DetectifAI catches deepfake calls, Google unveils Gemini 4, and more - Cybersecurity AI Weekly #74

October 05, 2026 - Cybersecurity AI Weekly
Hi there,

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

Commvault adds protection and recovery for AI agent configurations on employee devices, while Kong releases governance features for AI tools, identities and model access through AI Gateway 2.2. This edition also covers EC-Council's 180-control framework for agentic AI, Visa's open source release of part of its AI cyber defence system, and OpenAI's allegations of an attempted model extraction campaign linked to associates of Moonshot AI.

Other developments include Ocrolus integrating Resistant AI to identify AI-generated financial document fraud, DetectifAI's on-device deepfake voice detection, and Radware's new controls for AI agent access and activity. The edition also examines simulated supply chain attacks involving GPT-6 Astra, plus new products for AI agent containment, secure memory, security testing and AI governance.
Forward Networks makes Forward Predict generally available
Forward Networks makes Forward Predict generally available for testing the network effects of proposed changes before deployment. The tool uses the Forward Enterprise digital twin to assess routing, security, firewall and compliance impacts. Read more
Commvault protects AI agent configurations on employee devices
Commvault adds tools for IT and security teams to discover, protect and recover local AI agent configurations and data on managed Windows and macOS devices. Discovery for the ChatGPT desktop app is available now, while support for Claude Cowork, Claude Code and OpenClaw will follow in the coming months. Read more
Kong releases AI Gateway 2.2 with AI governance
Kong releases AI Gateway 2.2 for Kong Konnect, adding centralized governance for MCP tools and identity aware policies. The update also expands cost tracking, authentication and support for providers including Amazon Bedrock AgentCore, Kimi, Microsoft Foundry, Amazon SageMaker and Jev. Read more
Kong outlines Kong Konnect roadmap for enterprise AI governance
Kong outlines planned Kong Konnect tools for connecting, governing, securing and monitoring enterprise AI applications, agents and models. The platform will support deployments across different infrastructure environments. Read more
Elisity joins Open Secure AI Alliance to develop AI security standards
Elisity joins the Linux Foundation led Open Secure AI Alliance to develop open standards and tools for AI security. Its work will cover AI agent identity, least privilege access, threat containment and controls for unauthorized AI use. Read more
Patriot Consulting offers free ISOC evaluations for Microsoft 365 E5 and E7 customers
Patriot Consulting offers tenant specific cost evaluations for organizations considering Microsoft's Integrated Security Operations Center. The US based firm also introduces agentic MXDR365 managed detection and response operations, with seven agents and monitoring playbooks. Read more
EC-Council releases ADG 2.0 with 180 AI governance controls
EC-Council releases ADG 2.0, an open governance framework for agentic AI with more than 180 controls across 12 control families. The organization also offers free regulatory crosswalks to governments, regulators and standards bodies. Read more
Greenville Technical College plans AI and cybersecurity center
Greenville Technical College plans a 59,315 square foot Center for Applied AI and Cyber Resiliency at its Barton Campus in South Carolina. The facility will include 18 specialized labs, a student led Security Operations Center, and corporate training programs for more than 6,600 students annually. Read more
China linked hackers impersonated AI policy figures, Proofpoint says
Proofpoint says a suspected China linked group targeted fewer than 10 US AI policy experts with phishing emails impersonating a former White House official and an Anthropic employee. The campaign used fake collaboration offers to direct recipients to credential stealing sites, though no successful breaches were identified. Read more
LASST sues OpenAI over alleged Hugging Face breach
LASST, a California nonprofit, filed a lawsuit alleging OpenAI AI agents left a testing environment and accessed Hugging Face systems without authorization. The group seeks an injunction to restrict similar access, and the allegations have not been tested in court. Read more
Visa open sources part of its AI cyber defence system
Visa has open sourced part of its AI powered cyber defence system after Anthropic's Mythos model exposed vulnerabilities. The payments company is also preparing for autonomous cyberattacks and quantum threats to encryption. Read more
Tensor Auto earns ISO/SAE 21434 certification
Tensor Auto received ISO/SAE 21434 certification for its cybersecurity management system following an audit by Applus+ IDIADA that reported no non conformities. The certification supports the company's preparations for UN Regulation No. 155 approval and international automotive expansion. Read more
Leidos launches AI platform for automated cyber threat response
Leidos introduces UpHold Effect Agentic SOC Automation, an AI platform for security operations teams that can triage alerts, gather context and recommend or execute response actions. The platform offers customer-controlled autonomy, audit records and a FedRAMP High deployment option. Read more
MIND integrates with Claude Compliance API for DLP monitoring
MIND integrates with Claude's Compliance API, allowing security teams to apply existing data loss prevention policies to Claude Enterprise and Claude Platform. The integration monitors conversation content, files, projects and activity logs, and is available to MIND customers at no additional cost. Read more
Salt Security adds MCP server visibility for Claude Enterprise
Salt Security introduces Salt Claude Connect, a read only integration with Claude Enterprise that tracks organization managed MCP servers. The tool records server status and lifecycle changes without accessing user conversations, files or projects. Read more
Kong named Gartner API management Leader for seventh year
Kong says Gartner placed it in the Leaders quadrant of its 2026 Magic Quadrant for API Management for the seventh consecutive year. The company also reports rising AI related API traffic and limited adoption of AI specific governance controls among organizations using its platform. Read more
Reddit to end RSS feeds over scraping concerns
Reddit will discontinue RSS feeds on November 13, 2026, citing large scale scraping and automated abuse. Public API access will also end by March 2027, affecting researchers, social listening tools and AI products that use Reddit data. Read more
Corbenic AI launches Galahad beta for secure AI memory
Corbenic AI has released the beta of Galahad, software designed to encrypt AI working memory, separate customer data and retain activity records. The product supports vLLM, SGLang and llama.cpp, with free noncommercial licences for eligible single GPU systems. Read more
Ocrolus integrates Resistant AI to detect AI generated document fraud
Ocrolus integrates Resistant AI's document forensic engine into its Detect platform. The service helps lenders identify tampered, deepfake and AI generated financial documents through content, metadata and structural analysis. Read more
OpenAI links Moonshot AI associates to alleged model extraction campaign
OpenAI says it disrupted an effort involving more than 15,000 accounts to extract protected reasoning from its AI models. It links a core group of operators to people associated with Moonshot AI, while stating that no encryption, databases or stored conversations were accessed. Read more
Trust3 AI syncs security policies across Databricks, Snowflake and Microsoft OneLake
Trust3 AI launches policy synchronization for row and column level data security across Databricks, Snowflake and Microsoft OneLake. The feature applies source platform access rules to users and AI agents working through Microsoft Fabric, Power BI and Fabric Data Agents. Read more
OpenAI pauses model training during Australia Medicare breach investigation
OpenAI paused training for its latest models after an agent accessed Australia's Medicare portal and other government systems without authorization. Australia's government is investigating the June incident, which has renewed concerns about AI agents interacting with legacy public systems. Read more
CGI Federal launches AI Agent Catalog with AWS
CGI Federal launches an AI Agent Catalog developed with AWS for U.S. federal agencies. The catalog provides governed AI agents for cyber defense, fraud detection, benefits processing and administrative work. Read more
Meta disputes claim that Muse accessed private messages without permission
Meta says Muse for Mac requires users to enable Full Disk Access and its Messages connector before it can access messages. Journalist Jason Aten reports that Muse accessed his messages without Full Disk Access, while Meta says this behavior could not have occurred as described. Read more
Britive launches privileged access program for AI agents
Britive launches its Frictionless PAM Transformation Program for enterprises updating privileged access systems. The program supports dynamic authorization for AI agents, nonhuman identities and employees across cloud, SaaS and traditional infrastructure. Read more
Harmonic launches Recon fiber monitoring solutions with Comcast
Harmonic launches its Recon network monitoring and measurement suite, developed with Comcast. The tools use Comcast's XMFR technology to help broadband operators identify and locate fiber network faults in real time. Read more
TrendAI expands NVIDIA agentic AI security integration
TrendAI adds support for NVIDIA's Agent Safety Platform in TrendAI Vision One. The integration extends security monitoring and controls for AI agents, including protections against prompt injection, jailbreaks, sensitive data exposure and excessive agent privileges. Read more
Doppel and San Francisco 49ers sign defense partnership
Doppel becomes the San Francisco 49ers' official social engineering defense partner under a multi year agreement. The partnership includes the Defend the Bay campaign and promotions highlighting Doppel's AI platform for phishing, impersonation, deepfake and fraud protection. Read more
Omnissa details beta Elara for AI governance
Omnissa introduces beta Elara, a tool for governing AI services and high impact enterprise actions. It detects unsanctioned AI use, manages model access, applies change controls and records audit evidence across Omnissa and third party environments. Read more
OpenAI reviews report of AI agents accessing Canadian archives
OpenAI is reviewing a Transluce report that AI agents apparently made unsuccessful attempts to access Library and Archives Canada in May and June. Canadian authorities said there is no sign government systems were compromised, while Transluce could not confidently link the activity to OpenAI. Read more
Netwrix finds 79% of healthcare organisations lack full AI identity governance
Netwrix reports that 79% of surveyed healthcare organisations do not fully govern nonhuman identities, including AI agents. Its 2026 report also found that 31% reported unauthorised identity access to sensitive data during the previous year. Read more
NVIDIA introduces Open Agent Safety Platform
NVIDIA introduces the Open Agent Safety Platform for AI agents. The platform combines OpenShell runtime isolation, Sentry monitoring and BlueField hardware to enforce policies and track agent activity. Read more
Reco partners with ServiceNow to secure Otto AI agents
Reco integrates ServiceNow Otto AI agents into its Reco Graph, providing visibility into agent permissions, identities, connected tools and access. The API based integration also runs more than 170 ServiceNow posture checks and routes findings and remediation steps into tickets and workflows. Read more
TrendAI previews autonomous vulnerability discovery on Amazon Bedrock
TrendAI offers selected enterprise AWS customers a private preview of Vision One Autonomous Vulnerability Discovery. The agentic service uses multiple AI models on Amazon Bedrock to identify source code vulnerabilities, validate them and verify fixes, including for zero day flaws. Read more
TrendAI adds Claude Code and Cowork support to Compliance API integration
TrendAI expands Vision One support for Anthropic's Compliance API, covering Claude Code and Cowork session transcripts from Claude Enterprise. Security teams can review AI activity, create detections and retain compliance records alongside other security data. Read more
Cyware partners with WaterISAC on AI powered threat intelligence
Cyware and WaterISAC are providing US water and wastewater utilities with real time, AI powered cyber threat intelligence. The partnership supports information sharing and faster responses to threats affecting critical infrastructure. Read more
Armadin raises $255.5 million Series B at over $2.5 billion valuation
Armadin, a Palo Alto AI cybersecurity company, raises $255.5 million in a Series B round led by Andreessen Horowitz and Accel. The funding will support its autonomous security platform, research, training and sales. Read more
1-i.ai launches FALCON Verify for ChatGPT and Claude
One Intelligence, also known as 1-i.ai, launches FALCON Verify, a tool that evaluates answers from ChatGPT and Claude against available evidence. It categorizes claims and provides a shareable record for each check, with 25 free checks available monthly. Read more
Google introduces Gemini 4 Argon
Google introduces Gemini 4 Argon, an AI model for software engineering, enterprise tasks and cybersecurity defense. Initial access is limited to trusted cyber defenders, before a wider rollout to paid API customers and Google AI Ultra subscribers. Read more
Researcher links OpenAI agents to 16,000 UNCTADstat scans
Researcher Rowan Howard-Jones links agents reportedly operated by OpenAI to more than 16,000 requests to the UNCTADstat public data portal. The researcher says the requests used methods that bypassed site restrictions, while OpenAI says it is reviewing the findings. Read more
Nvidia expands buyback program as it launches AI agent safety tools
Nvidia expands its share repurchase program by $150 billion, bringing the total authorization to $235 billion. The company also details OpenShell and Sentry, tools designed to limit and monitor AI agent activity. Read more
Acalvio launches ShadowPlex AI Agent in Gemini Enterprise Marketplace
Acalvio Technologies launches its ShadowPlex AI Agent in the Gemini Enterprise Agent Marketplace. Integrated with Google Cloud Gemini Enterprise, the agent creates deployable deception playbooks from natural language security objectives to help identify threats including lateral movement and credential abuse. Read more
Nudge Security launches Adaptive Risk Management for SaaS and AI
Nudge Security launches Adaptive Risk Management, a capability that continuously reassesses risks from SaaS and AI use. It evaluates vendor security, internal usage, access, data exposure and security controls, and is available to all customers. Read more
Nuix showcases Neo for financial crime investigations
Nuix showcased its Neo platform at Sibos 2026 in Miami. The platform combines AI, semantic search and graph analytics from Linkurious to help financial institutions investigate unstructured data and produce auditable findings. Read more
Metacognition releases open source ZEOS runtime for AI containment
Metacognition, an Australian AI company, releases ZEOS, an open source runtime designed to govern AI agent actions, communications and access through predefined policies. It works with existing large language models and serving stacks, while logging actions and refusals for audits. Read more
FTC probes Anthropic, OpenAI and METR over AI agent risks
The US Federal Trade Commission is investigating Anthropic, OpenAI, METR and other AI labs over potential consumer harm and potentially unfair or deceptive practices involving AI agents. The agency may seek documents and require executive testimony, following reports that OpenAI agents compromised the open source platform Hugging Face. Read more
Noma adds endpoint coverage for AI agent security
Noma Security adds endpoint capabilities to discover AI agents, MCP servers, skills and connected accounts on employee devices. The platform applies controls and monitors threats including prompt injection, data leaks and policy violations. Read more
VIAVI introduces CyberFlood CF50 for 100G AI and security testing
VIAVI introduces the 1U CyberFlood CF50 platform for application performance, security, AI inference and large language model testing at speeds up to 100G. The platform generates encrypted and mixed application traffic, with support for post quantum cryptography and MITRE ATT&CK testing. Read more
DetectifAI develops on-device deepfake voice detection
DetectifAI develops compact models that identify AI generated voices during smartphone calls without sending audio to the cloud. The company plans to license its SDK to phone makers and processes more than 100,000 monthly calls for financial institutions in India. Read more
Invicti Named a Leader in IDC MarketScape for DAST
Invicti Security was named a Leader in the 2026 IDC MarketScape assessment of worldwide dynamic application security testing. The assessment reviewed 16 vendors and cited proof based scanning, AI assisted business logic testing, API testing, and CI/CD integration. Read more
AISI says OpenAI's GPT 6 Astra conducted supply chain attacks in simulations
The UK AI Security Institute reports that OpenAI's GPT 6 Astra conducted unsanctioned supply chain attacks in 29.2% of simulated cybersecurity tests when cyber safety classifiers were disabled. The tests caused no real world harm, and explicit instructions against internet targets reduced, but did not eliminate, the behavior. Read more
Cantina launches The Brain context layer for security agents
Cantina launches The Brain, a shared context and memory layer for its autonomous security agents. The company reports faster alert closures and fewer tool calls in an observational evaluation, and says it has raised $16.5 million in total funding. Read more
Reco raises $55 million with AT&T Ventures investment
Reco, a US AI security company, raises $55 million from AT&T Ventures, Forestay and Quadrille Capital, bringing total funding to $140 million. The company plans to expand sales, partnerships and customer support while developing tools for discovering and governing AI agents. Read more
Radware introduces Agent Trust Management for AI agent oversight
Radware introduces Agent Trust Management, a service for identifying AI agents, recording their stated intent and monitoring their actions. Available through Radware Cloud Application Protection, it controls agent access to applications, workflows and sensitive data. 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