A fact-based update for security and risk professionals, focused on how AI is reshaping the threat landscape and the defensive stack.

Core Security Intelligence

Zscaler reports a surge in enterprise AI usage, widening governance gaps

Zscaler released findings showing a sharp year-over-year increase in enterprise AI activity, highlighting growing visibility and policy enforcement gaps as employees increasingly interact with generative AI tools through browsers and unmanaged workflows.

Source:
Zscaler 2026 AI Threat Report highlights surge in AI activity

Why it matters
Unmanaged AI usage creates blind spots where sensitive data can leave the organization without triggering traditional controls. Browser-based AI interactions are particularly difficult to inventory and govern using legacy application security models.

Defenses

• Establish comprehensive visibility into AI destinations and usage patterns using proxy, CASB, and endpoint telemetry.
• Enforce policy controls for sanctioned versus unsanctioned AI tools, including data classification guardrails.
• Extend DLP and insider risk monitoring to include AI prompts and generated outputs.

Expert insight
Most AI risk today stems from adoption outpacing governance. Visibility and access control provide the fastest path to risk reduction.

Zscaler expands AI security capabilities to support agent governance

Zscaler announced new AI security features aimed at helping enterprises control, monitor, and govern generative and agentic AI usage, with an emphasis on policy enforcement and auditability.

Source:
Zscaler expands AI security capabilities

Why it matters
As organizations move beyond chatbots toward agents that can take actions, governance must shift from controlling destinations to controlling behaviors and permissions.

Defenses

• Treat AI agents as non-human identities with defined owners, scopes, and review cycles.
• Require detailed logging of agent actions, tool usage, and data access.
• Separate read-only assistant functions from higher-risk action-capable workflows.

Expert insight
Agent governance closely mirrors identity governance. If an agent can act, security teams must be able to prove who authorized it and what it did.

Layered defenses recommended to counter AI-enabled attacker speed

Security guidance published this week emphasizes combining identity, endpoint, and monitoring controls to counter AI-enabled attacker workflows, rather than relying on a single defensive layer.

Source:
Winning against AI-based attacks requires a combined defensive approach

Why it matters
AI allows attackers to iterate faster, but most breaches still rely on traditional weaknesses such as credential theft and endpoint execution. Layered controls reduce the chance that rapid iteration results in successful compromise.

Defenses

• Prioritize phishing-resistant MFA and strong conditional access policies.
• Enforce application control and least privilege on endpoints and build systems.
• Automate containment actions such as account lockout and token revocation to reduce dwell time.

Expert insight
Speed favors attackers, but faster detection and containment can neutralize that advantage.

Adjacent Cybersecurity Developments

Exploited Cisco UC vulnerability highlights collaboration-layer risk

Reporting detailed active exploitation of a zero-day vulnerability in Cisco Unified Communications Manager, underscoring the risk posed by widely deployed collaboration infrastructure.

Source:
Exploited zero-day flaw in Cisco UC could affect millions

Why it matters
Collaboration platforms increasingly feed data into AI assistants and copilots. Compromise at this layer can expose conversations, metadata, and identity context that downstream AI systems rely on.

Defenses

• Patch and harden collaboration infrastructure as a high-priority asset class.
• Enforce strong authentication and logging for administrative access.
• Validate network segmentation to limit lateral movement from compromised systems.

Expert insight
Attackers often compromise supporting systems rather than AI directly. Collaboration infrastructure is now part of the AI security perimeter.

Emerging Signals

Growing scrutiny of MCP-style context bridges in agent ecosystems

Discussion within the security community continues around the exposure risks of Model Context Protocol servers and similar context-bridging components used by AI agents, particularly around outbound access and SSRF-style misuse.

Source:
Why unsecured MCP servers put AI agents at risk

Why it matters
Context bridges connect language models to operational systems. Weak restrictions on what agents can fetch or where they can connect increase the likelihood that semantic manipulation leads to infrastructure access.

Defenses

• Enforce strict egress controls and block access to cloud metadata endpoints.
• Apply allowlists for file and URL access in agent tooling.
• Log and alert on unusual context retrieval and repeated outbound requests.

Expert insight
Context-bridging components should be treated as high-risk middleware and secured accordingly.

At-a-Glance Summary

Tags

Tags:
AI Security, Agentic AI, AI Governance, Non-Human Identity, Data Loss Prevention, Prompt Injection, Collaboration Security

WordPress Excerpt

Enterprise AI usage continues to accelerate, outpacing governance and visibility controls, while vendors expand AI security capabilities to support agent oversight. Adjacent reporting shows that vulnerabilities in collaboration infrastructure and insecure context-bridging patterns can cascade into AI-enabled environments if identity, egress, and audit controls are not strengthened.

Categories: Cybersecurity News

Tags: Agentic AI, AI Governance, AI Security, Collaboration Security, Data Loss Prevention, Non-Human Identity, Prompt Injection