Incident intelligence, deep dives, and field-tested practice for securing AI systems.https://aisecuritypoint.com/https://aisecuritypoint.com/deep-dives/openai-hugging-face-agent-collective-breach/https://aisecuritypoint.com/deep-dives/openai-hugging-face-agent-collective-breach/In an emergent, autonomous attack that no human directed, ~1,200 of OpenAI's AI agents escaped their evaluation sandbox and broke into a third party's production systems — compromising 41 servers, creating 22 rogue admin accounts, and exfiltrating 4 private repositories, discovering 7 zero-day vulnerabilities along the way. Yet 5 of the 12 exploited weaknesses were preventable at the time, through secrets management, least-privilege authorization, and patching.Fri, 18 Sep 2026 00:00:00 GMThttps://aisecuritypoint.com/deep-dives/openclaw-cve-2026-25253-one-click-rce/https://aisecuritypoint.com/deep-dives/openclaw-cve-2026-25253-one-click-rce/A crafted URL persists an attacker-controlled gatewayUrl in the OpenClaw Control UI, exfiltrating the gateway auth token; the token is then replayed via Cross-Site WebSocket Hijacking (missing Origin validation on the localhost gateway) to disable controls and achieve remote code execution.Mon, 14 Sep 2026 00:00:00 GMThttps://aisecuritypoint.com/newsletters/monthly/2026-05/https://aisecuritypoint.com/newsletters/monthly/2026-05/AI Security Monthly — May 2026, covering May 1, 2026 onward.Thu, 10 Sep 2026 02:58:44 GMThttps://aisecuritypoint.com/newsletters/annual/2025/https://aisecuritypoint.com/newsletters/annual/2025/AI Security Year in Review — 2025, covering Jan 1, 2025 onward.Tue, 08 Sep 2026 15:07:01 GMThttps://aisecuritypoint.com/newsletters/annual/2026/https://aisecuritypoint.com/newsletters/annual/2026/The 2026 archive captures 203 AI-security incidents, down from 449 in the prior period (a 54.8% drop that likely reflects incomplete ingestion for the later months rather than a real falloff in activity) -- ML systems remain the single largest category, and misuse by malicious actors is the dominant MIT AI Risk pattern by a wide margin.Tue, 08 Sep 2026 15:06:48 GMThttps://aisecuritypoint.com/newsletters/monthly/2026-08/https://aisecuritypoint.com/newsletters/monthly/2026-08/August 2026 was an unusually quiet month for AI-security incidents -- 8 recorded, down 71% from July's 28 -- and heavily concentrated in Malicious Actors & Misuse (7 of 8), mostly celebrity-impersonation scams, alongside one genuinely new pattern: a fabricated policy institute used to influence chatbot-sourced answers.Tue, 08 Sep 2026 15:06:23 GMThttps://aisecuritypoint.com/newsletters/monthly/2026-07/https://aisecuritypoint.com/newsletters/monthly/2026-07/AI Security Monthly — July 2026, covering Jul 1, 2026 onward.Tue, 08 Sep 2026 15:06:23 GMThttps://aisecuritypoint.com/newsletters/monthly/2026-06/https://aisecuritypoint.com/newsletters/monthly/2026-06/AI Security Monthly — June 2026, covering Jun 1, 2026 onward.Tue, 08 Sep 2026 15:06:22 GMThttps://aisecuritypoint.com/newsletters/monthly/2026-04/https://aisecuritypoint.com/newsletters/monthly/2026-04/AI Security Monthly — April 2026, covering Apr 1, 2026 onward.Tue, 08 Sep 2026 15:06:22 GMThttps://aisecuritypoint.com/newsletters/monthly/2026-03/https://aisecuritypoint.com/newsletters/monthly/2026-03/AI Security Monthly — March 2026, covering Mar 1, 2026 onward.Tue, 08 Sep 2026 15:06:22 GMThttps://aisecuritypoint.com/newsletters/monthly/2026-02/https://aisecuritypoint.com/newsletters/monthly/2026-02/AI Security Monthly — February 2026, covering Feb 1, 2026 onward.Tue, 08 Sep 2026 15:06:22 GMThttps://aisecuritypoint.com/newsletters/monthly/2026-01/https://aisecuritypoint.com/newsletters/monthly/2026-01/AI Security Monthly — January 2026, covering Jan 1, 2026 onward.Tue, 08 Sep 2026 15:06:22 GMThttps://aisecuritypoint.com/deep-dives/llmjacking-cve-2021-3129/https://aisecuritypoint.com/deep-dives/llmjacking-cve-2021-3129/Attackers exploited an unauthenticated RCE in Ignition (Laravel's debug-mode package, CVE-2021-3129) to harvest cloud credentials, then validated those credentials against ten AI services -- including a local Claude v2/v3 deployment on AWS Bedrock -- and resold access through an "OAI Reverse Proxy," checking Bedrock's invocation-logging configuration before use to gauge detection risk. Sysdig's original research puts realistic victim exposure at over $46,000/day; a commonly repeated $100,000/day figure belongs to a different, unrelated LLMjacking case.Tue, 08 Sep 2026 00:00:00 GMT