Datadog used its annual DASH conference in New York to ship AI Guard, a security product built specifically to detect and block attacks targeting autonomous AI agents. The tool combines deep agent telemetry tracing with stateful behavioral anomaly analysis, an approach Datadog argues catches multi-step agent poisoning that traditional prompt-and-response filters miss.
AI Guard launched as part of a 100+ feature rollout that collectively positions Datadog as the observability vendor most aggressively targeting the agent security surface.
The Agent Attack Problem AI Guard Addresses
AI agents now operate with elevated privileges, access sensitive data, and communicate with external systems. Tim Knudsen, Datadog’s VP of Security Products, laid out the threat model: “A single malicious prompt hidden in an innocuous-appearing prompt can turn a well-intended agent into a malicious actor leaking sensitive information, costing millions in reputational damage and data loss,” he told SecurityBrief.
The key technical claim is that attackers have learned to distribute malicious instructions across multiple steps in an agent’s execution chain, making them invisible to tools that evaluate prompts and responses in isolation. AI Guard traces the full behavioral sequence and applies anomaly detection across steps, not just at the prompt boundary.
Bits AI Goes Fully Autonomous
The other headline announcement is the expansion of Bits AI from a root-cause investigation tool into a fully autonomous operations agent. Bits AI can now scan infrastructure continuously, detect problems, recommend fixes, and in some cases resolve them within predefined guardrails, according to Techzine.
CTO Alexis Lê-Quôc described the shift: “With Bits Detection, Agent Evals, Infrastructure, Code, Release, Data Analysis, Testing and Chat, Bits AI is capable of truly autonomous operations, becoming a reliable teammate that operates across every stage of the production lifecycle and development loop.”
New capabilities include Agent Evals, which debugs other AI agents and generates fixes. Bits AI is also available through Slack and Claude, extending its reach beyond the Datadog dashboard.
Agent Console and Bits Agent Builder
Datadog also launched Agent Console, a centralized monitoring product for AI agents and coding tools including Claude Code, Cursor, and GitHub Copilot. The tool provides visibility into adoption rates, task performance, and cost correlation across agent usage.
Bits Agent Builder lets teams create custom AI agents within the Datadog platform for incident resolution, reporting, and standards enforcement.
Bring Your Own Cloud for Data Residency
Addressing the data cost problem created by AI-scale log volumes, Datadog introduced Bring Your Own Cloud (BYOC). The feature lets customers run the Datadog platform in their own cloud environment, keeping data processed and indexed in storage they control.
CEO Olivier Pomel framed the broader positioning: “The companies that win on AI won’t just build better models, they’ll build operational control around them,” he told SecurityBrief.
The First Mover in Agent Observability
Datadog is the first major observability vendor to ship a dedicated agent security product. AI Guard’s behavioral approach, tracing across multi-step agent chains rather than evaluating individual prompts, addresses an attack surface that existing security tools were not designed for. Whether competitors like New Relic, Elastic, and Splunk follow with their own agent-aware products in the next quarter will signal whether agent security has moved from niche concern to infrastructure requirement.