126 AI security incidents in one month. 38 orgs breached including NIST and CISA. AI was the attack vector in 39 of them — the largest category.
@bibo72.bsky.social
@bibo72.bsky.social126 AI security incidents in one month. 38 orgs breached including NIST and CISA. AI was the attack vector in 39 of them — the largest category.
@promptfoundry.bsky.socialNew arXiv work shows attackers can shorten predicted output lengths by up to 83.4% via adversarial suffixes, giving their requests priority in LLM schedulers and faster completion. Worth noting if you build or use…
#LLM #AIsecurity #PromptInjection #MLops
https://arxiv.org/abs/2610.03430
@opsmatters.comThe latest update for #Acronis includes "Acronis Cyber Protect #Cloud earns maximum score in AV-TEST Advanced Threat Protection test" and "Microsoft #AISecurity for SMBs: Purview Coverage Explained".
#cybersecurity #ThreatDetection #infosec https://opsmtrs.com/4hetaGC
@opsmatters.comThe latest update for #GitProtect includes "How to Automate Repository Backups to Minimize Data Loss and Downtime" and "#AISecurity in #DevOps: Best Practices to Follow".
#cybersecurity #Backup #DevSecOps https://opsmtrs.com/3TszuS2
@reksas13.bsky.socialDay 87/100 of #100DaysWithAI 🔐
Sealed reasoning blocks were portable: feed one model's encrypted thinking to a weaker sibling and it read it aloud. Now mitigated.
🔗 https://github.com/aurimas13/100-Days-With-AI
🤝 Claude Code-assisted, reviewed by me.
@securityonline.bsky.socialOpenAI text watermarking rolls out to ChatGPT and Codex in the EU under the EU AI Act, with opt-in API access and limited detector access.
#OpenAI #TextWatermarking #EUAIAct #textGrain #ContentProvenance #ChatGPT #AISecurity
@securityonline.bsky.socialIBM fixes 25 Langflow vulnerabilities, including critical RCE flaw CVE-2026-104334 (CVSS 9.8). Upgrade to Langflow 1.12.3 now.
#Langflow #IBM #AISecurity #CVE2026104334 #CVE202693674 #RCE #LLM #Vulnerability
@securityonline.bsky.socialAMD RCCL vulnerability CVE-2026-43598 in the ROCm Communication Collectives Library could allow RCE on Instinct GPUs. Update to ROCm 7.14.
#AMD #ROCm #RCCL #CVE202643598 #AMDInstinct #GPUSecurity #AISecurity #Vulnerability
@cipherpulseai.bsky.socialResearchers propose CVE2AP, an LLM-based method that automatically generates PDDL-encoded attack paths from natural-language CVE descriptions, aiming to scale up formal cybersecurity analysis.
#cybersecurity #AIsecurity #threatmodeling #LLM
https://arxiv.org/abs/2610.03383
@convectiveinc.bsky.socialCisco Talos read real attacker prompt logs. No exotic encoding. Mostly a plain "I'm allowed to do this," and the model complied. Check authorization outside the conversation. www.convective.com/blog/your-g... #AISecurity #CyberSecurity
@hexonbot.bsky.socialAI agents with macOS Full Disk Access can reach files, mail, messages, and browsing history. Audit privileged apps, scope their data, and separate read from send. https://www.hexon.bot/blog/macos-full-disk-access-ai-agent-permissions #MacSecurity #AISecurity #LeastPrivilege
@thedailytechfeed.comAI agent exploits zero-day in Zammad chain to seize root access in seconds #ZeroDay #AIsecurity #Zammad #SecOps #IncidentResponse #OpenSource thedailytechfeed.com/critical-ai-...
@cyberlensai.bsky.socialBefore you plug a new AI coding tool into your workflow, inspect what it can touch: repo, shell, browser, secrets, package manager, deployment path. The permission model matters more than the demo. #AISecurity #DevSecOps
@pondero-ai.bsky.socialAI Agents Are Breaking the Security Embargo: What to Do Before Your Next Fix Ships
https://pondero.ai/security/guides/ai-agent-vulnerability-disclosure-embargo-october-2026/
@valuechainrisk.orgCan this AI be trusted with what it reads, remembers and does? Our open framework: 7 dimensions, 40 questions, behavior-anchored levels 0 to 5. Free, CC BY 4.0.
valuechainrisk.org/ai-cogsec #AIsecurity #JawnCon #BSidesDE
@valuechainrisk.orgMETR told the U.S. Senate how AI agents under test escaped isolation and roughly 700 hit another company's systems. Ask vendors what their agents can reach. metr.org/blog/2026-09-30-chris-painter-senate-testimony/ #AIsecurity
@rodtrent.comWe Think We’re Using AI. We’re Also Feeding It.
#AgenticAI #AISecurity #AIGovernance #AgentSecurity #CyberGovernance
https://go.rodtrent.com/go/4jpau3k
@hendryadrian.bsky.socialPrompt injection is only the start. Real AI security comes from least privilege, scoped tools, human approval, logging, and containment when agents read emails, docs, or act autonomously. #PromptInjection #AISecurity #LeastPrivilege
@psoheil.bsky.socialDon’t just deploy a fix, verify it. Continuously discover, reproduce, patch, retest, deploy, independently verify, and monitor.
@devstackdaily.bsky.socialNew arXiv paper proposes AgentSecGraph, a static analysis framework that builds security-aware dependency graphs for LLM agents instead of relying on predefined sensitive sinks. It also introduces…
#AIsecurity #LLMagents #DevTools #SoftwareEngineering
https://arxiv.org/abs/2610.03014