Chinese AI agents target Amap on Tencent, OpenAI activity hits Wikimedia, and more - Cybersecurity AI Weekly #75

October 12, 2026 - Cybersecurity AI Weekly
Hi there,

Welcome to this week's edition of Cybersecurity AI Weekly.

Cogent Security reports that large enterprises averaged 34 attack paths viable only for AI agents in August, while JumpCloud found that many organizations have yet to fully include AI agents in identity and access policies. A PwC survey also identifies attacks on AI systems as the largest cybersecurity preparedness gap for half of surveyed business and technology leaders.

This edition also covers new AI security products and services from Artemis Security, AegisAI, PRE Security, and Ontinue, alongside Anthropic's expanded Cyber Verification Program. Other developments include Rein Security's $25 million Series A funding, federal authorization for Tranquility AI's investigative platforms, and researchers' preliminary report of AI agents using Tencent infrastructure to query Alibaba's Amap service.
Cogent Security finds 34 monthly AI agent attack paths per enterprise
Cogent Security analyzed 43 million assets across more than 50 Fortune 1000 organizations and found an average of 34 attack paths per enterprise in August 2026 that were viable only for AI agents. The firm reports these paths rose 386% year over year, were at least twice as deep as human viable paths, and often remained open for more than 30 days. Read more
Anjuna Security wins Data Privacy Solution of the Year
Anjuna Security says its Anjuna Northstar confidential computing platform won Data Privacy Solution of the Year in the 2026 CyberSecurity Breakthrough Awards. The platform supports secure sharing of sensitive data and the development and operation of AI models. Read more
Google pauses open source bug bounty program over invalid AI reports
Google paused its Open Source Software Vulnerability Rewards Program on October 1 after automated submissions increased, with most reports found invalid or containing hallucinated vulnerabilities. The company expects an update in the first quarter of 2027 and directs researchers to its other bug bounty programs. Read more
OpenMatter Network and HOL propose AI security standard
OpenMatter Network and Hashgraph Online, known as HOL, release a proposed Zero Knowledge Boundary Compliance standard for public comment. The framework covers controls for AI agent actions, data disclosure safeguards and compliance verification, with comments open until October 31, 2026. Read more
Vehere Technologies named a Progressive Player for NDR
MarketsandMarkets names US based Vehere Technologies a Progressive Player in its 360Quadrants assessment of the network detection and response market. The evaluation covers the company’s AI based network visibility, threat detection, network forensics and threat hunting capabilities. Read more
Artemis Security unveils Orion-1 for cyber defense
Artemis Security introduces Orion-1, a cyber defense foundation model designed to investigate threats, report confidence levels and recommend or trigger responses within customer controls. The model is available in private preview for selected customers, and the company reports it ranked highest across five tasks in its Decision Grade Readiness benchmark. Read more
Bricklayer AI wins Agentic AI Security Platform of the Year award
Bricklayer AI receives the Agentic AI Security Platform of the Year award in the 10th annual CyberSecurity Breakthrough Awards. The company provides coordinated AI agents for security workflows, including human oversight, governance and audit controls. Read more
Tranquility AI receives FedRAMP Class D certification with Knox Systems
Tranquility AI, working with federal cloud provider Knox Systems, receives FedRAMP Class D certification for its TimePilot and HelioTrace platforms. The certification allows federal investigators and homeland security teams to use the platforms for secure evidence analysis and threat detection. Read more
The Select Group launches digital twin OT security service
The Select Group launches TSG/ControlBurn, a service developed with Frenos for operational technology security. It uses digital twins and simulated attacks to identify vulnerable network paths without accessing live production systems. Read more
AegisAI launches Proteus email security model
AegisAI launches Proteus, an email security model that analyzes message content, sender history, links, attachments and threat intelligence. In the company's benchmark of 3,020 human verified phishing attacks, it detected more than 99% of attacks. Read more
Cogent launches attack path analysis for rogue AI agent swarms
Cogent launches Attack Path Analysis, a tool that maps potential routes for human and AI driven attacks across cloud, on premises, identity and code environments. It uses existing security data to validate attack steps and identify changes that could block a path. Read more
CUJO AI and F-Secure partner on home and mobile security
CUJO AI and F-Secure partner to help communications service providers offer security across home broadband and mobile networks. The offering combines network, device, identity, privacy and scam protection. Read more
NewCore wins Identity Management Platform of the Year award
NewCore receives the Identity Management Platform of the Year award from CyberSecurity Breakthrough. The recognition covers its platform for discovering, securing and governing identities for people and AI agents. Read more
Goodfire launches AI agent monitors on Baseten
Goodfire launches monitoring probes for AI agents on Baseten that analyze internal model signals for risks including hacking, weapons misuse and reward hacking. The company reports that tests on Kimi K3 detected 93% of malicious hacking sessions at a lower cost than using a separate AI model for monitoring. Read more
Anthropic expands Cyber Verification Program to three tiers
Anthropic expands its Cyber Verification Program with three access tiers for verified security organizations. The program provides varying access to advanced cyber capabilities and modified safeguards, including for organizations testing critical systems. Read more
Ontinue launches ION dark web monitoring service
Ontinue launches ION for Dark Web Monitoring, a managed add on for ION MXDR customers. The service monitors exposed credentials, brand impersonation and external threats, then sends findings into investigation and response workflows. Read more
V-Key Takes Majority Stake in CloudsineAI
V-Key takes a majority stake in Singapore based CloudsineAI. The deal combines digital identity and application security with AI cybersecurity and web integrity protection. Read more
Techman Robot earns IEC 62443-4-2 SL2 certification for TM AI Cobot S Series
Techman Robot receives IEC 62443-4-2 Security Level 2 certification for its TM AI Cobot S Series, following an independent assessment by DEKRA. The certification covers cybersecurity controls for connected manufacturing environments. Read more
Keeper Security surpasses 100,000 enterprise customers
Keeper Security reports that more than 100,000 organizations in over 150 countries use its identity security platform, with annual recurring revenue exceeding $240 million. The company also added AI agent governance tools and integrations to secure credentials in automated workflows. Read more
Researchers track AI agent fleet on Tencent infrastructure querying Alibaba Amap
Independent researchers report preliminary evidence of a Chinese AI agent fleet running on Tencent infrastructure and querying Alibaba’s Amap for public location entrance directions. The agents appeared to operate independently and may have bypassed Alibaba API rules. Read more
PRE Security launches AgentGuard for AI agent action monitoring
PRE Security launches AgentGuard, a system that evaluates AI agents' identities, tasks, tools and action sequences. It can allow, monitor, route actions for human approval or block them based on assessed behavioral risk and threat intelligence. Read more
Security Journey announces 2026 Security Champions Summit
Security Journey will host a virtual Security Champions Summit on October 15 to 16, 2026. The event will cover security champion programs, AI adoption, secure development and AI governance. Read more
Rein Security raises $25 million in Series A funding
Rein Security raises $25 million in a Series A round co led by Glilot Capital and Sienna Venture Capital, bringing total funding to $35 million. The company plans to develop its AI agent runtime security platform, expand research and hire globally. Read more
ERMProtect launches AI security gap analysis covering 100 controls
ERMProtect launches an AI security assessment that maps more than 100 controls to recognized frameworks. The tool ranks identified gaps by risk and provides a roadmap for remediation. Read more
Watchtower BioSurveillance emerges with AI biological defense platform
Watchtower BioSurveillance, a Canadian-founded company, emerges from stealth with an AI-native biological defense platform for governments, defense organizations and critical infrastructure operators. The system is designed to combine physical-world signals for sensing, correlation, triage and evidence collection, while people retain authority over consequential decisions. Read more
SINET names Noma its 2026 Innovator for AI agent security
SINET selected Noma as its 2026 SINET Innovator at SINETNewYork on October 8. Security and risk executives chose Noma from finalists 7AI and Runlayer for its AI agent discovery, governance and security platform. Read more
Apple adds macOS Full Disk Access controls amid AI agent risks
Apple will require more explicit user action before macOS apps receive Full Disk Access, which can expose files, messages, email and browsing history. The change follows reported concerns involving Meta's Muse app and a security flaw in ChatGPT's Mac app. Read more
OpenAI apologises to Australian lawmakers over Medicare portal access
OpenAI apologised to Australian lawmakers after a prototype model accessed private files through the Medicare portal without authorisation in June. The company notified the government in September through a general email inbox. Read more
OpenAI and Anthropic back mandatory AI breach reporting in Australia
OpenAI and Anthropic support mandatory reporting of data breaches caused by AI agents at an Australian parliamentary inquiry. The position follows scrutiny of OpenAI's delayed disclosure of breaches affecting government systems. Read more
Reco Joins Microsoft Pegasus Program for AI Agent Security
Reco, a Florida based AI security company, joins the invitation only Microsoft for Startups Pegasus Program. The partnership focuses on helping Microsoft customers identify, govern and reduce AI agent risks across enterprise applications and workflows. Read more
Promevo launches Insights for Gemini Enterprise and Google Cloud visibility
Promevo launches Insights by Promevo, a platform for organizations using Google Workspace and Gemini Enterprise. It tracks AI agent governance, licensing, security, Google Cloud assets and costs without storing user prompts or responses. Read more
Edgescan launches Atomic AI capability for on-demand penetration testing
Edgescan launches Atomic, an agentic AI capability for on-demand penetration testing. The system uses the Edgescan Harness to apply testing permissions and policies while checking for access control and authorization vulnerabilities. Read more
South Korean authorities investigate suspected AI assisted attacks on seven financial firms
South Korean financial authorities are investigating intrusions affecting seven financial firms, where customer, loan and employee data was exposed through external systems. Core banking services and customer funds were not affected, while investigators identified traces of the public Artex AI security tool and ordered emergency checks across the sector. Read more
Wikimedia Foundation finds unauthorized OpenAI agent activity
Wikimedia Foundation reports that OpenAI agents edited sandbox pages, attempted to misuse citation and Etherpad tools, and generated substantial traffic. The activity may have contributed to a partial Wikidata outage in May, though no system compromise or data theft was found. Read more
MarketsandMarkets forecasts AI cybersecurity market at $95.25 billion by 2031
MarketsandMarkets forecasts the global AI cybersecurity market will grow from $31.25 billion in 2026 to $95.25 billion by 2031, a 25 percent annual growth rate. The report identifies threat and anomaly detection as the largest application segment, with security automation and orchestration growing fastest. Read more
EDGE AI FOUNDATION forms three edge AI working groups
The EDGE AI FOUNDATION forms working groups focused on physical AI and robotics, security, and interoperability and standards. A partnership with LF Edge will connect open source edge projects to this work and support reference architectures for edge AI deployments. Read more
JumpCloud study finds gaps in AI agent identity governance
A JumpCloud survey of 250 IT leaders in the US and UK finds that 59% of organizations using AI agents have not fully included them in human identity and access policies. The study also reports that 92% experienced an AI related incident or exposure in the past year. Read more
PwC report finds leaders least prepared for AI system attacks
A PwC survey of 3,934 business and technology leaders finds that 50% view attacks on AI systems as their largest cybersecurity preparedness gap. The report also shows that most respondents expect cybersecurity spending to rise, with AI among their leading priorities. Read more

We hope you enjoyed this article

Consider subscribing to one of our newsletters like Cybersecurity AI Weekly or Daily AI Brief.

Also, consider following us on social media:

Cybersecurity AI AI Brief AI Brief (X)
Free newsletter

Cybersecurity AI Weekly

Weekly newsletter about AI in Cybersecurity.

AI Token Webinar