Datadog launches AI agent security platform

Observability vendor Datadog has released new security capabilities targeting the emerging threat landscape created by artificial intelligence workloads.

The monitoring platform provider's latest tools address what it identifies as a fundamental shift in threat models.

Every microservice can now spawn autonomous agents capable of generating secrets, deploying code and calling external Application Programming Interfaces (APIs) without human oversight.

"AI has exponentially increased the ever-expanding backlog of security risks and vulnerabilities organisations deal with," Prashant Prahlad, vice president of products for security at Datadog, said.

The company's analysis suggests that AI-native applications operate non-deterministically, creating what Prahlad describes as "more of a black box" with expanded attack surfaces vulnerable to prompt and code injection.

Datadog Code Security is also now generally available.

It enables development and security teams to detect vulnerabilities in custom code and open-source libraries.

The tool uses AI to drive remediation of complex issues across both AI and traditional applications from development through production.

It prioritises risks based on runtime threat activity and business impact, with integrations for Integrated Development Environments (IDEs) and GitHub to enable vulnerability remediation without disrupting development pipelines.

Datadog Large Language Model (LLM) Observability, also generally available, monitors AI model integrity and performs toxicity checks across prompts and responses within organisational AI applications.

The platform's Cloud Security component detects and remediates risks including misconfigurations, unpatched vulnerabilities and unauthorised access to data, applications and infrastructure.

Sensitive Data Scanner prevents personally identifiable information from leaking into LLM training or inference datasets, with support for Amazon Web Services S3 and Relational Database Service instances now available in preview.

The company has introduced Bits AI Security Analyst, an AI agent integrated into Datadog Cloud Security Information and Event Management that autonomously triages security signals generated by AWS CloudTrail.

The agent performs investigations of potential threats and provides recommendations to help teams mitigate risks more quickly.

Datadog's Workload Protection continuously monitors interactions between LLMs and their host environments.

Meanwhile, new LLM Isolation capabilities, available in preview, detect and block vulnerability exploitation whilst enforcing guardrails to secure production AI models.