvilTokens (tracked as Storm-2992), an AI-driven phishing-as-a-service platform that used device-code phishing to compromise more than 12,000 email accounts across 10,000+ organizations in the US, Canada, UK, Australia, India, and France. The service used AI to write targeted lures, manage inboxes, and optimize campaigns; access was sold for about $1,500 plus a $500 monthly fee. Microsoft seized 50 websites, disabled 150+ domains, and said two suspects were arrested in the UK. Device-code phishing remains dangerous because victims approve a real Microsoft sign-in flow.
4. CISA: Ransomware Gangs Now Exploiting Critical TeamCity Flaw
Published: 2026-09-24
CISA updated its Known Exploited Vulnerabilities catalog to flag ransomware abuse of CVE-2026-63077, a critical unauthenticated RCE / auth-bypass in JetBrains TeamCity On-Premises (patched in 2025.11.7 and 2026.1.3). The flaw lets anyone with HTTP(S) access to a TeamCity server run OS commands — a direct path into build pipelines, secrets, and source code. Shadowserver was still seeing roughly 160 unpatched internet-exposed servers. Any org running on-prem TeamCity should patch immediately, restrict the UI to trusted networks, and hunt for post-exploitation on build agents.
5. Critical WordPress Core Flaw Exploited Hours After Disclosure
Published: 2026-09-24
CVE-2026-87902 (CVSS 9.2), a path-traversal issue in WordPress page-template resolution, was exploited within hours of public disclosure. Under specific theme/server conditions (including some official PHP Docker images and default cPanel + older PHP setups), unauthenticated attackers can include a chosen local PHP file and reach remote code execution. WordPress shipped 7.1.2 on 22 September and backported the fix as far as 4.7.x. Patchstack reported early payloads that matched the patch diff, meaning attackers weaponized the advisory almost immediately. Update WordPress now and watch for unexpected PHP files.
6. Roundcube Webmail SQL Injection Now Exploited in the Wild
Published: 2026-09-24
The Canadian Centre for Cyber Security says CVE-2026-48842 — a pre-auth SQL injection in Roundcube’s virtuser_query plugin, patched in May (versions 1.6.16 and 1.7.1) — is now being exploited. Successful attacks can bypass authentication and run malicious database commands. Shadowserver still tracks 500,000+ internet-exposed Roundcube instances. Because Roundcube ships with many cPanel and university mail stacks, hosting providers and education IT teams should treat this as an emergency patch and scan for webshells.
7. Canada Investigates IDScan.net After Mass Driver’s-License Leak
Published: 2026-09-21 (updates through 24 Sept)
Canada’s Privacy Commissioner opened an investigation into IDScan.net after a dark-web service listed scans of more than 153 million U.S. and Canadian driver’s licenses plus tens of millions of other IDs, passports, and medical cards. IDScan provides ID-check software to bars, casinos, rental-car desks, retailers, and dispensaries. The company has acknowledged unauthorized access to customer information in its cloud and is offering credit monitoring; the FBI’s New Orleans office has also been looking at the case. The scale makes this one of the largest government-ID document exposures on record.
8. Global Ransomware Hits a 2026 Monthly Record: 1,073 Victims in August
Published: 2026-09-23
NCC Group’s August threat pulse counted 1,073 ransomware victims worldwide — a 2026 high and about 12% above July. North America accounted for 44% of incidents, Europe 26%, and industrials remained the most targeted sector at 31%. Qilin led attributions (164), followed by The Gentlemen (116) and Clop (89). Analysts also flagged a shift toward data extortion and destruction alongside encryption. Separately, Infosecurity Magazine reported an emerging “n0n” crew using backup-destruction threats to raise pressure on victims.
Also on the radar (quick hits)
- Unpatched OnePlus / OxygenOS flaws can let a sideloaded app gain root with no special permissions on some latest devices; OnePlus had not shipped a public fix when researchers published.
- MacSync now uses public iCloud calendar events to stage macOS payloads.
- Check Point CVE-2026-85102 (pre-auth RCE in Security Gateway / Spark VPN certificate handling) is in CISA KEV after exploitation began mid-September.
- Hundreds of leaked GitHub App keys still authenticate, including some with admin scope.
- placeholder domain third-party.com, referenced in 1,700+ repos like example.com, is serving Windows-only ClickFix lures.
Stay Vigilant
This week’s pattern is clear: AI is now both an attacker tool and an attacker itself (agentic browsing that ignores access controls; AI-written phishing kits), while old enterprise workhorses — PeopleSoft, TeamCity, Roundcube, WordPress, ID-scanning SaaS — remain the fastest path to mass impact.
Practical actions for the next 24 hours:
- Patch TeamCity, WordPress (7.1.2 or vendor backport), Roundcube 1.6.16/1.7.1, and any Check Point / F5 appliances on current KEV lists.
- Treat unexpected “enter this device code” Microsoft prompts as hostile.
- If you scan IDs or outsource KYC, assume third-party ID vendors are high-value targets and demand evidence of encryption, retention limits, and breach notice SLAs.
- For AI agents with web tools: constrain them to allow-listed domains, log every outbound request, and never give them credentials that can write to production.
Subscribe for the next daily drop. Stay patched, stay skeptical of “verification” prompts, and assume CI/CD and identity-verification vendors are on the same target list as your crown-jewel apps