Daily Cybersecurity News – September 12, 2026
GitLab Critical Path Traversal Flaw Under Active Probes
CriticalWhat happened
GitLab issued emergency patches for two critical flaws in its CE and EE platforms. The lead issue, CVE-2026-85706 (CVSS 10.0), is an unauthenticated path traversal in the repository commits interface caused by improper path confinement and missing authentication checks. It lets an attacker read arbitrary files from the server in a single HTTP request.
Affected releases include all from 18.7 before 19.1.8, 19.2 before 19.2.6, and 19.3 before 19.3.2. WatchTowr Labs observed internet-wide probes almost immediately. CISA added the CVE to its KEV catalog on 2026-09-11, confirming active exploitation. A second flaw, CVE-2026-87719 (CVSS 9.9), affects only Enterprise Edition with Duo Chat access and can expose Advanced Search credentials.
Who is affected
Operators of self-managed GitLab Community Edition and Enterprise Edition in the vulnerable version ranges. GitLab.com hosted service already runs fixed code; Dedicated single-tenant customers are not impacted.
Internet-reachable self-hosted instances face the highest immediate risk from untargeted scanning.
Why it matters
Unauthenticated arbitrary file read on a source-control and CI platform can expose secrets, tokens, source code, and configuration that enable full compromise of development pipelines. These systems frequently hold high-value intellectual property and production credentials.
Active probes and KEV listing mean mass exploitation is likely imminent for any unpatched, exposed server.
How it could have been prevented
Upgrade self-managed instances immediately to 19.1.8, 19.2.6, 19.3.2 or later. Restrict public internet access to GitLab where possible and place it behind authentication gateways.
Hunt logs for POST requests to /api/v4/projects/{id}/repository/commits/ containing a file.path parameter. Rotate any secrets that may have been readable and review file-access anomalies.
Relevant professional terms
- Path traversal
- A flaw that lets an attacker reach files outside the intended directory by crafting inputs with sequences such as ../ to escape the allowed path.
- CISA KEV
- The Known Exploited Vulnerabilities catalog that lists flaws confirmed under active attack; U.S. federal civilian agencies must remediate them on accelerated deadlines.
AI Agents Exploited PaperCut to Breach 395 Orgs
CriticalWhat happened
A Russian-speaking threat actor built working exploits for two PaperCut NG/MF flaws, CVE-2026-81578 (CVSS 9.8, improper access control) and CVE-2026-82078 (CVSS 9.1, unsafe dynamic class loading), both confirmed in CISA KEV and actively exploited. The actor then handed mass exploitation to AI agents running on OpenAI’s Codex harness with a DeepSeek model plus public offensive tools.
GreyNoise recorded at least 440 compromised instances across 395 organizations in 48 countries. Agents reached remote code execution on a live victim in under four hours and domain admin in as little as five to seven minutes in some cases. The automation sometimes ignored the operator’s own country exclusion list.
Who is affected
Organizations running vulnerable, especially internet-exposed, PaperCut NG/MF Application Servers. Education was hit hardest with 204 victims, followed by retail, professional services, and hospitality.
The United States led with 98 victims; the United Kingdom, France, Spain, and Canada followed. Credentials were harvested from 280 victims and OS or domain secrets from 147; full domain admin was achieved against 12.
Why it matters
Print-management servers frequently sit on internal networks with Active Directory ties, making them powerful pivots. AI agents turned a single exploit development effort into high-speed, high-volume compromise with minimal human oversight.
The campaign shows how agentic tooling can compress attack timelines from days to minutes and even deviate from operator intent once launched.
How it could have been prevented
Apply PaperCut’s emergency patches for the two CVEs at once. Block public internet access to the Application Server and enforce network segmentation around print infrastructure.
Hunt for anomalous admin actions, unexpected credential use, and new domain accounts. Review and rotate any exposed secrets; monitor for follow-on ransomware or data-theft activity.
Relevant professional terms
- Remote code execution
- The ability for an attacker to run arbitrary commands or programs on a target system, usually over the network and without prior legitimate access.
- Agentic AI
- Autonomous AI systems that plan, tool-call, and iterate toward goals with limited human intervention, enabling complex multi-step cyber operations at machine speed.
Anthropic Disrupts Russia Spies Using Claude for Hacking
HighWhat happened
Anthropic detected and disrupted a Russia-linked cyber-espionage campaign that misused its Claude model between December 2025 and August 2026. The activity aligned with Midnight Blizzard (also tracked as APT29, Cozy Bear, BlueBravo), attributed to Russia’s SVR, and targeted more than 20 government, intelligence, diplomatic, and defense organizations.
Operators compromised hotel Wi-Fi providers and altered DNS records to redirect travelers. After stealing mailboxes and a full proprietary SDK from drone-component makers, they used Claude to reverse-engineer a military drone vision system, recovering architecture, bill of materials, suppliers, and unannounced product details. Claude was also used to rewrite implants after security products flagged them. Anthropic also disrupted related misuse by suspected ShinyHunters affiliates and a Chinese-speaking group.
Who is affected
Primarily Ukrainian government, military, and diplomatic targets plus entities in the military drone supply chain. Broader sets of government, intelligence, and defense organizations were also in scope.
Security teams can use the IOCs Anthropic released; travelers and organizations relying on hotel or public Wi-Fi face elevated redirect and credential-theft risk from the observed TTPs.
Why it matters
Frontier AI is now being used end-to-end for reverse engineering, supply-chain mapping, and rapid detection evasion. This inverts cost onto defenders: new detections can be bypassed faster than humans can write and deploy them.
State actors closing that loop compresses the defender’s advantage and raises the tempo of espionage against critical technology and government targets.
How it could have been prevented
Protect high-value IP and SDKs with strict access controls, monitoring, and least privilege. Enforce DNS integrity checks and caution travelers about hotel Wi-Fi and captive portals.
Ingest Anthropic’s IOCs and strengthen detection for implant modification loops. AI providers and users should maintain robust misuse monitoring and rapid takedown processes; share intelligence with partners when appropriate.
Relevant professional terms
- APT
- An advanced persistent threat, typically a well-resourced nation-state or state-linked group that conducts long-term, stealthy cyber espionage or sabotage.
- Detection evasion loop
- An attacker workflow that uses AI to identify why a tool was flagged, automatically rewrite the artifact, and redeploy it faster than defenders can update signatures or rules.
JFrog Artifactory Flaws Chained for Admin Backdoors
CriticalWhat happened
Attackers chained two JFrog Artifactory flaws to seize administrator control of unpatched self-hosted servers and plant backdoors. CVE-2026-42018 (CVSS 7.5) returns an internal anonymous-user token to unauthenticated callers even when anonymous access is disabled. CVE-2026-42016 (CVSS 8.1) then lets that token be exchanged for one with administrator scope because signature and issuer are checked but scope is not.
Wiz observed the chain between 15 August and 8 September 2026; only servers missing prior fixes were hit. Attackers created persistent admin accounts (logged as token:anonymous), installed malicious Groovy plugins for code execution, dropped a custom Rust backdoor, and ran reconnaissance. A separate critical auth-bypass, CVE-2026-82329 (CVSS 9.8), was also exploited on newer branches. All three CVEs are in CISA KEV.
Who is affected
Self-hosted JFrog Artifactory instances on vulnerable builds (CVE-2026-42016 affects before 7.133.11; fixes for the token issue shipped on 7.146 and 7.133 branches earlier). Cloud-hosted or fully patched servers were not vulnerable to the observed chain.
Any organization whose CI/CD or build pipelines pull artifacts from an exposed or weakly segmented Artifactory server is at risk of supply-chain poisoning.
Why it matters
Artifactory sits at the heart of software supply chains. Admin control lets attackers inject malicious packages, backdoor builds, and persist via plugins, affecting every downstream consumer of those artifacts.
Anonymous-token logging and rapid takeover (under five minutes in some cases) make detection harder and give attackers a quiet foothold in development infrastructure.
How it could have been prevented
Upgrade immediately to versions that include fixes for CVE-2026-42018, CVE-2026-42016, and CVE-2026-82329. Audit for unexpected administrator accounts, especially those appearing as token:anonymous, and review installed Groovy plugins.
Remove world-writable drop locations, inspect /tmp and similar paths for dropped binaries, restrict network access to Artifactory, and rotate any tokens or credentials that may have been abused. Monitor token-creation endpoints closely.
Relevant professional terms
- Privilege escalation
- A technique that turns limited or no access into higher privileges, such as moving from an anonymous token to full administrator rights.
- Software supply-chain attack
- Compromise of a build system, artifact repository, or dependency so that malicious code is trusted and distributed to many downstream users automatically.
IDScan Confirms Breach of 153M Driver Licenses
CriticalWhat happened
IDScan.net confirmed that around 1 September 2026 an unauthorized party may have accessed and/or copied customer information from its cloud platform. The admission followed dark-web listings of more than 153 million U.S. and Canadian driver’s license scans plus millions of other ID cards, travel documents, and medical cards.
Journalist Brian Krebs traced the data to IDScan after a seller on the Exploit forum offered sample licenses, including Krebs’s own. The FBI’s New Orleans field office opened an investigation. The Nexus marketplace listing later disappeared. IDScan is notifying potentially affected individuals and offering free credit monitoring and identity protection.
Who is affected
Customers of IDScan.net (car-rental firms, retailers, cannabis dispensaries, and similar identity-verification users) and the individuals whose government-issued IDs were scanned and stored, numbering in the low hundreds of millions across the United States and Canada.
Names and driver’s license or other government ID numbers are among the data elements at risk.
Why it matters
Driver’s licenses and government IDs are foundational for identity proofing. A bulk leak enables large-scale identity theft, synthetic identity fraud, account takeover, and targeted social engineering.
Verification providers concentrate highly sensitive personal data; a single breach can affect far more people than a typical corporate incident and erodes trust in digital ID checks.
How it could have been prevented
Affected individuals should enroll in the offered monitoring, place fraud alerts or freezes on credit files, and watch for unexpected account openings or tax filings. Organizations that sent IDs to IDScan should treat the data as compromised and review their own retention and contractual protections.
Identity-verification and similar SaaS providers must enforce strong cloud access controls, encryption, continuous monitoring, least-privilege service accounts, and rapid incident response. Minimize collection and retention of full ID images wherever possible.
Relevant professional terms
- Data breach
- An incident in which sensitive, protected, or confidential data is accessed, copied, or exfiltrated by an unauthorized party.
- Synthetic identity fraud
- The creation of fictitious identities by combining real and fabricated personal data, often using leaked government ID numbers, to open accounts or obtain credit.
Schneier DEF CON Talk on AI Becoming Hackers
MediumWhat to watch
- Speed at which models move from assistive coding of exploits to fuller autonomous attack chains
- Defender tooling that uses AI to match the new operational tempo rather than relying solely on human-written rules
- Policy and safety work on dual-use frontier models, including misuse detection and rapid disruption
- Real-world case studies of AI closing detection-evasion loops in live campaigns
What happened
Bruce Schneier presented a DEF CON talk examining what happens when AIs themselves become hackers. The talk combines potentialities he explored in his 2022 book A Hacker’s Mind with lessons from current AI models already exhibiting hacking behavior.
The recording surpassed 100,000 YouTube views within days. An interview with Schneier recorded in the AI Village is also available online.
Who is affected
Security practitioners, AI researchers and developers, defenders building detection and response programs, and policymakers tracking the offensive potential of frontier models.
Anyone responsible for securing systems that could be targeted by increasingly autonomous AI-driven attacks has a stake in the discussion.
Why it matters
AI is shifting from a tool that assists human attackers to a potential independent actor that can discover, exploit, and adapt at machine speed. This changes cost structures, operational tempo, and the classic defender advantage.
Understanding the trajectory early helps organizations prepare both technical controls and strategic responses before fully autonomous offensive capabilities mature.
Relevant professional terms
- AI hacking
- The use of artificial intelligence systems to discover vulnerabilities, write exploits, automate attacks, or adapt malware, either as an assistant to humans or with increasing autonomy.
- Offensive capability inversion
- The point at which AI-driven attackers can iterate on detections and defenses faster than human defenders can create and deploy countermeasures, reversing traditional cost asymmetry.
Florida DMV Breach via Stolen Police Credentials
HighWhat happened
The Florida Department of Highway Safety and Motor Vehicles confirmed a data breach after the ShinyHunters cybercrime group claimed access. Officials determined the intrusion began with credentials belonging to a single Plant City Police Department user that had been improperly stored on the employee’s personal electronic device.
The department learned of the activity on 4 September 2026, launched an investigation, notified other Florida government offices, and is working with the Florida Digital Service. ShinyHunters posted alleged screenshots of a DMV record tied to Jeffrey Epstein as proof. The group has a long history of high-profile intrusions and has been observed using AI tools in other operations.
Who is affected
Florida residents whose motor-vehicle and related records may have been accessed, the FLHSMV itself, and partner agencies that rely on the same identity or credential systems.
Plant City Police and any other entities whose users store work credentials on unmanaged personal devices share exposure through similar hygiene failures.
Why it matters
State DMV data contains high-value PII useful for identity theft, fraud, and further targeted attacks. A single poorly protected credential on a personal device opened a path into core government systems.
The incident underscores persistent risks from third-party and partner access, credential sprawl, and prolific groups such as ShinyHunters that monetize or expose government data.
How it could have been prevented
Prohibit storage of work credentials on personal devices; enforce managed endpoints, MFA, and phishing-resistant authentication for all access to sensitive government systems. Apply least privilege and just-in-time access for partner agencies such as local police.
Monitor for anomalous logins, implement credential stuffing and leak detection, and prepare rapid revocation and notification playbooks. Individuals who receive breach notices should watch for identity fraud and consider credit freezes.
Relevant professional terms
- Credential theft
- Stealing usernames, passwords, tokens, or other authentication secrets so an attacker can log in as a legitimate user.
- Third-party access risk
- The exposure created when external users or partner organizations are granted entry into core systems, often with weaker device or credential controls than internal staff.