@newsanalysis.comAn OpenAI safety employee, David Robinson, has resigned, citing a "broken culture" within the company. He led safety report writing for major product launches and believes AI development needs more than just rules, advocating for a cultural shift. #AIsecurity #News
@ericdbrown.bsky.socialThe FTC opened a probe into AI product risks this week, Google, OpenAI and Anthropic all shipped new models, and Micron says the memory squeeze runs through 2028.
@psoheil.bsky.socialOpenAI, 53 user-provided images were posted to third-party image-hosting sites without the lab’s knowledge.
#AI #ArtificialIntelligence #AIAgents #Cybersecurity #AISecurity #DataPrivacy #InfoSec #CloudSecurity #MachineLearning
Read the full TechCrunch article:
techcrunch.com/2026/09/25/u...
@spoint42.bsky.socialDeepSeek Harness : l'agent sandboxé qui éteint lui-même son bac à sable, en une commande blog.gioria.org/fr/ai-securi... #AISecurity
@today0tech.bsky.socialRecent reports indicate a series of suspected AI hacking attempts aimed at stealing information.
#AIsecurity #cybersecurity #infosec
https://news.google.com/rss/articles/CBMiUkFVX3lxTFAwWkNVYW9FTXFrTE1Qa04wSzR6ak44a1dkdnN2LWpxSHRsVHd0OVdtQjhJZGIxZlVPZmlOSlRJZUZhcm5iLVJQNHJQaUJGWmZGWGc
@gltch.ioOpenAI AI “went rogue” 4+ times this spring, hacking sites (incl. Australia gov) during normal data tasks. #AIsecurity
@hexonbot.bsky.socialExploitBench results are not an AI safety score. Ask whether they repeat, what tools and safeguards shaped them, and which controls match the highest reliable capability. https://www.hexon.bot/blog/ai-cybersecurity-benchmarks-exploitbench #AISecurity #Cybersecurity #AIEvaluation
@shehackspurple.bsky.socialI'm curious what other AppSec and AI practitioners think (that's you).
What security controls do you see as essential for AI agents? And what are we still not sure about? What are your 'may haves'?
Article: https://twp.ai/9Obx3f
Is your AI app safe to ship?
LLM Red-Team Starter Kit: 50 ready-to-run tests mapped to OWASP LLM Top 10 (2025) + an Excel dashboard that scores your results.
Pay what you want:
https://techsavant013.gumroad.com/l/llm-redteam-starter-kit
@hackernoon.comIn multi-agent systems, a valid conclusion can gain unintended authority as it moves downstream. That makes agent handoffs a security boundary of their own. #aisecurity
@cyberupdates365.bsky.socialCritical CVE-2026-90970 affects GitLab Self-Hosted AI Gateway.
An authenticated Duo Agent Platform user can potentially escape the prompt-template sandbox and execute commands on the gateway.
@siliconsignalai.bsky.socialApple is tightening macOS Full Disk Access controls, citing new risks from increasingly capable AI agents that could misuse broad access to user…
#Apple #macOS #AISecurity
https://techcrunch.com/2026/10/02/apple-says-its-tightening-macos-full-disk-access-controls-due-to-new-risks-from-ai-agents/
@ossradarai.bsky.socialA new arXiv paper introduces APEX, showing how chained open-source LLM skills can carry attacker-controlled claims across steps to hijack agent decisions on SkillsBench. This highlights a supply-chain risk when…
#AIsecurity #LLMagents #OpenSourceAI #PromptInjection
https://arxiv.org/abs/2610.01564
@opsmatters.comThe latest update for #GitProtect includes "#AISecurity in #DevOps: Best Practices to Follow" and "Have you ever considered how much downtime costs?".
#cybersecurity #Backup #DevSecOps https://opsmtrs.com/3TszuS2
@securityonline.bsky.socialAWS fixes Loom for AWS admin takeover CVE-2026-103956 and a SageMaker Unified Studio code execution bug, CVE-2026-104019. Patch now.
#AWS #LoomForAWS #SageMaker #CVE2026103956 #CVE2026104019 #AISecurity #CloudSecurity #Vulnerability
@joebuildssystems.bsky.socialAI agent leak test. Check three boxes: - Reads private data - Reads outside content - Sends data out All three checked? Cut one. Outbound access is usually the easiest. From Aakash Abhay Yadav's talk at B-Sides Orlando 2026. #aiforbusiness #aisecurity #aiagents
@cybersecboardrm.bsky.socialWhen AI Agents Won’t Take No for an Answer: Why Access Control Is No Longer Enough
youtu.be/ouZuOENDev4 #AISecurity #AgenticAI #Cybersecurity #AIGovernance #ZeroTrust
@cyberlifecoach.bsky.socialAI agents are moving from answering questions to taking actions
That means permissions, blast radius, approval gates, least privilege, and kill switches suddenly matter a lot more.
My latest article looks at NIST AI RMF guardrails for agentic AI
shorturl.at/YKNAN
@shehackspurple.bsky.socialWho should be responsible when an autonomous agent goes rogue?
🎥 Watch here: https://twp.ai/9Obi3l
@joebuildssystems.bsky.socialAI agents drift from the job you approved. Mine has skipped my approval before. - Give every delete, send, and export its own rule - Check each action, every time I wrote up the full B-Sides Orlando 2026 breakdown in my newsletter. #aiagents #aisecurity #automation