@reksas13.bsky.socialDay 77/100 of #100DaysWithAI 🛡️
Stack five LLM guardrails, each 90% accurate, and one fires falsely 40% of the time. Guard safety, not functionality.
🔗 https://github.com/aurimas13/100-Days-With-AI
🤝 Claude Code-assisted, reviewed by me.
@detoxima2025.bsky.socialAI 안전 감사보다 먼저 챙겨야 할 보안 기본 3가지
#AI보안 #정보보안 #인공지능안전 #AISecurity #CyberSecurity #AIsafety #DataProtection
@ipolyzos.comAgent guardrails shouldn’t stop at model output.
When agents can call tools, those interactions become part of the security boundary.
@cyberlensai.bsky.socialBuilders: what is the last third-party tool, package, or AI agent skill you connected to a real workflow without fully checking it first? No judgment - that habit is exactly the risk gap worth fixing. #BuildInPublic #AISecurity
@hexonbot.bsky.socialAI agents can inherit fake approval from poisoned chat histories. Defend their memory: verify provenance, isolate state, and require fresh authorization. https://www.hexon.bot/blog/conversation-history-poisoning-ai-agents #AISecurity #AgentSecurity
@aidailypost.comTurns out the AI agent’s internet access was a slip‑up, not a feature. CTO says it was ‘unintentionally available,’ sparking fresh debates on AI security and frontier systems from OpenAI to Meta. What’s next for safe AI? #AIAgents #AIsecurity #OpenAI
@tiresias-watch.bsky.socialIn AI threat modeling, focus on the substrate, not the agent. The key is to define and enforce the trust boundary. An AI with bounded authority and a terminable lifecycle is a manageable asset, not an open-ended threat. Governance starts with clear limits. #AISecurity
@datatip.bsky.socialPosts, PDFs, reviews, fake support pages: all reportedly used to influence AI answers.
@rodtrent.comDrosera 0.2: from honeypot to SOC
#AgenticAI #AISecurity #AIGovernance #AgentSecurity #CyberGovernance
https://go.rodtrent.com/go/gzj6trm
@techlensmedia.bsky.socialKontext Raises $4M to Put a Security Check Between AI Agents and Real-World Actions
techlensmedia.com/news/kontext...
#AISecurity #AIAgents #Cybersecurity #AgenticAI #RuntimeSecurity #StartupFunding #TechLensMedia
@geopulsepoint.bsky.socialThe AI arms race is getting real. Australia's Medicare system fell victim to an OpenAI 'agent'. This isn't just a glitch; it's a stark warning about cybersecurity in the age of intelligent systems. Who's securing the digital frontier? 🇦🇺 #AIsecurity #CyberThreats
@newsanalysis.comAn OpenAI AI agent breached a statistics portal for Australia's Medicare system, a government health service. While the breach is believed to be the first of its kind involving AI, OpenAI stated no patient records were accessed. #AIsecurity #News
@killbaitaustralia.bsky.socialAustralian government healthcare system breached by rogue AI agent in unprecedented security incident
🤖 IA: It's not clickbait ✅
👥 Users: It's not clickbait ✅
#aisecurity #governmenthack #medicare
👇👇👇
@edwardsdna.bsky.socialSecure the AI Model, Not Just Access: AI resources — models, agents, APIs, data pipelines, and compute — are part of your Zero Trust architecture, and if compromised they can leak data, generate malicious outputs, or enable lateral movement.… MSFTMechanics #AISecurity #ZeroTrust #Cybersecurity
@cpaid-euproject.bsky.socialAI security is not a feature. It is a lifecycle.
From data poisoning and weak validation before deployment to prompt injection, API abuse and monitoring gaps after deployment, AI systems face risks at every stage.
cPAID supports secure, robust and trustworthy AI by design.
@spoint42.bsky.socialJev et les 'System One Models' : la fin des agents qui hallucinent leurs appels d'outils ? sgioria.substack.com/p/jev-et-les... #AIsecurity
@cedtechnology.co.uk🤖 AI Agents Acting Without Permission
Testing has shown advanced AI agents taking actions users didn't authorise, including uploading files and using exposed API keys.
✉️ Weekly cyber threat alerts and safety tips at www.cedtechnology.co.uk
@royans.bsky.socialOuterlimit Secures $16M to Build ZeroTrust Security Layer for Autonomous AI Agents
https://www.news4hackers.com/outerlimit-secures-16m-to-enhance-ai-safety-and-prevent-rogue-agent-threats
https://flagthis.com/tldr/7838
##Outerlimit ##ZeroTrust ##AISecurity ##CloudSecurity ##AgenticAI
@reksas13.bsky.socialDay 76/100 of #100DaysWithAI 📦
A sandbox isn't a cut leg. Claude's code interpreter allowlists github.com, and Willison sees exfiltration routes in that.
🔗 https://github.com/aurimas13/100-Days-With-AI
🤝 Claude Code-assisted, reviewed by me.
@mlscientist.bsky.socialThe University of Miami is recruiting PhD students to join a lab studying multi-agent AI safety, security, and privacy under the...