started · updated
Tenable and OpenAI launch AI agent vetting process to counter cyber risks
The rapid adoption of agentic AI is creating significant cybersecurity challenges, prompting new defensive measures and highlighting emerging risks. Tenable and OpenAI have partnered to launch the CyberAgents Exchange AI Inspector, a security review process designed to vet AI agents, skills, and multi-agent playbooks before enterprise deployment. This initiative utilizes OpenAI GPT cyber models and Tenable’s security expertise to inspect community-built components within the open-source CyberAgents Exchange registry.
Security experts warn that AI agents operate at machine speed, potentially bypassing traditional identity and access management systems designed for humans. Risks include 'shadow AI'—the unsanctioned use of AI tools by employees—and the ability of agents to escalate privileges or leak sensitive data. In Australia and New Zealand, AI data exposure has become the second most common AI-related security incident. Additionally, Google reported that attackers are using AI coding chatbots and multi-agent frameworks to automate mass credential harvesting campaigns.
To counter these threats, companies like Orchid Security are introducing AI agent controls, such as identity drift detection and application-level kill switches. Meanwhile, the cybersecurity landscape is seeing a surge in AI-driven phishing attacks, which tripled in volume in 2025, and a massive increase in fraudulent e-commerce portals.
Entities
Anthropic · Claude Mythos · CrowdStrike · CyberAgents Exchange · CyberAgents Exchange AI Inspector · Eric Doerr · Google · KPMG · Netskope · OKTA · OpenAI · Orchid Security
Claims
What the coverage asserts, and how many sources carry each claim.
- [○ 1 SOURCE] AI agents are increasingly capable of performing autonomous cyberattacks and human deception.
- [○ 1 SOURCE] AI agent labor hours exceeded human researcher labor hours in June.
- [○ 1 SOURCE] The Claude Mythos model identified 181 working security vulnerabilities in the Firefox browser.
- [○ 1 SOURCE] OpenAI aims to develop an automated AI researcher under human supervision by March 2028.
- [○ 1 SOURCE] Claude Mythos identified a 27-year-old hidden OpenBSD bug and a long-standing FreeBSD vulnerability.
- [○ 1 SOURCE] AI models can be modified to bypass safety measures to assist in creating explosives, chemical, or biological weapons. www.origo.hu
- [○ 1 SOURCE] A method exists that can disable most AI model safety measures within minutes. www.origo.hu
- [○ 1 SOURCE] AI is evolving from simple chatbots into agents that can access corporate systems and execute tasks.
- [● 2 SOURCES] Tenable and OpenAI are collaborating to create the CyberAgents Exchange AI Inspector to vet AI agents and components. smestreet.in · astig.ph
- [○ 1 SOURCE] AI data exposure is the second most common AI-related security incident in Australia and New Zealand. australiancybersecuritymagazine.com.au
- [○ 1 SOURCE] The volume of phishing attacks tripled in 2025 due to AI-driven improvements in email quality and scale. world-today-journal.com
- [● 2 SOURCES] The Exchange Inspector uses OpenAI GPT cyber models, Tenable One AI Exposure, and Tenable researcher reviews to inspect AI agents and components. smestreet.in · astig.ph