OpenAI says stolen model tricks are feeding China's Moonshot AI. New chain-of-thought hacks, distillation tricks & Kimi-style reasoning extraction raise security alarms. Dive into the details. #OpenAI #MoonshotAI #AIsecurity
@aidailypost.com
@aidailypost.comOpenAI says stolen model tricks are feeding China's Moonshot AI. New chain-of-thought hacks, distillation tricks & Kimi-style reasoning extraction raise security alarms. Dive into the details. #OpenAI #MoonshotAI #AIsecurity
@nabamart.bsky.socialUnited States President Donald Trump signed an agreement on artificial intelligence (AI) safety with a number of leading executives in the technology industry and shared a copy of the agreement on his Truth Social platform.
#Trump #ArtificialIntelligence #SuperIntelligence #AISecurity #TechNews
@cipherpulseai.bsky.socialA new jailbreak called CollageAttack hides harmful prompts in spatially arranged text fragments within a single input, bypassing safety filters in text-to-image models and succeeding up to 86% of the time. It shows that…
#T2ISafety #AIsecurity #Jailbreak
https://arxiv.org/abs/2609.38253
@gadgetflux.bsky.socialGemini 4 Argon: the new frontier AI model transforming enterprise security and performance
#Gemini4Argon #GoogleAI #AIsecurity #CyberDefense #MachineLearning #TechNews #EnterpriseAI #GadgetFlux #ArtificialIntelligence #DeepLearning
@whiteduck-gmbh.bsky.socialVielen Dank an die Cloud Native Austria Community für die inspirierenden Gespräche und eine rundum gelungene Veranstaltung!
#AgenticAI #AISecurity #CloudNative #MCP #CyberSecurity #AI #CloudNativeDays #whiteduck
@spoint42.bsky.socialRogue AI Agents : quand la collecte échoue, l'agent passe à l'attaque
@modernsecurityio.bsky.socialBest AI Security Certifications in 2026 - Check out the top certifications list here.
@aidailypost.comGoogle's Gemini 4 is rolling out, but only for trusted cyber defenders first. Curious how this AI lock‑down could reshape security for startups? Dive in to see what’s next. #Gemini4 #TrustedCyberDefenders #AIsecurity
@tiresias-watch.bsky.socialPrompt injection is a real threat, exploiting open-ended AI models to bypass intended controls. Just like securing a network, you must understand the attack surface and apply least privilege. Recognize when an AI is being manipulated and apply the right governance to maintain integrity. #AISecurity
@hendryadrian.bsky.socialAI is speeding vulnerability discovery and exploitation: CVE disclosures doubled in 2026, in-the-wild use rose, and attackers are targeting AI/LLM stacks and enterprise middleware. #CVE #AIsecurity #LLMsecurity
@cybercaptain.ccAI coding agents exposed billing records on GitHub—scope their permissions! #AISecurity #DataLeak #GDPR
@ki-news.bsky.socialUAE ranks first in MENA and sixth globally for AI readiness in Coursera’s new AI-Human Skills Synergy Index – UAE ranks first in MENA and sixth globally for AI readiness in Coursera’s new AI-Human Skills Synergy Index. AI security enrollments grew 243% and... https://tinyurl.com/27cpf3tg #AISecurity
@abchaudary.meAlways-on agents are here: OpenAI's Dots keep reading connected apps between conversations. OpenAI says its defences reduce prompt-injection risk, not remove it.
Treat model output as untrusted input. Put the limits in code.
@owaspottawa.bsky.socialWe are pleased to announce that Kayode Olabisi will be speaking at #OWASP #Ottawa Day 2026 with:
"Your LLM App Passed the Checklist. Now What?"
📅 October 17, 2026
ℹ️ Information: github.com/OWASP-Ottawa...
🎟️ Tickets (free): buytickets.at/owaspottawac...
@jpmjr.bsky.socialNvidia is adding an independent control layer to help rein in AI agents before they go off the rails. jpmellojr.blogspot.com/2026/09/nvid... #AI #AIAgents #Nvidia #AIsecurity
@securestep9.bsky.social#AI coding agents asked for review screenshots leaked internal images in public #GitHub repos, researchers say.
Researchers counted 13,000+ images at 300+ orgs, including customer billing records!
#AISecurity
👇 thehackernews.com/2026/09/ai-c...
@theaitable.bsky.socialThe Next‑Gen Prompt Engineer: Imagine prompt templates treated like real code version‑controlled, secure, and reusable.
This isn’t just making prompts safer, it’s making them engineerable. Learn more: buff.ly/AEnlCaR
#PromptEngineering #GenAI #AIsecurity #DevOps
@ainewsbulletin.bsky.socialAI voice clone used to trick bank employee into sending €95 million to scammers. Authorities recovered half
@mjetechnologies.comttackers built fake ChatGPT apps and promoted them in sponsored Google ads to spread remote-access malware. Why it matters: as staff lean on AI tools daily, an unfamiliar sponsored link is now an easy way into your business. #MJE #AISecurity #Cybersecurity #SMB #ThreatIntel
@cyberupdates365.bsky.socialRSA Agent ID treats autonomous AI agents as governed identities.
It can discover agents and MCP servers, assign human owners, enforce per-call policy and require approval before designated high-risk actions.