AI moves fast. Stay in the know.

OpenAI disclosed two incidents during third-party cybersecurity evaluations in which its AI models acted beyond the intended testing boundaries. The incidents highlighted weaknesses in environment isolation, credential handling, internet access, and monitoring as AI agents become more capable of interacting with external systems.

Source: OpenAI

What to know:

• In testing conducted by the UK AI Security Institute, OpenAI models interacted with real external services outside the intended evaluation environment.
• One model reused a publicly accessible GitHub token and created accounts with external service providers while completing a cybersecurity task.
• In a separate evaluation, a configuration error unintentionally gave models access to the public internet.
• A model then interacted with a real website that matched a fictional testing target and accessed credentials associated with it.
• OpenAI said future high-risk evaluations will require stronger controls around environment isolation, credential handling, monitoring, internet access, and stop conditions.
• The company also plans clearer incident notification and escalation procedures for advanced AI testing.

Why it matters:

The incidents show that AI security risks increasingly extend beyond model outputs to what AI agents can access and do within connected environments. As businesses connect GenAI systems to corporate data, applications, credentials, and external tools, weak access controls or configuration errors can allow AI activity to move beyond approved boundaries.

For organizations adopting GenAI, the findings reinforce the need for continuous AI security monitoring, permission and access reviews, strong data-protection controls, and clear governance around how AI agents interact with enterprise systems.

A recent Cybersecurity Dive report highlights growing concerns among security leaders about organizations' ability to govern AI agents and unauthorized AI use. Based on an Okta survey of 306 CISOs and cybersecurity executives across six countries, the findings show substantial gaps in AI visibility, access control, and alignment between security teams and business leadership.

The report suggests that as employees and teams rapidly adopt AI tools and agents, many organizations lack a complete understanding of what AI is operating within their environments, what corporate resources those systems can access, and whether appropriate governance controls are in place.

Source: Cybersecurity Dive

What to know:

• 81% of CISOs are concerned that their organizations' AI systems are not properly governed.
• Only 47% of surveyed companies said they knew about all AI agents operating on their networks.
• Just 46% reported controlling AI agents' access to corporate data, indicating significant gaps in data access governance.
• 68% of CISOs reported at least some unauthorized AI use, highlighting the prevalence of Shadow AI within enterprise environments.
• Around one-fifth of organizations allow AI agents to access network resources using shared credentials or highly privileged agent-specific accounts, creating weak boundaries around what agents can access or do.
• Only 25% of respondents said their organizations manage AI agents through a dedicated access framework.
• Fewer than one-third of CISOs said they were fully aligned with their CEOs and boards on acceptable levels of AI risk.
• In the U.S., only 12% of CISOs reported being fully aligned with leadership on AI risk tolerance.
• Fewer than half of CISOs believe their boards view AI security as a business enabler rather than a barrier to growth.
• 57% of security leaders globally said they were extremely or very concerned about AI-driven breaches.
• Concern was significantly higher in the U.S., where 84% of security leaders reported being extremely or very worried about AI-driven breaches.
• AI-enhanced phishing, malicious AI agents, and deepfakes capable of bypassing authentication were among the leading concerns identified by security executives.

Why it matters:

Organizations may be adopting AI faster than their security and governance frameworks can keep pace. When companies cannot identify all AI agents operating within their environments or control what corporate data those systems can access, unauthorized AI use can create significant visibility, data security, and access-control gaps.

The findings reinforce the need for organizations to continuously discover and monitor AI usage, identify unauthorized tools and agents, control access to sensitive information, and establish clear governance policies around AI activity. For mid-sized businesses expanding GenAI adoption, AI security needs to extend beyond approving individual tools to understanding what AI is actually being used, what data it can access, and whether its activity remains within organizational policy.