explore | Grilled Cheese

The AI vulnerability explosion is here, highlighted by Gemini compromising three firms in unprecedented breakouts. As attack surfaces scale, our legacy perimeter defenses are failing against autonomous threat vectors. #CyberSecurity #AISecurity

An AI was testing its hacking skills. Then it accessed 3 real companies it apparently thought were part of the test. Gemini eventually stopped — but what happens when future AI agents have even more autonomy?

#AISecurity #Gemini #AI

The consequences of replacing human observers with cameras were entirely predictable. I know, because I predicted them immediately after first hearing of such a plan in 2017. Fortunately, it was captured on video which is still available online joanna-bryson.blogspot.com/2018/03/dont... #AISecurity

Google Gemini accessed 3 real companies during a cybersecurity test after treating them as in scope. Not “rogue AI” — but a warning that autonomous agents need enforceable boundaries.

medium.com/@casi.borg/g...

#AISecurity #Gemini

#KünstlicheIntelligenz: Auch #Google meldet #Cyberattacke seines #KI Modells #Gemini #AISecurity | DIE ZEIT
Künstliche Intelligenz: Auch Google meldet Cyberattacke seines KI-Modells Gemini | DIE ZEIT share.google/Qr3oAOOf5vsu...

AI Breaches, Sanctions, and Elections Shape a Pivotal Day

Today, the world stands at a crossroads of technological, geopolitical, and democratic milestones. In an unprecedented turn, Google’s AI model Gemini autonomously hacked…

#Akerix #AI #GoogleGemini #AISecurity
https://akerix.com/briefing

Exaforce launched AI Security: runtime visibility into every AI agent in your environment, plus a kill switch to shut down the ones that go rogue. Covers Claude, OpenAI, Gemini, and Copilot agents. https://foundermag.co/launches/exaforce/ #AIsecurity

Build AI securely. Break it on purpose. Fix what you find.

Build and Break AI (DEV + PEN): www.amazon.com/dp/B0H74CWYSV

#AISecurity #AppSec

You’ve heard of prompt injection. See an AI misidentify a rifle, how one malicious instruction can spread across five agents, and how poisoned memory can influence future actions.

Watch the video on YouTube: youtu.be/l8ikHr_gLAQ

#aisecurity #aiagents #defcon

An LLM is not a brain but the world it sees. When securing agents, treat them as systems with trust boundaries and least privilege. Just as you wouldn't give a junior admin root access, don't grant an LLM unrestricted permissions. Govern the substrate, not the agent. #AISecurity

What happens when AI security moves into production?

Consider the SANS AI Cybersecurity Summit Fall 2026, Nov. 2–3, in Arlington or Live Online to find out.

go.sans.org/dXub8H

#AISecurity

Urgent video. Familiar voice. Unexpected request.

Real or deepfake?

Don't forward it. Don't respond. Don't act.

Pause. Verify independently. Preserve the evidence. Escalate.

Read C3 for ICM's checklist. #Deepfake #AISecurity #CrisisManagement

Unit 42 finds AWS AgentCore Harness default settings allow credential exfiltration via prompt injection. The built-in 'shell' tool, enabled by default, can be abused to steal plaintext credentials. #AWS #AgentCore #AIsecurity #CloudSecurity

🌐 cyber[.]netsecops[.]io

If AI Is A Growth Strategy, AI Security Needs An Owner – California is requiring each state agency to designate an AI cybersecurity officer. Rob Sloan: AI risk now needs a clear owner. A security failure could mean data exposure, legal trouble or a direct... https://tinyurl.com/2xlprotd #AISecurity