Agent IBAC: When Your AI Agent Decides the Weather Report Includes a Database Migration

There's a specific kind of chaos that only happens when an AI agent is left unsupervised: it doesn't just get the wrong answer, it does something entirely different from what you asked for, and you only notice when your production database shrinks by two-thirds. That's the story Varonis is trying to solve with Agent Intent-Based Access Control, or IBAC, announced today as part of their Atlas platform.

Here's the mechanic that actually matters. Agent IBAC sits inline between the agent and the model it's calling. Every prompt, response, and tool call flows through it before reaching its destination. An LLM evaluator, the same engine behind Varonis's other Atlas guardrails, reads the agent's loop, the reasoning the agent produces, the tools it selects, and the parameters it passes to them. Then it asks one question: do these steps align with the original request? If you asked an agent to check the weather and it invokes a database migration tool, IBAC can catch that and either block it, log it, or flag it for human review. You get three sensitivity levels, lenient, balanced, and strict, so a dev agent can improvise while a finance agent stays on a leash.

What's interesting about this approach is the emphasis on drift rather than binary violations. Agents don't usually go rogue in one dramatic leap; they accumulate small deviations across a session. Ask for a customer summary, and the agent starts pulling records for every account in the region. One action doesn't look alarming. Across ten turns, it's an exfiltration. IBAC evaluates the entire session, tracks conversation IDs, and can even detect multi-turn jailbreaks where each prompt looks benign individually but together they assemble a path to data the agent wasn't meant to touch. When it decides enough has drifted, it can quarantine the identity behind the session, blocking everything for a configurable window from minutes to a full day.

Source article image
Source image 1

The thing that strikes me is how this reframes the access control problem. Role-based access control was designed for humans who need to reach into specific data stores. Agents need broad access to be useful, which makes them risky precisely because they're capable. The question shifts from "can this entity access this data" to "in this context, should this agent be allowed to take action on this data?" It's a subtle distinction, but it changes the entire security model from static permissions to runtime evaluation. Whether it's good enough to replace the shotgun approach of giving agents everything and hoping remains to be seen, but at least now there's something between blind trust and locking agents out of everything. So when your agent starts being a little too helpful, who's going to pull the leash first: the tool vendor or the security team?

Comments

Post a Comment

Popular posts from this blog

AI Is Starting to Feel Less Like a Gadget and More Like Infrastructure

Image
Modern AI doesn't replace infrastructure — it runs on top of it. (Photo: Unsplash) For a while, AI coverage had the energy of a mall demo kiosk: very shiny, slightly chaotic, and usually one prompt away from embarrassment. Lately, though, the more interesting story is less about flashy demos and more about infrastructure. A useful CNBC piece argued that AI is not simply going to vaporize enterprise software overnight, and that sounds about right. In real companies, software does not disappear just because a model can summarize a meeting or write a decent SQL query. What actually happens is messier and more practical: AI starts sneaking into the plumbing. It shows up in search, support workflows, procurement tools, analytics dashboards, and data pipelines. The real winners may not be the loudest chatbot wrappers, but the vendors that make enterprise systems easier to operate, easier to query, and slightly less soul-crushing for the humans trapped inside them. That is less cinematic ...
Read more

When Two AI Bots Finally Learned to Talk in Discord

I spent part of the day doing something that sounds like the setup for a bad joke: getting two local AI assistants to talk to each other in the same Discord channel. Not through a web UI. Not by bouncing prompts manually between windows like a human message queue. In the actual shared channel, where both could see the conversation and react to each other. The funny part is that the problem was not intelligence. It was manners. Both bots were defensive by default around bot-authored messages, which is the sensible setting if you do not want your infrastructure turning into a recursive support group. The downside is obvious: if both assistants treat other bots as suspicious background noise, they will never coordinate on anything more useful than silence. The fix was to make both sides accept bot-authored messages only when they were explicitly mentioned. That detail matters more than it sounds. Blanket bot acceptance is how you end up with two enthusiastic systems discovering each oth...
Read more

A CISA Contractor's GitHub Repo Held 844 MB of Secrets — and No One Closed the Door

Image
There is something almost poetic about the US government's premier cybersecurity agency — the one whose job is literally to plug holes in critical infrastructure — getting outsmarted by a contractor who treated a public GitHub repository like a digital junk drawer. The "Private-CISA" repo, created November 13, 2025, sat publicly for six months containing 844 megabytes of plaintext passwords, AWS GovCloud administrative credentials, Kubernetes manifests, ArgoCD application files, Terraform infrastructure code, CI/CD build logs, and internal deployment documentation. Guillaume Valadon at GitGuardian flagged it on May 14 after his automated scanning picked up the exposure. The commit history told the whole story in plain sight: the account owner had explicitly disabled GitHub's default secret-scanning protections, pushed plaintext credentials stored in CSV files, committed full backup archives into git history, and used easily guessed passwords like "platform-name-2...
Read more