Exabeam doubled its AI-focused detection coverage to 90 detections and added monitoring support for Anthropic's Claude, the company announced July 1, 2026, per SiliconANGLE. The update extends existing coverage for ChatGPT, Google Gemini, Microsoft Copilot, and GitHub Copilot, and maps detections to the OWASP Top 10 for Agentic AI inside its Outcomes Navigator tool. Exabeam also released Observra, an open-source telemetry library that normalizes AI agent activity and enriches it with cost, redaction, and risk signals for routing to security platforms, following its Praxen pre-deployment agent-verification tool released June 23. New Nova features include a natural-language Nova Rules Creator and early-access Nova Related Cases for grouping connected security incidents.
As AI agents increasingly act on systems with valid credentials at machine speed, the harder security problem is no longer detecting unauthorized access but distinguishing risky agent behavior from legitimate automation after the fact; Exabeam's push toward open telemetry standards and natural-language rule authoring is a direct response to that shift, not just a coverage-count milestone.
Exabeam said it doubled its AI-focused detection coverage to 90 and added monitoring support for Anthropic PBC's Claude, extending prior coverage for ChatGPT, Google Gemini, Microsoft Copilot, and GitHub Copilot that the company added in April, according to SiliconANGLE. The new detections target anomalous person-agent interactions and unauthorized autonomous activity, including suspicious prompt patterns, unusual tool-invocation sequences, abnormal consumption patterns, and "denial-of-wallet" indicators, per SiliconANGLE's report and Exabeam's own release.
SiliconANGLE reports Exabeam mapped its detections to the OWASP Top 10 for Agentic AI inside Outcomes Navigator, the tool security teams use to assess detection coverage gaps. The company also introduced Nova Rules Creator, which converts natural-language descriptions and existing Sigma rules into correlation analytics, and Nova Related Cases (early access), which groups related incidents and shared entities such as IP addresses or hosts. Exabeam released Observra, an open-source telemetry library that captures agent activity across major frameworks, normalizes it into events, and enriches it with cost, redaction, and risk signals before routing to security operations platforms; the library follows Praxen, a pre-deployment agent-verification tool Exabeam open-sourced on June 23. "Organizations are rapidly moving from AI experimentation to autonomous AI agents operating across the enterprise," Exabeam CEO Pete Harteveld said, according to SiliconANGLE. "Security teams need visibility not only into human activity, but into how agents behave, interact and make decisions."
inconsistent telemetry formats across agent frameworks, high false-positive rates from legitimate automated workflows, and slow rule-authoring cycles. Observability libraries like Observra and natural-language rule builders like Nova Rules Creator address the telemetry and authoring problems directly, though they do not eliminate the harder challenge of establishing behavioral baselines for agents that vary widely by task.
Whether other SIEM/SOAR vendors adopt Observra as a shared telemetry format, how Nova Rules Creator performs against noisy production agent traffic, and whether OWASP Top 10 for Agentic AI mappings become a standard checklist for agent-security posture assessments across the industry.
A substantive product update expanding agent-security detection coverage and introducing an open-source telemetry standard (Observra), which has practical value for SOC teams managing agentic AI risk. Adjusted slightly down from 6.5: this is a vendor product release corroborated by trade press (SiliconANGLE) and the company's own press release, without independent benchmarking of detection efficacy.
Public references used for this report.
A 5-minute Tuesday brief on AI & data science. Curated, no fluff.
No spam. Privacy.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
News on Let's Data Science is compiled from multiple public sources with editorial oversight. See our Editorial Standards and Corrections Policy.
Leave a Reply