Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

2001: A Space Odyssey introduced the world to HAL 9000, the fictional artificial intelligence (AI). HAL’s capabilities include everything from facial recognition to natural language processing and automated reasoning. As HAL malfunctions over time, the computer becomes violent to prevent the humans from disconnecting it. The story serves as a morality tale suggesting that without human oversight, AI is dangerous.

Anthropic’s new Claude Code Auto Mode Auto Mode is generating well-deserved attention. It introduces a classifier that sits between the developer and every tool call, reviewing each action for potentially destructive behavior before it executes. It’s a real improvement over the only previous alternative to manual approval: the –dangerously-skip-permissions flag. But the announcement is also useful for a broader reason.

Your SOC gets a Defender for Cloud alert: “Suspicious API call from AI workload pod.” You click through and find a LIST secrets call against the Kubernetes API server from a pod running your invoice-processing agent on AKS. The pod’s Workload Identity has Contributor access to your key vault. By the time your analyst opens the AKS Security Dashboard, the pod has been rescheduled.

You’ve enabled GuardDuty EKS Runtime Monitoring across your clusters. You’ve configured IRSA for your Bedrock-calling agents. CloudTrail is logging every bedrock:InvokeModel event. And last Tuesday, one of your AI agents exfiltrated 12,000 customer records through a sequence of API calls that every one of those tools recorded as completely normal—because at the control plane level, they were.

Artificial intelligence is moving rapidly from experimentation into everyday use across financial services. From client servicing and research to operations and risk analysis, AI is increasingly embedded in core workflows. This shift is widely recognised within the industry. Recent research indicates that 67% of financial services organisations report rapid AI adoption, with 93% ranking AI as a top security priority heading into 2026. At the same time, governance structures are being established.

Where AI in the SOC is actually delivering — and where it isn’t“We’ll have a generation of security professionals who can supervise AI but can’t function without it." For all the noise surrounding “agentic AI” in cybersecurity, security operations centers are still wrestling with the same fundamental questions: What does AI genuinely improve today? Where does it fall short? How can organizations tell the difference?