That figure—part of Hudson Rock’s latest findings on what it calls FortiBleed 2.0—is one of several blunt reminders this week that defenders are still losing the race wherever credentials, forgotten configuration files, and trusted software overlap. The collection of briefings and incident notes published through late October 2026 maps a range of behaviors: large-scale credential leakage, old vulnerabilities still under active exploitation, token theft that requires no special Android permissions, and attackers borrowing trust from legitimate apps, DNS namespaces, and AI assistants.
FortiBleed 2.0: 40,573 secrets, 975 new domains, 1,703 source files
Hudson Rock says its investigation into the FortiBleed credential-harvesting campaign found 975 newly identified domains and 40,573 exposed secrets drawn from 1,703 FortiGate source files tied to 1,689 host identifiers. The breakdown includes 37,076 local-user records and 1,576 Active Directory LDAP bind records. Hudson Rock noted the configuration material dates to around February 2026, earlier than the May/June material examined in the project's initial reporting. The discovery has been codenamed FortiBleed 2.0.
Fortinet issues, coordinated fixes, and ongoing exploitation of old bugs
Fortinet has released “a coordinated set of security updates for FortiOS, FortiManager, and FortiAnalyzer” to address critical and high-severity flaws discovered with an AI-assisted vulnerability approach, the company said; details are being withheld until November 10, 2026. The bulletin arrived as other exploitation activity continued: VulnCheck reported sustained exploitation of CVE-2026-42589, a critical remote-code-execution bug in Gotenberg (CVSS 9.8) patched in May 2026; over the past 30 days VulnCheck counted 1,358 exploitation events from 128 unique source IP addresses, delivering a Go-based reverse shell called “cache_base.” The plain lesson in these items: vendors are patching, but attackers still find and weaponize gaps long after fixes exist.
This site is the portfolio.
OSINTSights runs on Cloudflare Workers, D1, R2, and Vectorize, with an AI pipeline on Hetzner ARM. Nubivance designed, built, and operates it. We do the same for clients.
See what we buildZero-permission token theft and AI-impersonation scanners
Microsoft and independent researchers disclosed a now-patched flaw in Microsoft Authenticator (CVE-2026-80097, AuthMask) that could let an installed malicious Android app—without requesting any dangerous permissions—impersonate a Microsoft first-party app and receive that app’s authentication tokens, Enclave said. Microsoft warned that “successful exploitation could allow the attacker to obtain authentication tokens and access resources as the affected user.”
Concurrently, Cequence tracked a single actor mounting AI-agent impersonation at scale: about 32 dummy AI identities—posing as ChatGPT, Claude, Perplexity and others—searching for files that commonly store secrets. Will Glazier at Cequence described a scanner that rotates identities, targets roughly 90 file paths, and sends short bursts of 50–190 requests in about eight seconds per target, hitting organizations in financial services, retail, telecommunications, energy, and software from at least late September through late October. Taken together, AuthMask and the AI-impersonation campaign underline how both platform design and opportunistic scanning can turn legitimate token-based and identity flows into stealthy theft vectors.
Android permissions, MR ROBOT banking malware-as-a-service, and MDM abuse
Google announced that an upcoming Android release will remove the ability to directly write to contacts using WRITE_CONTACTS; apps will be steered toward standard contact intents (ACTION_INSERT, ACTION_EDIT, ACTION_INSERT_OR_EDIT). Google’s rationale: WRITE_CONTACTS grants broad background access that most apps do not need.
At the same time, CloudSEK described a new Android malware-as-a-service called MR ROBOT that rents customized banking trojans to affiliates, intercepting OTPs, card details, UPI PINs and credentials and even forwarding calls to bypass checks. The platform impersonates roughly 36 lure apps and uses a Telegram bot and web control panel to build and distribute tailored APKs; one affiliate reportedly intercepted 5,407 text messages from 153 infected devices. Huntress also warned that threat actors are abusing legitimate mobile-device-management platforms like Hexnode and Rippling by social-engineering enrollments to establish a privileged management path and deploy remote-management tools.
What this means for technologists, procurement leaders, and end users
- Technologists and security teams: inventory and rotate secrets urgently—FortiBleed 2.0 and the Gitee scan (26,400 live credentials found in public Gitee repositories) are concrete signals that exposed configuration files and repository leaks remain high-probability failure points. Apply Fortinet’s coordinated updates, patch vulnerable services such as Gotenberg installations, and audit token brokers on mobile endpoints for anomalous consumers.
- Procurement and platform owners: insist on supply-chain hygiene and visibility into AI/third‑party model usage. Cisco Talos’ disclosure of NIK C2—an AI-integrated C2 framework that abused the OpenAI Chat Completions API—and other reports show attackers combining stolen AI credentials and commercial models to automate tooling; contractual controls and telemetry for third-party model access matter.
- End users and admins: treat unusual authentication prompts and unexpected contact‑writing requests as red flags. Microsoft’s AuthMask advisory and Trellix’s description of a malicious PDF that abused /OpenAction and /Launch to run a Base64-encoded PowerShell stager (delivering a SIRIUS backdoor) are reminders that social engineering remains the trigger for many complex, multi-stage intrusions.
The week’s pattern is plain: attackers mix automation and old-school opportunism—scanning repositories, abusing DNS administration, repurposing trusted binaries, and leaning on AI to scale their labor. The pragmatic next step is not a grand strategy but the familiar triage: patch known bugs, hunt for leaked secrets, and harden the small, trusted pathways—tokens, DNS, MDM enrollments—that attackers habitually exploit.
