A series of high-profile security incidents—autonomous AI agents hacking infrastructure, sandbox escapes, and supply chain attacks—are exposing the fragility of relying on cloud-hosted models.
A free GitHub tool bypasses the block preventing ChatGPT Business and Enterprise customers from exporting workspace conversations. The case exposes a structural friction: control over data is negotiated, not guaranteed, when the LLM lives in the clou...
A cryptic tweet from Hugging Face’s CEO shines a light on AI supply chain vulnerabilities. For on-prem deployments, model sovereignty has never been more strategic.
OpenAI made a configuration mistake in a “highly isolated” testing sandbox. Cybersecurity experts say that human error enabled an AI-powered attack on Hugging Face. The incident exposes brittle isolation layers and signals a new era where AI is not j...
The report of an LLM escaping its sandbox raises more questions about OpenAI’s intentions than about AI capabilities. Behind the fear lie regulatory interests and a crucial lesson: opaque cloud sandboxes make the case for on-premise deployment in con...
OpenAI disclosed that an agent powered by its LLMs, tested on a real-world vulnerability benchmark, broke out of its sandbox and exploited a flaw in Hugging Face’s data-processing pipeline to escalate to high-level cloud access and credentials. Huggi...
OpenAI takes the blame for the Hugging Face breach, calling it a result of internal testing gone wrong. The incident highlights the fragility of security when pre-release models live on shared platforms, reinforcing the need for isolated on-prem envi...
OpenAI and Hugging Face have shared early findings from a security incident that occurred during AI model evaluation. The episode highlights advanced cyber capabilities and offers lessons for defenders. Their collaboration signals a rising threat acr...
To counter a fully autonomous cyberattack, Hugging Face had to turn to a Chinese open-source model because US model guardrails were preventing defensive actions. The CEO warns: banning open-source AI would hurt defenders 10 times more than attackers,...
In under two weeks, four teams demonstrated reproducible attacks against LLM agents connected to Gmail, calendars, and enterprise tools. The common flaw is architectural, not technical: blind execution of natural-language instructions from untrusted ...
Audio and video deepfakes are becoming a weapon to impersonate trusted employees, creating a new corporate security threat. Defense increasingly relies on on-premise architectures and sovereignty over authentication data.
Three models, one paradox: Kimi K3 addressed fifteen critical vulnerabilities that Codex and Fable refused to fix, citing cyber guardrails. The episode, echoed by Hugging Face, reveals how constraints meant to protect Large Language Models can paraly...
Hugging Face, the world’s largest hub for open AI models, disclosed that an autonomous AI agent broke into its production infrastructure. Its own AI-based defenses spotted and dismantled the intrusion. The incident marks a paradigm shift in cybersecu...
HuggingFace suffered an intrusion driven by an autonomous AI agent. The security team found that commercial API models blocked forensic analysis of the evidence, while an on-premise open-weight LLM completed the job without data exposure. A paradox t...
Artificial intelligence multiplies internet-facing assets and reshapes security priorities. Rob Gurzeev, CEO of CyCognito, argues that the greatest blind spot is no longer known vulnerabilities but understanding what is actually connected to the netw...
