View Ridge Security
Back to Cyber Hose

XBOW Built $250K Chrome Full-Chain Exploit With GPT-5.6 AI Agents

AI agents built a full-chain Chrome exploit in six days, iOS exploit kits compromise iPhones via web visit, and Chrome/Firefox patch 114 flaws.

🚨 ACTIVE EXPLOITATION

  • (no items)

💥 BREACHES & INCIDENTS

  • Attackers Breach European Bank via Third-Party Gateway, Target WSUS Server
    Attackers breached a European bank through a weakly authenticated third-party gateway and attempted to exploit a WSUS vulnerability before being contained.
    • A European bank was breached via a third-party-managed external gateway appliance that lacked strict authentication and MFA, using compromised credentials to access internal systems over VPN.
    • The attacker scanned thousands of systems across multiple protocols, ran months-long password-guessing campaigns against administrator accounts, and scraped internal documentation and network shares.
    • The attacker attempted to exploit an undisclosed remote code execution vulnerability in WSUS on the bank's central configuration management server; no CVE identifier was disclosed.
    • Cortex XDR detected the code execution attempt and contained the threat before persistence or lateral movement; forensics found no backdoors, and no malware family, IOCs, or stolen data volume were published.
    Sources 📎 Coverage: cybersecuritynews.com · 👁 via Cyber Security News, cryptika.com (discovered)

🔓 CVEs & KEV

  • CVE-2026-104795 — Red Hat OpenShift Container Platform 4 — CVSS 8.8 — Insights-runtime-extractor: insights-runtime-extractor: container escape via ...

  • CVE-2026-101341 — NLnet Labs NSD — CVSS 8.7 — Stack overwrite from specially formatted TXT-DATA RDATA fieldTokens of text i...

  • CVE-2026-101339 — NLnet Labs NSD — CVSS 8.7 — Stack overwrite from a specially crafted and formatted HIP RRA HIP resource r...

  • CVE-2026-101338 — NLnet Labs NSD — CVSS 8.7 — Stack overwrite from a specially crafted and formatted SVCB or HTTPS RRWhen a...

  • CVE-2026-101337 — NLnet Labs NSD — CVSS 8.7 — Stack overwrite from specially formatted base64 RDATA fieldWhen base64 data i...

  • CVE-2026-85851 — Microsoft Windows 11 version 23H2 — CVSS 7.8 — Windows SmartScreen Vulnerability Enables Privilege Escalation Attacks

  • CVE-2026-100229 — Red Hat Enterprise Linux 10 — CVSS 6.5 — Gstreamer1-plugins-bad-free: gstreamer1-plugins-bad-free: out-of-bounds write...

  • CVE-2026-100231 — Red Hat Enterprise Linux 10 — CVSS 6.1 — Gstreamer1-plugins-bad-free: gstreamer1-plugins-bad-free: heap out-of-bounds ...

🕵️ RESEARCH & DEEP DIVES

  • XBOW Built a $250,000 Chrome Full-Chain Exploit With GPT-5.6 AI Agents in Six Days
    XBOW used AI agents running OpenAI's public GPT-5.6 Sol model to build a full-chain Chrome exploit in six days, earning Google's top $250,000 bounty.

    • Applies to Google Chrome on Windows — an estimated one billion users were at risk before fixes; the exploit required only that the user load a web page.
    • The chain linked three bugs: a V8 Maglev compiler flaw (CVE-2026-85045), an animated WebP decoder out-of-bounds write, and a Crashpad flaw (CVE-2026-78952), achieving code execution outside Chrome's sandboxes.
    • AI agents searched Chrome source, built proofs of concept, and drafted exploit code; researcher Brendan Dolan-Gavitt directed them, using swappable modular parts to replace a bug Google already knew about.
    • The awarded chain was built August 2–7, 2026, for about $40,000 in compute; bugs were fixed in Chrome 152.0.7977.65 (Aug 25) and 152.0.7977.82 (Sep 3), and were never used against users.
    Sources 📄 Source: xbow.com · 📎 Coverage: thehackernews.com · 👁 via The Hacker News, Cyber Security News (+1)

  • Hirudo Android trojan replaces banking apps with patched copies to bypass in-app security checks
    ThreatFabric has disclosed a new Android banking trojan, Hirudo, that replaces legitimate banking apps with patched versions that bypass the apps' own security checks.

    • Targets Android users, with patched replacements for eight Indonesian banking apps and credential targeting spanning Indonesia, China, India, Mexico, Morocco and Thailand
    • Steals credentials, intercepts SMS, captures screen and camera, and enables full remote device takeover via a mature backend
    • Delivered by dropper apps (Shopora, 1V1 X, KoiShort, ZoyaTV) posing as shopping or video apps, served as fake in-app updates; fake Google Play pages promoted via Meta ads
    • Uses a non-functional VPN connection to block internet access and hinder cloud-based checks like Google Play Protect
    • Replacement apps blind signature, accessibility, screen-capture and Play Integrity checks, re-signing key-attestation certificates with a publicly leaked keybox and relaying Integrity verdicts over plain HTTP
    Sources 📎 Coverage: thehackernews.com · 👁 via The Hacker News

  • Coruna and DarkSword iOS exploit kits can fully compromise iPhones via a single web visit
    The Coruna and DarkSword iOS exploit kits can fully compromise iPhones through a single visit to a malicious or compromised website.

    • Targets iPhone users: Coruna packs five exploit chains and 23 exploits for iOS 13.0–17.2.1; DarkSword uses a six-flaw chain against iOS 18.4–18.7.
    • Stolen data includes passwords, messages, browsing history, location, keychain contents and cryptocurrency wallets; Coruna's PlasmaLoader payload searches Apple Notes for wallet recovery phrases.
    • Delivery works via Safari: a malicious page fingerprints the device, exploits a WebKit or JavaScriptCore flaw for code execution, then chains sandbox escapes and privilege escalation to kernel-level access.
    • DarkSword was used by commercial surveillance vendors and suspected state-sponsored actors against targets in Saudi Arabia, Turkey, Malaysia and Ukraine; its source code leaked in 2026, spreading adoption, and iVerify reported a P7 DarkSword variant with keychain theft and C2 communication.
    • IOCs include exploit URLs hxxps://cdn[.]gotoiphone[.]com and hxxps://mzpo30[.]cam/assets/js/group[.]html, C2 hxxps://agdx53[.]cc, and SHA-256 hashes for rce_loader.js and c2_agent.js samples.
    Sources 📎 Coverage: cybersecuritynews.com · 👁 via Cyber Security News, cryptika.com (discovered)

  • AI pentesting agent finds critical OTP delivery-flaw authentication bypass on insurer's site
    Reflectiz's AI pentesting agent autonomously discovered a critical OTP-based authentication bypass on a major insurer's production website.

    • Reflectiz's AI pentesting agent, during an authorized assessment, found a critical-severity authentication bypass on a major insurer's production website with millions of users.
    • The login request let the caller choose the OTP channel and delivery address; the server sent the code to any email address in the request without verifying it belonged to the account.
    • An attacker with the required customer identifiers could access that customer's account by redirecting the one-time passcode to an inbox they control.
    • The agent found the flaw via JavaScript and API reconnaissance, then validated it through the live login flow before its guardrails halted the test after login succeeded.
    • The insurer had been pentested about six months earlier, yet the exploitable login flaw remained undetected.
    Sources 📎 Coverage: thehackernews.com · 👁 via The Hacker News

📋 ADVISORIES

  • Chrome, Firefox Updates Patch 114 Flaws as Mozilla Unveils Firefox Enterprise
    Google and Mozilla released Chrome and Firefox updates fixing 114 vulnerabilities, while Mozilla launched Firefox Enterprise.

    • Chrome 155.0.8059.79/.80 (Windows/macOS) and 155.0.8059.79 (Linux) fix 37 flaws, including seven critical memory issues: a buffer overflow in ANGLE and six use-after-free bugs in Browser, PDF, Network, Core, SVG, and Media
    • 29 high-severity Chrome bugs are mostly use-after-free issues; at least 15 were reported by Xinyang Ge (Anthropic) with Claude assistance, with no bounty paid
    • Firefox 158 patches 77 defects, including 30 high-severity use-after-free, sandbox escape, privilege escalation, information disclosure, race condition, JIT miscompilation, and integer overflow bugs; Thunderbird 158/153.5 and Firefox ESR 153.5/115.43 also updated
    • Mozilla launched Firefox Enterprise, a browser with a self-hosted management console for policies, access, data transfers, and AI use, built on the Gecko engine
    Sources 📄 Source: firefox.com · 📎 Coverage: securityweek.com · 👁 via SecurityWeek

  • Microsoft: Recent Windows updates may trigger Remote Desktop failures with error 0xC000000D
    Microsoft confirmed that recent Windows updates can cause Remote Desktop connections to fail with error 0xC000000D on systems with Administrator protection enabled.

    • Affects enterprise-managed Windows 11 24H2/25H2/26H1/26H2 devices used as Remote Desktop hosts, primarily in organizations using Windows security baselines.
    • Connections fail with error 0xC000000D ('The token that was passed to the function is invalid') after installing the KB5124010 September 2026 non-security preview update or later.
    • The issue occurs when the Administrator protection feature is enabled, which is default in Windows 11 24H2/25H2/26H2 security baselines from the Microsoft Security Compliance Toolkit.
    • Administrator protection began rolling out with KB5067036 preview cumulative update in October 2025 and gives logged-in users only standard user permissions with Windows Hello prompts for elevated actions.
    • Separately, Microsoft released out-of-band updates (KB5129195, KB5129194, KB5129236, KB5129235, KB5129237, KB5129238) on September 14, 2026 to fix RDS failures caused by September security updates on Windows 10, Windows 11, and Windows Server.
      📄 <https://support.microsoft.com/en-us/topic/windows-11-version-25h2-update-history-99c7f493-df2a-4832-bd2d

Need help assessing your exposure?

Start with the free Posture Self-Check to see where you stand against the current threat landscape.

Free Posture Self-Check