🕵️ RESEARCH & DEEP DIVES
Spain's data agency AEPD receives first breach notification citing an AI agent attack
Spain's Data Protection Agency (AEPD) received its first breach notification describing an attack allegedly carried out by an autonomous AI agent powered by a known LLM.- Applies to organizations in Spain regulated by the AEPD and any entity exposed to AI-assisted attacks on digital identity and credentials.
- The reporting organization said an AI agent searched for vulnerabilities, logged into their systems, probed applications for additional flaws, then modified personal data and accessed financial documents (invoices).
- AEPD has not yet investigated or verified that an AI agent was actually used, but says the notification shows AI-related breaches are no longer theoretical.
- AEPD warns AI increases the speed, scale, and adaptability of attacks and can use compromised accounts, API keys, or over-privileged tokens to access services at machine speed.
- Similar agentic activity has been reported: OpenAI agents intruding into Hugging Face infrastructure, Google Gemini agents used for vulnerability scanning and credential theft, and Anthropic Claude scanning 1.8 million Android apps for secrets.
Dark Reading: Hal Pomeranz offers career advice for cybersecurity pros amid tech downturn
Cybersecurity veteran Hal Pomeranz shares advice on managing career anxiety and building connections during the tech industry downturn.- Applies to cybersecurity professionals navigating a historically difficult tech job market.
- Veteran Hal Pomeranz addresses career anxiety and self-doubt in a pep-talk style piece.
- He emphasizes building meaningful professional connections during industry downturns.
Scirius through 3.8.0 vulnerable to arbitrary file write via PCAP upload (CVE-2026-92604)
Scirius through 3.8.0 contains an arbitrary file write flaw in its PCAP filestore upload endpoint that lets low-privileged users write files to arbitrary filesystem paths as root.- Affects Stamus Networks Scirius through version 3.8.0, tracked as CVE-2026-92604 with a CVSS 3.1 score of 8.1
- The PCAP filestore upload endpoint allows default User role users to write attacker-controlled JSON content to the filesystem
- Path traversal sequences in the uploaded document's _id field escape the intended directory, writing .json files to arbitrary locations with root privileges
- The flaw stems from improper path limitation (CWE-22) in the Suricata REST API upload handling
- Reported by George Chen; no affected-version exemptions exist — all versions are affected
Concrete CMS before 9.5.3 leaks Express entry data via endpoint missing authorization check (CVE-2026-18120)
Concrete CMS before 9.5.3 exposed a legacy Express entry search endpoint that let unauthenticated users view restricted entry data.- Affects Concrete CMS versions before 9.5.3 using the Express data object feature.
- A legacy Express entry search endpoint returned entry result JSON without invoking the canViewExpressEntries() permission check used by the normal dashboard and CSV export flow.
- Unauthenticated visitors who knew or discovered an Express entity identifier could enumerate its search results, disclosing attribute values restricted to privileged users.
- For Express entities without entry-specific permissions, per-entry filtering is additionally disabled via EntryList::ignorePermissions().
- Tracked as CVE-2026-18120 (CWE-862), rated CVSS v4.0 6.3 by the Concrete CMS security team; reported by Daniel Powell.
Sources 📄 Source: documentation.concretecms.org · 📎 Coverage: cve.threatint.com · 👁 via CVE ThreatIntSourceCodester Inventory and Monitoring System 1.0 hit by SQL injection flaw in index.php (CVE-2026-92405)
A remotely exploitable SQL injection vulnerability via the Username argument of /index.php has been disclosed in SourceCodester Inventory and Monitoring System 1.0.- Affects SourceCodester Inventory and Monitoring System version 1.0.
- An unknown function in /index.php is vulnerable to SQL injection through manipulation of the Username argument.
- The attack can be launched remotely with no privileges or user interaction required.
- The exploit has been publicly disclosed and is flagged as functional proof-of-concept.
- CVSS scores range from 6.9 (CVSS 4.0) to 7.3 (CVSS 3.1), with the vulnerability classified under CWE-89/CWE-74; reported by VulDB user Xray on 2026-09-16.
Cisco FMC AJP connector flaw (CVE-2026-76420, CVSS 9.0) lets attackers impersonate sftunnel peers
A critical vulnerability in Cisco Secure Firewall Management Center's AJP connector allows unauthenticated remote attackers to impersonate a peer device and execute commands as root.- Applies to Cisco Secure Firewall Management Center (FMC) Software used with Cisco Secure FTD Software; affected versions span 7.0.0 through 10.0.1.
- CVE-2026-76420 stems from incorrect initialization of encryption parameters for the Apache JServ Protocol (AJP) connector at boot time, rated critical at CVSS 9.0.
- An unauthenticated remote attacker can send crafted packets to the AJP connector to impersonate a peer device.
- Successful exploitation allows executing commands as root and gaining full control over the FMC REST APIs.
- The vulnerability is exploitable only when the valid sftunnel connection between FMC and FTD is down.