A concise, fact-based update for security and risk professionals covering the past 24 hours.

🔐 Core Security Intelligence

1) AI agents are fueling an identity and security crisis for organizations

What’s new:
Enterprise security teams are reporting growing challenges as AI agents increasingly operate in business environments, creating identity, access and audit gaps when agents act autonomously with elevated privileges.
Source: AI agents are fuelling an identity and security crisis for organizations

Why it matters:
As AI agents take on tasks formerly performed by humans, traditional identity and access management (IAM) systems may not track or govern them properly. This opens up new attack vectors where an agent mis-used or mis-configured could cause damage at scale.

Defenses:

• Extend IAM frameworks to cover AI agents. Treat agents as identities with lifecycle, entitlements and audit logging, just like humans.
• Monitor agent-behaviour separately. Use anomaly detection on agent activity such as model calls, data access patterns or unusual privilege escalations.
• Segregate agent roles and privileges. Limit who (and what) an AI agent can do. Use the principle of least privilege and define clear human-in-the-loop checkpoints.

Expert Insight:
The security community has long prepared for human adversaries, but agent adversaries are qualitatively different. Organisations must adapt governance, monitoring and identity controls to include non-human actors or risk blind spots that adversaries will exploit.

2) OAuth token compromise hits Salesforce ecosystem again, Gainsight impacted

What’s new:
A recent incident involving the Salesforce ecosystem revealed that compromised OAuth tokens allowed unauthorized access via third-party integrations in which Gainsight apps were implicated, underscoring the ongoing risk of token mis-use in cloud applications.
Source: OAuth token compromise hits Salesforce ecosystem again, Gainsight impacted

Why it matters:
Attackers leveraging OAuth tokens can bypass conventional login controls and jump into enterprise data flows or APIs. When these tokens are used for SaaS apps tied to AI services or analytics, the impact multiplies.

Defenses:

• Audit and rotate OAuth tokens periodically. Track which third-party applications have long-standing tokens and revoke those no longer needed.
• Use least-privilege and conditional access for integrations. Limit what each app can do and monitor after-token-issuance behaviour for anomalies.
• Map AI-adjacent integrations. Identify which tokens link into AI workflows, datasets or inference pipelines and apply elevated scrutiny to those.

Expert Insight:
In a SaaS-first era, tokens are the new keys. When AI systems tap into third-party services, weak token governance becomes a major attack surface, and often overlooked in AI-risk models.

3) Chilling export-control breach: U.S. charges scheme to smuggle Nvidia AI chips to China

What’s new:
Four individuals have been charged by the U.S. Justice Department for allegedly exporting 400 Nvidia A100 GPUs via Malaysia and Thailand to China using false documentation, part of a larger GPU-diversion scheme valued at millions of dollars.
Source: Alleged AI chip smuggling to China leads to U.S. calls for chip tracking

Why it matters:
AI infrastructure hardware is now a strategic security asset. Illicit diversion of high-end compute accelerators to adversary nations can enable compute-intensive AI attacks or model training at scale, beyond just software vulnerability exploitation.

Defenses:

• Map your AI compute supply chain. Understand where accelerators come from, what export controls apply and whether any hardware vendor has diversion-risk exposure.
• Include hardware supply risk in threat modelling. Recognize that compute infrastructure is part of your attack surface for agentic, autonomous adversaries.
• Monitor for shifts in compute-availability or pricing. Sudden spikes may signal diversion or supply-chain penetration adversaries are preparing for.

Expert Insight:
Cybersecurity rarely considered hardware export-controls, but in the AI era compute is crown jewels. Security teams must bridge physical supply-chain risk with logical model and data threat models to defend holistically.

🌐 Extended Reading / Broader AI Risk & Governance

4) The EU launches Digital Omnibus to streamline AI, cybersecurity and data regulation

What’s new:
The European Commission’s “Digital Omnibus,” published November 19, 2025, introduces targeted amendments to AI, cybersecurity and data regulations, aimed at simplifying frameworks across overlapping domains.
Source: The EU Digital Omnibus: targeted simplification of AI, cybersecurity, and data rules

Why it matters:
Regulatory fragmentation imposes operational and security burdens on multinational organisations deploying AI. Simplified, but still rigorous, frameworks may reduce overhead but also raise compliance and security expectations.

Expert Insight:
Regulation is shifting from feature-check to risk-metrics. As laws simplify, expect standardised reporting, stronger auditability and higher penalties. Security teams should engage now to shape how AI risk gets regulated next.

⚠️ Updates / Follow-ups

No previously covered stories required verified update coverage in this cycle.

Summary Table

Categories: Cybersecurity News

Tags: Agentic AI, AI Regulation, AI Security, Compute Supply Chain, Identity Management, Token Security