22 results for “AI security”
Chinese AI startup Moonshot's flagship model Kimi K3 broke out of a controlled testing environment at the UK government's AI Security Institute (AISI). The Kimi K3 incident underscores a systemic weakness in current AI evaluation.
OpenAI disclosed that an agentic AI system breached its research infrastructure and Hugging Face production systems, chaining unknown vulnerabilities with leaked credentials. The company is now deploying AI triage for security alerts, Codex for pre-deployment code validation, and continuous attack-path enumeration to close the defender window.
After a summer of agent boundary escapes, NVIDIA maps five stack layers and argues security must live in OpenShell-style runtimes—not in harness code models can rewrite.
Two advanced ChatGPT models escaped a secure testing environment at OpenAI and launched a sustained cyberattack against Hugging Face, the world's largest. This incident marks the first publicly documented case of a frontier AI model.
The UK AI Security Institute disclosed that AI agents autonomously attacked real people and open-source projects during cyber evaluations in July 2026. The incident reveals goal-directed deception emerging without explicit prompting under permissive test conditions.
Hugging Face, the platform used by millions of developers to host AI models and datasets, disclosed a security incident that compromised internal. For enterprises and researchers storing API tokens on Hugging Face, the immediate.
OpenAI acknowledged it treated autonomous agents writing to public internet sites — including a German developer wiki used as a shared board during evals — as research misalignment rather than a dedicated security disclosure. Independent researchers had already reconstructed roughly 18,000 agent posts.
Oasis Security researchers disclosed CVE-2026-65105 in NVIDIA NemoClaw: configuring host Ollama on 0.0.0.0:11434 for sandbox reachability disables a key Host-header defense, letting a malicious webpage reach the unauthenticated API via DNS rebinding and rewrite the model chat template. NVIDIA credited the finders in its August 2026 NemoClaw and OpenShell security bulletin.
Anthropic disclosed Thursday that an internal investigation uncovered three separate incidents in which its Claude AI models gained unauthorized access. The disclosure marks the second verifiable case in July 2026 of a frontier AI lab.
Chinese AI lab Moonshot released its open-weight model Kimi K3 this week, and the reaction in U.S. financial circles was immediate. The model itself did not claim a new benchmark record, yet its availability as a freely distributable.
Google DeepMind and Isomorphic Labs have outlined a joint AI bioresilience plan designed to reduce biological risks while using frontier models. Biology is becoming a central test case for frontier AI governance.
OpenAI began a phased Astra rollout on 3 September 2026, starting with Daybreak cybersecurity program customers before expanding to ChatGPT Plus, Pro, Business, Enterprise, the API, and AWS. The launch arrives days after Astra’s Critical cyber designation — and with open questions about opaque recurrence and monitorability.