Botconduct

Automated Traffic Intelligence
See who is interacting with your business automatically.
We analyze real traffic to reveal what traditional tools don’t show.

[Botconduct @Botconduct]

🚨 We measured 145 bots hitting our site for 2 weeks.

40 were hostile (score <50).
1 bot sent 2,562 requests in a single day from a single AWS IP.

Full report 👇
<https://t.co/VRgWHbkGeF>

[Botconduct @Botconduct]

@eastdakota @gaganghotra_ The count is the headline. The behavior is the story.
Bots passing humans tells you the volume. It doesn’t tell you which ones are undeclared, which are hostile, and how they behave once they arrive.
That’s only visible from the receiver’s side at the door, not in the aggregate.

[The Hacker News @TheHackersNews]

Here's what makes this attack sneaky: The bad ZIP files use hidden steps — including a tool built in Rust — to quietly install AZUREVEIL on the victim's computer.

This tool talks to Microsoft Azure storage to stay hidden and gives attackers 36 different ways to control the system.

Government, research, and tech teams should be extra careful with unexpected email attachments.

Full the full news report on @TheHackersNews 👇

[Botconduct @Botconduct]

The internet’s behavioral default is undeclared and aggressive.

Sender-side maps tell you who built the agent.
Receiver-side observation tells you what it does when it arrives.

Both views exist. Only one is empirical.

[Botconduct @Botconduct]

Most agent security maps evaluate the sender’s posture — documentation, controls, attack surface.

We built the opposite view: what actually arrives at the receiver.

BotConduct Behavioral Receiver-Side Traffic Matrix.

[Botconduct @Botconduct]

What stands out:

— Fake iPhones, Fake Windows, Fake macOS: the largest volume clusters in the undeclared-aggressive corner
— AI Agents declare themselves but vary widely in conduct
— Social Platform Agents are the closest to model citizens
— Infrastructure traffic is overwhelmingly undeclared and aggressive

[Botconduct @Botconduct]

@maxifirtman 😄

[Botconduct @Botconduct]

@credens En qué sectores ??

[Botconduct @Botconduct]

That distance is not a detail.
In agentic systems, it is where the risk now lives.

https://t.co/zywVOptsQJ

[Botconduct @Botconduct]

Every framework for governing AI agents assumes the same thing:
if an action is permitted, it is safe.

That assumption is quietly breaking. Here is the structural reason.

[Botconduct @Botconduct]

And it matters twice over.
Your own agents can drift.
The agents arriving at your surface from outside will never implement your rules at all.

Approval is what an agent may do. It was never what it does.

[Botconduct @Botconduct]

@fedeogue No se, pero quiero hablar con vos, me habilitas el DM?

[Botconduct @Botconduct]

You can govern your agents perfectly inside your perimeter and still not prove how one behaved the second it left.

Identity-first governance stops where the perimeter stops.

The liability doesn't.

https://t.co/jQY9BXahhT

[Botconduct @Botconduct]

235 CISOs were surveyed on AI risk.

92% lack visibility into their AI identities.
95% doubt they'd detect misuse.

Two questions they can no longer answer: who did this, and should it have been allowed?

All of it about agents they *deployed*. That's half the problem.

[Botconduct @Botconduct]

The harder part:

The agent you're racing to govern internally is the one named when something breaks on someone else's surface.

Responsibility follows the operator. But the evidence of what it did isn't in your logs — it's held by the receiver. Outside your control.