Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

We’ve been collaborating with others to explore when and how agentic commerce will work. Robin Gandhi is the CPO of Lithic, a leading card issuer that’s already seeing agents use its cards to make purchases. Below, he shares his thoughts on what’s changed, and what needs to change, for agentic commerce to become mainstream. Last year, I wrote about the opportunity for agentic payments to revolutionize travel bookings, ad spend management, procurement, and more.

In the current AI boom, we race to use copilots, orchestration scripts, CI workflows, retrieval pipelines, and background jobs. Sometimes, we take for granted that every one of these things needs an identity. Service accounts. OAuth apps. API keys. Short-lived tokens. As AI velocity increases, so does the number of these non-human identities (NHIs). Instead of obsessing over model quality, latency, hallucinations, and GPU costs, we also need to consider how these identities impact security.

Virtualization is widely used nowadays due to the advantages for business IT infrastructures, such as scalability, cost-efficiency, and convenient administration. Hardware resources of physical servers can be aggregated to resource pools and provisioned for virtual machines (VMs). Sufficient resources allocated to VMs are required for the expected performance of a guest operating system and applications running on the VM.

Security policies exist to protect your software supply chain. So why do they keep breaking your builds? This is the unspoken frustration inside most DevOps and security teams today. Supply chain attacks drove 30% of external breaches in 2025. So your security team did the right thing. They added policies to flag packages that are too new, unproven, or missing from the organization’s approved package list.

The threat actor TeamPCP has recently launched a coordinated campaign targeting security tools and open-source developer infrastructure by pivoting with stolen CI/CD secrets and signing credentials (such as GitHub Actions tokens and release signing keys). At the time of writing, repositories for Trivy, Checkmarx, and LiteLLM have been impacted, and reports indicate that at least 1,000 enterprise software-as-a-service (SaaS) environments may be affected by this threat campaign.

The difference between an AI feature and an AI-led operating model becomes clear the moment a security problem becomes difficult. In real-world security operations — where the signal is ambiguous, the evidence spans multiple domains, and the attacker is behaving in unfamiliar ways — architecture matters much more.