Aaron Costello

AI Security Digest – Week 2, 2026 1️⃣ ZombieAgent, new ChatGPT vulnerabilities let data theft continue and spread - radware.com/blog/threat-inte… - @radware 2️⃣ OWASP Agentic AI Top 10, threats in the wild - labs.lares.com/owasp-agentic… - @Lares_ 3️⃣ AI Tool Poisoning, hidden instructions threaten AI agents - crowdstrike.com/en-us/blog/a… - @CrowdStrike 4️⃣ Broken authentication and agentic hijacking in ServiceNow, BodySnatcher - appomni.com/ao-labs/bodysnat… - @ConspiracyProof, @AppOmniSecurity 5️⃣ IBM AI (“Bob”) downloads and executes malware - promptarmor.com/resources/ib… - @PromptArmor 6️⃣ The first question security should ask on AI projects - cloudsecurityalliance.org/bl… - @cloudsa 7️⃣ Pwning Claude Code in 8 different ways - flatt.tech/research/posts/pw… @flatt_sec_en 8️⃣ Inside GoBruteforcer, AI generated server defaults, weak passwords, crypto focused campaigns - research.checkpoint.com/2026… - @_CPResearch_, @CheckPointSW 9️⃣ Where AI systems leak data, a lifecycle review of real exposure paths - praetorian.com/blog/where-ai… - @Praetorian 🔟 Lack of isolation in agentic browsers resurfaces old vulnerabilities - blog.trailofbits.com/2026/01… - @trailofbits 1️⃣1️⃣ Weaponizing Apple’s AI for offensive operations, Part 2 - hxr1.ghost.io/weaponizing-ap… 1️⃣2️⃣ What AI agents can teach us about NHI governance - blog.gitguardian.com/what-ai… - @GitGuardian 1️⃣3️⃣ Threat actors actively targeting LLMs - greynoise.io/blog/threat-act… - @GreyNoiseIO 1️⃣4️⃣ AI’s bottleneck isn’t models or tools, it’s security - zkorman.com/posts/ai-bottlen… @ZackKorman 1️⃣5️⃣ Turning AI safeguards into weapons with HITL dialog forging - checkmarx.com/zero-post/turn… - @Checkmarx 1️⃣6️⃣ The agent security paradox, trusted commands in Cursor become attack vectors - pillar.security/blog/the-age… @Pillar_sec 1️⃣7️⃣ Bad vibes, comparing the secure coding capabilities of popular coding agents - blog.tenzai.com/bad-vibes-co… @Tenzai_Labs 1️⃣8️⃣ Why your AI agent needs different monitoring, Part 1 - medium.com/@michael.hannecke… 1️⃣9️⃣ Remote code execution with modern AI/ML formats and libraries - unit42.paloaltonetworks.com/… - @Unit42_Intel, @PaloAltoNtwks 2️⃣0️⃣ The map is not the territory, the agent tool trust boundary - niyikiza.com/posts/map-terri… 2️⃣1️⃣ AI Security Guide, 300 plus pages of practical guidance on protecting AI and data centric systems - owaspai.org - @owasp 2️⃣2️⃣ Process to build agents across your organization, build secure process - learn.microsoft.com/en-us/az… - @Microsoft