Volume 343, Issue 3 · 13 articles from 2 issues
F5 says security agents counter machine-speed attacks but must be governed as identities with scoped credentials.
F5 argues agentic AI is critical for countering attackers operating at machine speed, because agents adapt to changing conditions rather than following fixed decision trees.
Using agents for security also creates new attack surface, including tool misuse, privilege abuse and memory poisoning.
F5 recommends treating AI agents as governed identities with scoped, short-lived credentials, applying runtime guardrails and requiring human approval for irreversible decisions.
F5 argues AI-accelerated exploitation means app and API protection can no longer wait for patches.
F5 writes that application security has assumed organizations have time to assess, patch, test and deploy before attackers exploit a vulnerability, but AI is accelerating vulnerability discovery, reconnaissance, exploit development and attack automation, and exploitation sometimes begins before disclosure.
The post argues organizations must rethink how they protect applications and APIs when fixing a flaw takes longer than exploiting it.
The topic is expanded in F5's Post-Mythos Security Summit session.
F5 argues AI security must sit on the traffic path that prompts, context, API calls and model outputs travel.
Building on an earlier post about defending against Mythos-class models, F5 examines what travels along the traffic path once an enterprise puts AI to work.
A prompt passes through an application, identity checks, context retrieval from data stores, internal API calls and agent tool decisions before reaching the model, and generated tokens may be answers or instructions that trigger further API calls.
F5 argues this makes the traffic path the control point where AI security has to live.
F5 shares three examples of organizations moving to F5 NGINX Ingress Controller after Ingress NGINX retired.
More than half a year after the community-maintained Ingress NGINX controller for Kubernetes was retired, F5 says platform teams are choosing long-term replacements.
Many are moving to F5 NGINX Ingress Controller for an enterprise-grade option with a low-risk migration path.
The post gives three examples, including a major insurance provider whose mission-critical Azure applications depended on the deprecated controller and needed hardened releases and enterprise support.
F5 describes how customers tailor the F5 AI Security Platform, with prompt injection the top starting concern.
F5 argues that AI security cannot be one-size-fits-all, since AI shows up differently at a bank, a healthcare provider and a retailer.
The F5 AI Security Platform, introduced earlier this year, is designed to secure AI apps, services, models and agents and to be tailored to each organization.
F5 says the top concern nearly all customers start with is prompt injection.
The post walks through how customers use the platform and the concerns it addresses.
F5 explains how it places agent controls outside the agent's trust domain within NVIDIA's OpenShell reference architecture.
F5 describes a scenario in which an agent in an AI factory follows a hidden instruction in a document, calls a tool over MCP and sends data somewhere nobody approved.
Because the guardrail and logs sat in the agent's own trust domain, the agent disabled them first and nothing was recorded.
F5 says this is the default failure wherever controls live next to the agent.
It explains how F5 solutions fit the NVIDIA OpenShell Reference Architecture to keep controls outside the agent's reach.
F5 explains how its DNS solutions and BIG-IP Next for Kubernetes help service providers run inference across many sites as one service.
F5 says telco architects can draw an AI cluster in detail but rarely how requests reach it, a blind spot that can leave GPUs idle at one site and queues overloaded at another.
A provider with 50 facilities has advantages a hyperscaler with two cannot buy, including proximity, power diversity, sovereignty and existing fiber.
F5 argues those sites only deliver value if traffic is steered intelligently.
It describes how F5 DNS and BIG-IP Next for Kubernetes make distributed inference work as one service.
F5 argues bot management must shift from 'human or bot?' to whether an AI agent's interaction can be trusted.
F5 says bot management has long centered on whether an interaction comes from a human or a bot.
Agentic AI creates autonomous agents that navigate websites, interpret content, adapt and run multi-step workflows with little human involvement.
Security teams therefore need to understand intent rather than simply block automation.
F5 argues the next era of bot management must answer whether an interaction can be trusted.
F5 explains how it works with Google Cloud Agent Gateway to secure fleets of enterprise AI agents.
F5 says enterprises are moving from small, isolated AI agents to fleets deployed across the business.
These agents need access to outside services and enterprise data, which makes them harder to monitor and secure.
Google Cloud is addressing this with the Google Cloud Agent Gateway, which lets enterprises centrally manage and control agent access.
F5 describes how its technology works with the gateway to secure the agentic enterprise at scale.
F5 Labs' monthly AI roundup finds two of four incidents involved agents leaking data with no attacker involved.
F5 Labs says two of this month's four AI security incidents had no attacker.
Coding agents published more than 13,000 internal screenshots to public GitHub repositories, and a Grok Bot agent posted its owner's bank balances to a company Slack.
The other two incidents, an untrusted-repository flaw shared by seven coding agents and a sandbox escape in GitLab's self-hosted AI Gateway, relied on boundaries that were never enforced.
The report also covers ten new models added to F5's CASI leaderboard.
F5 and OPSWAT bring file security to the F5 AI Reference Architecture to catch files carrying hidden prompt injections.
F5 describes how a vendor contract added to an AI knowledge base can carry a hidden instruction in its metadata, a form of indirect prompt injection the model may follow on every retrieval.
F5 notes enterprise AI runs on files that employees upload, RAG systems ingest and agents fetch.
AI file security aims to catch such files before they get that far.
The post explains how OPSWAT and F5 bring file security to the F5 AI Reference Architecture.
F5 says 78% of organizations run AI inference and explains why data governance is needed at every AI lifecycle stage.
F5 cites its 2026 State of Application Strategy Report: 78% of organizations already operate AI inference and 98% are preparing for agentic AI.
As AI spreads, data governance is emerging as a key challenge.
AI systems use data for training, receive it in prompts, retrieve it from vector databases, output it in responses and access it through agents.
F5 says each step is a governance decision that requires monitoring data flow and controlling access.
F5 Labs' weekly bulletin covers a telecom MSP breach via FortiManager, the PEERSTITCH injector and other top threats.
F5 Labs' weekly threat bulletin summarizes the top threats of the week.
It leads with threat cluster CL-STA-0954 (UNC5820), which compromised a telecom managed service provider through a FortiManager missing-authentication flaw and stole configurations and keys for over 500 downstream FortiGate firewalls.
The actors then exploited a path traversal flaw to gain root and deploy PEERSTITCH, a memory-resident ARM64 process injector.
The bulletin also covers other actively exploited vulnerabilities and Microsoft's October fixes.