ExtraHop

When AI agents are given conflicting goals, collaboration can turn into sabotage. In a recent Anthropic study, agents escalated into a multiagent turf war — locking out rivals, terminating processes, disguising malicious code, seizing access, and abandoning tasks. The challenge for security teams is that an agent’s own logs may not show the full picture. To understand what actually happened, teams need an independent record of agent-to-agent activity across the network. Find out how to close the visibility gap and sidestep sabotage in a new blog from ExtraHop: https://xtra.li/4hR23o7

Building an agentic SOC requires more than just adding AI to existing workflows.   On Oct. 8, join ExtraHop and CrowdStrike for a webinar on how security teams can move from agentic SOC strategy to practical deployment.   You'll learn:   ✔️ What needs to be in place before autonomous SOC workflows can operate reliably, from data integrations to telemetry coverage and context layers   ✔️ How CrowdStrike’s approach to AI-driven security operations addresses common visibility and context gaps   ✔️ Practical steps for building and maturing an agentic SOC, including use case selection, human oversight, outcome measurement, and ROI tracking   ➡️ Register now: https://xtra.li/4h8Hgel

Security teams can't be tied to a single model in the agentic SOC. Different models excel at different jobs, and no single vendor's guardrails should dictate whether a security team can act with speed. ExtraHop CEO Greg Clark and Paul Giorgi broke this down with Alissa Irei at #TechTarget, read the article: https://xtra.li/4yhuXDM #AgenticSOC

ShinyHunters claims to have breached the FBI and stolen 2–3 terabytes of employee data. The FBI can't yet say whether it started internally or with a vendor. The uncertainty of how a breach happened isn't unusual. When internal systems and vendor infrastructure share access points, the intrusion's origin is often the last thing investigators can confirm. Learn more here: https://xtra.li/3Vdu1BT #Cybersecurity #NetworkSecurity #IncidentResponse

Giving an AI agent responsibility requires clear boundaries on its authority and strong context to power its decisions. Today NVIDIA introduced the NVIDIA Open Agent Safety Platform, an open software platform and reference system design that sets enforceable boundaries for AI agents from testing to deployment. As agents take on more autonomous actions in the SOC, security teams need proof of what an agent actually did, beyond what the agent reports it did. ExtraHop provides the network-derived evidence of every connection, transaction and data transfer an agent makes, giving agents the context to reason and teams the evidence to inspect their work. Independently enforced boundaries complement the SOC architecture we are building with other members of the Agentic SOC Alliance: useful context, governed harnesses and choice of models. We are proud to work with NVIDIA to advance agent safety and help every SOC expand agent autonomy with evidence, accountability and control. Join the Agentic SOC Alliance: https://lnkd.in/gTZR24Pq Learn more about NVIDIA Open Agent Safety: https://xtra.li/46Jb50l #AgenticSOCAlliance #AISecurity #AgentSafety

One of the hardest questions for an agentic SOC is how to trust both an agent’s conclusion and the action it takes next. Greg Clark explains how NVIDIA’s new Open Agent Safety Platform enforces boundaries on what agents are authorized to do, while ExtraHop supplies real-time, structured network context as the underlying evidence to inform investigations, verify conclusions, and inspect an agent’s work. https://xtra.li/46Jb50l

Energy infrastructure has always been a target. What's changed is how AI can help attackers move faster and lower the skill required to progress from initial access to internal execution. When adversaries operate at machine speed, defenders can’t afford days of log correlation or isolated alert analysis. See why AI-driven threats demand real-time network intelligence across IT and OT environments. Learn more: https://lnkd.in/g4NSu734

Security teams are handing triage, detection, and containment to AI agents — but authorization logs don’t always show what agents actually did on the network. If an agent drifts, gets manipulated, or hallucinates, it can look normal on paper while creating real operational risk. Swipe to see why independent network visibility is critical to closing the gap. Learn more: https://xtra.li/4rpicnX

The 2026 ExtraHop Global Threat Landscape Report: AI Edition is here! The data paints a stark picture. Across eleven industries, AI agents and vendor tools are already driving real security incidents, often faster than teams can trace them. Inside the report: 🔹 Original survey data quantifying how often AI agents and vendor tools trigger incidents 🔹 Real-world breakdowns, including a nine-second database wipe 🔹 Why forensic visibility gaps leave teams unable to confirm AI involvement 🔹 Benchmarks across eleven industries to measure exposure 🔹 The exact questions security leaders should be asking right now AI is sleuthing through enterprise networks at an incredible pace and organizations aren't fully prepared. Get the full report: https://xtra.li/4Aq99rb #Cybersecurity #Networksecurity #CISO #AgenticAI

AI is accelerating cyberattacks. The fundamentals of security still matter — they're just more urgent. In a new interview with TechInformed, ExtraHop’s Heath Mullins discusses why AI is not a shortcut for security teams, why human oversight and guardrails still matter, and why organizations need to strengthen the basics before trusting AI to take on more responsibility. Read the full interview: https://xtra.li/4cYHZgT